BlV
The Johns Hopkins
ACG System
®
Applications Guide
Version 9.0
December 2009
Important Warranty Limitation and Copyright Notices
Copyright 2009, The Johns Hopkins University. All rights reserved.
This document is produced by the Health Services Research & Development Center at
The Johns Hopkins University, Bloomberg School of Public Health.
The terms The Johns Hopkins ACG® System, ACG® System, ACG®, ADG®, Adjusted
Clinical Groups®, Ambulatory Care GroupsTM, Aggregated Diagnostic GroupsTM,
Ambulatory Diagnostic GroupsTM, Johns Hopkins Expanded Diagnosis ClustersTM,
EDCsTM, ACG Predictive Model™, Rx-Defined Morbidity Groups™, Rx-MGs™, ACG
Rx GapsTM,, ACG Coordination Markers™, ACG-PM™, Dx-PM™, Rx-PM™ and DxRxPM™ are trademarks of The Johns Hopkins University. All materials in this document
are copyrighted by The Johns Hopkins University. It is an infringement of copyright law
to develop any derivative product based on the grouping algorithm or other information
presented in this document.
This document is provided as an information resource on measuring population morbidity
for those with expertise in risk-adjustment models. The documentation should be used
for informational purposes only. Information contained herein does not constitute
recommendation for or advice about medical treatment or business practices.
No permission is granted to redistribute this documentation. No permission is granted to
modify or otherwise create derivative works of this documentation.
Copies may be made only by the individual who requested the documentation initially
from Johns Hopkins or their agents and only for that person's use and those of his/her coworkers at the same place of employment. All such copies must include the copyright
notice above, this grant of permission and the disclaimer below must appear in all copies
made; and so long as the name of The Johns Hopkins University is not used in any
advertising or publicity pertaining to the use or distribution of this software without
specific, written prior authorization.
Disclaimer: This documentation is provided AS IS, without representation as to its
fitness for any purpose, and without warranty of any kind, either express or implied,
including without limitation the implied warranties of merchantability and fitness for a
particular purpose. The Johns Hopkins University and the Johns Hopkins Health System
shall not be liable for any damages, including special, indirect, incidental, or
consequential damages, with respect to any claim arising out of or in connection with the
use of the documentation, even if it has been or is hereafter advised of the possibility of
such damages.
Documentation Production Staff
Senior Editor: Jonathan P. Weiner, Dr. P.H.
Managing Editor: Chad Abrams, M.A.
Production assistance provided by: David Bodycombe Sc.D., Klaus Lemke, Ph.D.,
Patricio Muniz, M.D., MPH, MBA, Thomas M. Richards, Barbara Starfield, M.D., MPH
and Erica Wernery.
Special thanks to Lorne Verhulst M.D., MPA, of the British Columbia Ministry of Health
in Vancouver, Canada, for his contribution to the chapter titled Practitioner Profiling:
Assessing Individual Physician Performance Provider Performance Assessment.
Additional production assistance and original content provided by Rosina DeGiulio, Lisa
Kabasakalian, Meg McGinn, and Amy Salls of DST Health Solutions, LLC. The ACG
Team gratefully acknowledges the support provided by our corporate partner in helping
to move this publication forward.
If users have questions regarding the software and its application, they are advised to
contact the organization from which they obtained the ACG software. Questions about
grants of rights or comments, criticisms, or corrections related to this document should be
directed to the Johns Hopkins ACG team (see below). Such communication is
encouraged.
ACG Project Coordinator
624 N. Broadway - Room 607
Baltimore, MD 21205-1901 USA
Telephone (410) 955-5660
Fax: (410) 955-0470
E-mail: askacg@jhsph.edu
Website: http://acg.jhsph.edu
H
H
Third Party Library Acknowledgements
This product includes software developed by the following companies:
Health Plus Technologies (http://www.healthplustech.com)
Karsten Lentzsch (http://www.jgoodies.com)
Sentintel Technologies, Inc. (http://www.healthplustech.com)
This product includes software developed by The Apache Software Foundation
(http://www.apache.org)
This product includes the Java Runtime Environment developed by Sun Microsystems
(http://java.sun.com)
This product includes the following open source:
JDOM library (http://www.jdom.org)
iText library (http://www.lowagie.com/iText)
JasperReports library (http://www.jasperforge.org)
i
Table of Contents
1 Introduction...................................................................................................... 1-i
Introduction to The Johns Hopkins ACG® System.................................... 1-1
Applications Guide Objective ...................................................................... 1-1
Applications Guide Navigation .................................................................... 1-1
Applications Guide Content ......................................................................... 1-2
Installation and Usage Guide Content ........................................................ 1-3
Technical Reference Guide Content............................................................ 1-4
Customer Commitment and Contact Information .................................... 1-5
2 Health Status Monitoring................................................................................ 2-i
Health Status Monitoring ............................................................................. 2-1
Epidemiology of Disease Within a Single Population ................................ 2-2
Comparing Disease Distribution Across Two or More Subpopulations.. 2-5
Age/Sex-Adjusted Comparison of Disease Distributions Across
Populations – Standardized Morbidity Ratios (SMRs) ............................. 2-9
Calculating Age/Sex Adjusted Standardized Morbidity Ratios ............. 2-10
Using a Combination of EDCs and ACGs to Support Case
Management and Disease Management.................................................... 2-12
3 Performance Assessment................................................................................. 3-i
Introduction ................................................................................................... 3-1
Goals and Objectives .................................................................................... 3-1
Theory and Background............................................................................... 3-1
Software-Produced Weights and Their Uses.............................................. 3-2
Concurrent ACG-Weights ........................................................................... 3-4
Customizing Risk Scores Using Local Cost Data ....................................... 3-5
Applications Guide
The Johns Hopkins ACG System, Version 9.0
ii
How to Take a Population-Based Approach to Practitioner
Profiling........................................................................................................ 3-12
Examples: Profiling Primary Care Physicians........................................ 3-18
Introducing Primary Responsible Physician (PRP)................................. 3-21
Comparing Specialists to Specialists – Intra-specialty Expected
Level of Service and Costs.......................................................................... 3-23
Using EDCs in the Context of Practitioner Profiles................................. 3-24
Summary...................................................................................................... 3-29
4 Clinical Screening by Care and Disease Managers ...................................... 4-i
Introduction ................................................................................................... 4-1
High-Risk Case Identification for Case Management ............................... 4-1
The ACG Predictive Model’s Probability Score ........................................ 4-7
Risk Stratification ....................................................................................... 4-12
Disease Management Candidates .............................................................. 4-14
Case Mix Control ........................................................................................ 4-16
Technical Considerations ........................................................................... 4-17
5 Managing Financial Risk for Pharmacy Benefits ......................................... 5-i
Managing Financial Risk for Pharmacy Benefits ...................................... 5-1
6 Capitation and Rate Setting............................................................................ 6-i
Capitation and Rate Setting ......................................................................... 6-1
7 Final Considerations........................................................................................ 7-i
Introduction ................................................................................................... 7-1
Art of Risk Adjustment ................................................................................ 7-1
Time Frames and Basic Population Perspectives....................................... 7-2
Handling New or Part-Year Enrollees ........................................................ 7-4
Sample Size .................................................................................................... 7-5
The Johns Hopkins ACG System, Version 9.0
Applications Guide
iii
Handling High Cost or Outlier Cases ......................................................... 7-6
Constructing Resource Consumption Measures ........................................ 7-6
Index .............................................................................................................IN-1
Applications Guide
The Johns Hopkins ACG System, Version 9.0
iv
This page was left blank intentionally.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Introduction
1-i
1 Introduction
Introduction to The Johns Hopkins ACG® System.................................... 1-1
H
H
Applications Guide Objective ...................................................................... 1-1
H
H
Applications Guide Navigation.................................................................... 1-1
H
H
Applications Guide Content......................................................................... 1-2
H
H
Installation and Usage Guide Content ........................................................ 1-3
H
H
Technical Reference Guide Content............................................................ 1-4
H
H
Customer Commitment and Contact Information .................................... 1-5
H
Applications Guide
H
The Johns Hopkins ACG System, Version 9.0
1-ii
Introduction
This page was left blank intentionally.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Introduction
1-1
Introduction to The Johns Hopkins ACG® System
The ACG (Adjusted Clinical Groups) System was developed by faculty at the Johns
Hopkins Bloomberg School of Public Health to help make health care delivery more
efficient and more equitable. Because the ACG System can be used for numerous
management, finance, and analytical applications related to health and health care, it has
become the most widely used, population-based, case-mix/risk adjustment methodology.
Precisely because of the diversity of ACG applications, one size does not fit all in terms
of methodology. Like health management and analysis itself, using case-mix or risk
adjustment methods involves art as well as science, and these applications are particularly
context and objective driven. We hope this documentation will provide you with much of
the guidance you will need in order to apply the ACG System to most effectively meet
the risk adjustment and case-mix needs of your organization.
Applications Guide Objective
The Applications Guide was designed to assist analysts, programmers, or other personnel
who are responsible for applying ACG methods to data. The objective of this manual is
to illustrate common applications of The ACG System. Whenever possible, basic
instructions for common post-processing calculations are provided.
Applications Guide Navigation
Locating information in the Applications Guide is facilitated by the following search
methods:
•
Master Table of Contents. The master table of contents contains the chapter names
and principal headings for each chapter.
•
Chapter Table of Contents. Each chapter has a table of contents, which lists the
principal headings and subheadings and figures and tables.
•
Index. Each chapter is indexed and organized alphabetically.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
1-2
Introduction
Applications Guide Content
For your convenience, a list of the Applications Guide chapters is provided.
•
Chapter 1: Introduction.
•
Chapter 2: Health Status Monitoring. This chapter demonstrates the application
of the ACG System markers and analyses to measures disease prevalence and support
health status monitoring.
•
Chapter 3: Performance Assessment. This chapter outlines the basic steps to
taking a population-based approach to profiling.
•
Chapter 4: Clinical Screening by Care and Disease Managers. This chapter
demonstrates the application of the ACG System markers to high risk case selection,
risk stratification and amenability.
•
Chapter 5: Managing Financial Risk for Pharmacy Benefits. This chapter
describes use of the ACG System markers to the application of managing pharmacy
risk.
•
Chapter 6: Capitation and Rate Setting. This chapter describes various methods of
applying the ACG System to capitation and rate setting.
•
Chapter 7: Final Considerations.
•
Index
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Introduction
1-3
Installation and Usage Guide Content
For your convenience, a list of the Installation and Usage Guide chapters is provided.
•
Chapter 1: Getting Started. Provides a general overview of the physical
organization of the manual as well as content.
•
Chapter 2: Overview of the ACG Toolkit. Intended for all users, this chapter
provides a brief overview of the ACG toolkit to provide a baseline introduction to the
nomenclature of the ACG System components.
•
Chapter 3: Installing the ACG Software. Intended for the programmer/analyst,
this chapter discusses the technical aspects of installing the software.
•
Chapter 4: Basic Data Requirements. Intended for the programmer/analyst, this
chapter discusses at a high level the minimum data input requirements and other
necessary data requirements for performing ACG-based risk adjusted analyses.
Included are discussions of augmenting or supplementing diagnosis information with
optional user supplied flags as well as consideration of the use of pharmacy
information.
•
Chapter 5: Using the ACG Sample Data. Intended for the programmer/analyst, this
chapter walks through the example of processing the sample data provided with the
installation. This sample data is provided to allow users to understand the outputs of
the ACG System and demonstrate system functionality prior to the availability of the
user’s own input data.
•
Chapter 6: Using Client Data. Intended for the programmer/analyst, the purpose of
this chapter is to describe the process of importing user-supplied data files into the
sytsem.
•
Chapter 7: Operating the ACG Software. Intended for the programmer/analyst,
this chapter discusses the technical aspects of using the software, and importing and
exporting data and reports.
•
Chapter 8: Validating Results. Intended for the programmer/analyst, the purpose of
this chapter is to provide examples of the ACG System functionality that was
designed to assist in the validation of user-supplied data.
•
Chapter 9: Troubleshooting. Intended for the programmer/analyst, the purpose of
this chapter is to leverage prior user experience with the software to describe the
symptom and solution to common user issues.
•
Appendix A: Output Data Dictionary.
•
Appendix B: Report Detail.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
1-4
Introduction
•
Appendix C: Batch Mode Processing.
•
Appendix D: Java API.
•
Index
Technical Reference Guide Content
For your convenience, a list of the Technical Reference Guide chapters is provided.
•
Chapter 1: Introduction. Provides a general overview of the physical organization
of the manual as well as content.
•
Chapter 2: Diagnosis and Code Sets. This chapter discusses the applicability of
diagnosis data to the field of risk assessment and the challenges of managing multiple
standards for diagnosis coding.
•
Chapter 3: Adjusted Clinical Groups (ACGs). This chapter provides a brief
overview of the history of the clinical origin of the ACG System and describes the
details of the ACG assignment algorithm.
•
Chapter 4: Expanded Diagnosis Clusters (EDCs). This chapter explains the
development and evolution of the EDC methodology.
•
Chapter 5: Medication Defined Morbidity. This chapter discusses the
applicability of pharmacy claims in risk assessment and defines how drug codes are
assigned to morbidity groups.
•
Chapter 6: Special Population Markers. This chapter discusses the definitions and
clinical criteria for the HOSDOM, Frailty, Chronic Condition Count, and Chronic
Condition Markers.
•
Chapter 7: Predicting Resource Use. This chapter discusses the methods and
variations of ACG Predictive Models developed to predict resource use.
•
Chapter 8: Predictive Modeling Statistical Performance. This chapter
demonstrates the ACG predictive models statistical performance while describing the
various ways in which they can be applied in health care applications.
•
Chapter 9: Predicting Hospitalization. This chapter discusses the methods and
variations of ACG Predictive Models developed to predict hospitalization risk.
•
Chapter 10: Coordination. This chapter discusses the methods for evaluating
coordination of care.
•
Chapter 11: Gaps in Pharmacy Utilization. This chapter discusses the methods
and metrics associated with medication possession and gaps in pharmacy adherence.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Introduction
1-5
•
Appendix A: ACG Publication List
•
Appendix B: Variables Necessary to Locally Calibrate the ACG Predictive
Models
•
Index
Customer Commitment and Contact Information
As part of our ongoing commitment to furthering the international state-of-the-art of riskadjustment methodology and supporting users of the ACG System worldwide, we will
continue to perform evaluation, research, and development. We will look forward to
sharing the results of this work with our user-base via white papers, our web site, peerreviewed articles, and in-person presentations. After you have carefully reviewed the
documentation supplied with this software release, we would welcome your inquiries on
any topic of relevance to your use of the ACG System within your organization.
Technical support is available during standard business hours by contacting your
designated account representative directly. If you do not know how to contact your
account representative, please call 866-287-9243 or e-mail acg@dsthealthsolutions.com.
We thank you for using the ACG System and for helping us to work toward meeting the
Johns Hopkins University’s ultimate goal of improving the quality, efficiency, and equity
of health care across the United States and around the globe.
H
Applications Guide
The Johns Hopkins ACG System, Version 9.0
1-6
Introduction
This page was left blank intentionally.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Health Status Monitoring
2-i
2 Health Status Monitoring
Health Status Monitoring............................................................................. 2-1
H
H
Table 1: Movers Analysis—Tracking Morbidity Burden Over Time ...... 2-1
H
H
Epidemiology of Disease Within a Single Population................................ 2-2
H
H
Table 2: Distribution of EDCs Within a Commercial HMO
Population*................................................................................................. 2-3
Table 3: Summary Descriptive Statistics for a Commercial
HMO’s EDC Distribution* ........................................................................ 2-4
H
H
H
H
Comparing Disease Distribution Across Two or More
Subpopulations .............................................................................................. 2-5
H
H
Table 4: Member Demographic and Plan Features for Four
Populations in Three Exemplary Plans ...................................................... 2-5
Table 5: Major EDC Prevalence for Four Populations in Three
Exemplary Health Plans ............................................................................. 2-7
H
H
H
H
Age/Sex-Adjusted Comparison of Disease Distributions Across
Populations – Standardized Morbidity Ratios (SMRs) ............................. 2-9
H
H
Table 6: Observed to Expected Standardized Morbidity Ratio
by MEDC ................................................................................................... 2-9
H
H
Calculating Age/Sex Adjusted Standardized Morbidity Ratios ............. 2-10
H
H
Using a Combination of EDCs and ACGs to Support Case
Management and Disease Management.................................................... 2-12
H
H
Table 7: Distribution of RUB Co-Morbidity Levels Within
Selected EDC Disease Categories and Relative Resource Use
Morbidity Ratios for Each EDC/RUB Category...................................... 2-13
H
H
Applications Guide
The Johns Hopkins ACG System, Version 9.0
2-ii
Health Status Monitoring
This page was left blank intentionally.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Health Status Monitoring
2-1
Health Status Monitoring
Monitoring the health status of a population may be desirable for purposes of setting
health policy or demonstrating value to health purchasers. As a population ages, health
may be expected to decline, but interventions to improve population health may improve
or reverse that trend. The ACG System describes population health in a unique,
aggregate way that can be trended over time.
In the example below, the case-mix for the population demonstrates a sharp increase in
case-mix from 1.02 to 1.17. Using a “movers analysis,” Resource Utilization Bands that
stratify the population into low, moderate and high morbidity categories, can be used to
show changing morbidity patterns within a population (see Table 1). For example, in the
prior period there were 758 patients assigned to the low morbidity category – 405 of
these individuals stayed in the low morbidity category, 329 moved to the moderate
morbidity bucket, and 24 moved to the high morbidity bucket. For those who went from
low to high, their average cost went from $2,383 to $14,183. Similarly, there were 2,271
moderate morbidity patients in the prior period. Roughly half stayed the same and
slightly less then half moved to low morbidity categories, but 10% moved to high
morbidity categories and tripled their resource use.
Table 1: Movers Analysis—Tracking Morbidity Burden Over Time
Current Period (Case Mix = 1.17)
Low
Morbidity
Low
Morbidity
Prior Period
(Case-Mix
=1.02)
Moderate
Morbidity
Applications Guide
Moderate
Morbidity
High
Morbidity
405
329
24
12.0%
9.7%
0.7%
P: $618
P: $705
P: $2,383
C: $1,382
C: $1,512
C: $14,183
986
1074
211
37.6%
41.0%
8.1%
P: $2,116
P: $2,123
P: $3,599
C: $2,549
C: $1,844
C: $9,507
The Johns Hopkins ACG System, Version 9.0
2-2
Health Status Monitoring
Epidemiology of Disease Within a Single Population
The Expanded Diagnosis Clusters (EDCs) provide a means to standardize condition
definitions for purposes of health status monitoring. Table 2 presents the frequency
distribution of selected EDCs and provides a count of persons and annual prevalence
rates (Nos. persons with EDC/1,000 in population) for selected EDC and MEDC
categories. The numerators for the prevalence rates are the number of persons with each
EDC/MEDC and the denominators for these rates are the total number of persons
(defined in terms of unique ID numbers fed into the ACG grouper) in the target
population. These prevalence rates are population-oriented: the denominator includes
both users and non-users of services. In Table 2, rates per 1000 population for the
summary MEDC categories do not necessarily reflect the sum of the EDC rates that
comprise that MEDC. If a person has more than one EDC within the MEDC, that person
is only counted once within the broader category. Note: Table 2 presents only a sample
of the full EDC and MEDC categories for illustrative purposes.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Health Status Monitoring
2-3
Table 2: Distribution of EDCs Within a Commercial HMO Population*
EDC Distribution
(note only a small subset of total EDCs are displayed)
MEDC
EDC
No. of
Persons
with EDC
Description
Administrative
Prevalence**
per 1,000
Population
48,933
496.59
Surgical aftercare
3,778
38.34
Allergy
8,720
88.49
ALL01
Allergic reactions
1,447
14.68
ALL03
Allergic rhinitis
5,501
55.83
Cardiovascular
9,362
95.01
CAR01
Cardiovascular signs and symptoms
1,132
11.49
CAR03
Ischemic heart disease
801
8.13
CAR04
Congenital heart disease
192
1.95
CAR05
Congestive heart disease
179
1.82
ADM
ADM02
ALL
CAR
*There were a total of 98,539 12-month enrollees in the analysis.
**The numerator is the count of persons having one or more ICD codes that are categorized into the
specific EDC. The denominator: the total number of persons (defined by unique ID numbers) within the
study population (i.e., health plan). The quotient is multiplied by 1,000 to express the ratio in terms of
prevalence per 1,000 population.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
2-4
Health Status Monitoring
Table 3 presents an alternative way of summarizing EDC information at the population
level and presents the distribution (and percent) of the population by number of EDC
categories. The average number of EDCs per individual is also tabulated. So as to
minimize the effects of non-users, typically fifteen to thirty-five percent of the
population, descriptive statistics are calculated separately for those with and without an
EDC assignment.
Table 3: Summary Descriptive Statistics for a Commercial HMO’s
EDC Distribution*
Description
Frequency
Percent
Number of people with 0 EDCs
15,932
16.17
Number of people with 1 EDC
14,811
15.03
Number of people with 2 EDCs
15,019
15.24
Number of people with 3 EDCs
13,430
13.63
Number of people with 4 EDCs
10,930
11.09
Number of people with 5 EDCs
8,403
8.53
Number of people with 6 EDCs
6,090
6.18
Number of people with 7 EDCs
4,273
4.34
Number of people with 8 EDCs
3,040
3.09
Number of people with 9 EDCs
2,049
2.08
Number of people with 10+ unique EDCs
4,562
4.63
Average
Average number of EDCs per person
3.41
Of those with an EDC, average number of EDCs per
person
4.07
*There were a total of 98,539, 12-month enrollees in the analysis.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Health Status Monitoring
2-5
Comparing Disease Distribution Across Two or More
Subpopulations
A logical extension of this application is to compare the prevalence of EDCs across two
or more populations, say by product-line, employer group, medical groups, or individual
practitioners. These population (or plan) subgroups can be compared to each other, or to
the full population.
To illustrate this point, Table 4 presents member demographics and plan features for four
sample populations: Medicaid SSI (Supplemental Security Income, primarily disabled
persons), Medicaid TANF (Temporary Assistance for Needy Families, primarily women
and their children), Commercial Fee-for-Service, and Commercial HMO. The
commercial products are from separate insurers. Since every population will be unique,
the reader is cautioned that these distributions are not intended to be normative
benchmarks.
Table 4: Member Demographic and Plan Features for Four
Populations in Three Exemplary Plans
Medicaid
SSI
Medicaid
TANF
Commercial Commercial
FFS
HMO
Commercial Plan
No
No
Yes
Yes
Mental Health Carve-out
No
No
Yes
Yes
1996
1996
1995
1997
47,287
92,461
552,284
98,539
92.4
96.4
92.5
94.8
# Enrollees with 1+ EDC
75.6
85.2
70.7
83.8
Avg. # of ICDs per Enrollee
11.4
6.3
3.7
5.7
Avg. # of EDCs per Enrollee
7.0
4.4
2.9
4.1
41.1 (20.5)
14.2 (11.7)
31.8 (17.6)
30.2 (17.3)
1 – 111
1 – 96
0 – 64
0 – 96
Plan Features
Analysis Period
Demographics
# of 12 Month Enrollees
% Retained Next Year
For Concurrent Enrollees:
Mean Age (Std. Dev.)
Age Range
Applications Guide
The Johns Hopkins ACG System, Version 9.0
2-6
Health Status Monitoring
Medicaid
SSI
Medicaid
TANF
Commercial Commercial
FFS
HMO
% Under Age 18
16.5
72.0
28.2
30.3
% Age 65 and Older
12.8
<0.1
0.0
0.7
% Female
52.2
60.7
50.0
52.3
Table 5 presents the MEDC distributions for these four sample plans. Note the large
differences in prevalence rates for the Administrative MEDC ranging from a low of 200
persons per 1,000 in the Commercial FFS plan compared to a high of 564 persons per
1,000 in the Medicaid TANF population. Though marked differences, this result is
perhaps not particularly surprising given the higher prevalence of pediatrics in the TANF
population and the emphasis in this program on routine wellness/immunization visits.
Other differences may be harder to explain. The generally high prevalence rates in the
HMO relative to the FFS plan may be a consequence of a greater morbidity burden,
richer benefits, patient preferences, variation in physician practice patterns, or utilization
rates. Such differences underscore the need for caution in developing and applying
references.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Health Status Monitoring
2-7
Table 5: Major EDC Prevalence for Four Populations in Three
Exemplary Health Plans
Medicaid SSI
Commercial
FFS
#
#
Persons
Per 1,000
Populatio
n
Person
s w/
EDC
Persons
Per 1,000
Populatio
n
16,353
345.82
52,137
563.88
110,51
1
200.10
48,933
496.59
4,829
102.12
11,791
127.52
28,624
51.83
8,720
88.49
14,715
311.18
4,879
52.77
38,207
69.18
9,362
95.01
Dental
1,066
22,54
2,977
32.20
4,947
8.96
1,039
10.54
Developmenta
l
3,486
73.72
2,539
27.46
376
0.68
480
4.87
Ears, Nose,
Throat
10,016
211.81
35,931
388.61
132,13
1
239.24
30,001
304.46
Endocrine
6,520
137.88
1,904
20.59
14,288
25.87
3,802
38.58
Eye
5,721
120.98
8,173
88.39
62,855
113.81
19,929
202.24
Female
Reproductive
4,577
96.79
14,781
159.86
47,929
86.78
13,251
134.47
Gastrointestin
al/
Hepatic
7,990
168.97
9,309
100.68
26,544
48.06
7,251
73.59
General Signs
and
Symptoms
10,546
223.02
12,094
130.80
33,035
59.82
10,577
107.34
General
Surgery
11,442
241.97
12,774
138.16
64,274
116.38
15,770
160.04
Genetic
287
6.07
44
0.48
141
0.26
48
0.49
Genito-
5,115
108.17
7,079
76.56
26,346
47.70
6,821
69.22
Administrativ
e
Allergy
Cardiovascula
r
Applications Guide
#
Person
s w/
EDC
#
Commercial
HMO
#
Major EDC
#
Person
s w/
EDC
Medicaid
TANF
Persons
Per 1,000
Populatio
n
#
Person
s w/
EDC
#
Persons
Per 1,000
Populatio
n
The Johns Hopkins ACG System, Version 9.0
2-8
Health Status Monitoring
Medicaid SSI
Major EDC
#
Person
s w/
EDC
Medicaid
TANF
#
#
#
Persons
Per 1,000
Populatio
n
Person
s w/
EDC
Persons
Per 1,000
Populatio
n
Commercial
FFS
#
Person
s w/
EDC
#
Persons
Per 1,000
Populatio
n
Commercial
HMO
#
Person
s w/
EDC
#
Persons
Per 1,000
Populatio
n
urinary
Hematologic
3,082
65.18
2,594
28.06
4,809
8.71
1,296
13.15
Infections
3,735
78.99
8,959
96.89
16,344
29.59
4,950
50.23
Malignancies
1,868
39.50
489
5.29
6,165
11.16
1,237
12.55
Musculosketal
12,715
268.89
14,270
154.34
123,84
0
224.23
21,933
222.58
Neurologic
9,294
196.54
5,846
63.23
28,376
51.38
8,286
84.09
Nutrition
2,309
48.83
2,584
27.95
3,345
6.06
2.430
24.66
Psychosocial
15,226
321.99
15,769
170.55
3,375
6.11
10,195
103.46
Reconstructiv
e
3,011
63.68
6,025
65.16
17,631
31.92
3,289
33.38
Renal
2,535
53.61
1,140
12.33
1,954
3.54
679
6.89
11,815
249.86
18,277
197.67
82,190
148.82
18,856
191.36
Rheumatologi
c
2,057
64.65
935
10.11
6,194
11.22
1,298
13.17
Skin
8,615
182.19
21,934
237.22
80,893
146.47
20,785
210.93
Toxic Effects
1,214
25.67
2,463
26.64
2,478
4.49
696
7.06
Respiratory
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Health Status Monitoring
2-9
Age/Sex-Adjusted Comparison of Disease Distributions
Across Populations – Standardized Morbidity Ratios (SMRs)
Beyond just comparing prevalence rates, the next step in assessing disease or morbidity
burden of populations would be to introduce rudimentary case-mix to account or adjust
for differences that might occur naturally within the population. Age-sex observed-toexpected standardized morbidity ratios (or SMRs) are an epidemiological adjustment
process that provides a framework for comparing EDC prevalence across subpopulations.
Such information can be used to help practitioners and managers understand which
specific conditions are more or less common than average (i.e., beyond statistical chance)
among subpopulations of interest.
Table 6 presents the observed-to-expected standardized morbidity for a few selected
MEDC categories. The columns include the observed prevalence per 1,000, the age-sex
expected prevalence, the standardized morbidity ratio (SMR), and a 95% confidence
interval. In the last column, the asterisks (*) are noted for those confidence intervals that
do not cross 1.0, indicating that the age-sex adjusted prevalence for the Major EDC is
significantly (at the .05 level) higher or lower than the ratios with the underlying
population.
Table 6: Observed to Expected Standardized Morbidity Ratio by
MEDC
The following table is for Medical Group A (13,237 enrollees) within a sample health
plan.
(1)
Major EDC
(2)
Observed
Prevalence
per 1,000
(3)
Age-sex
Expected
Prevalence
per 1,000
(4)
Standardized
Morbidity
Ratio
(SMR)
(5)
95% SMR
Confidence Level
Lower Limit
Upper Limit
Administrative
566.06
474.91
1.218
1.190
1.245*
Allergy
115.89
78.15
1.483
1.409
1.557*
Cardiovascular
93.22
124.84
0.747
0.705
0.788*
Dental
14.13
11.00
1.284
1.100
1.469*
1.89
1.05
1.804
1.097
2.511*
Developmental
Note: The reference population of XYZ Health Plan is comprised of 211,773, 12-month enrollees under
age 65 with similar insurance contracts. In the last column, the asterisks (*) are noted for those CIs that
do not cross 1.0, indicating that the age-sex adjusted prevalence for the Major EDC is significantly (at
the .05 level) higher or lower than the ratios with the underlying population.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
2-10
Health Status Monitoring
Calculating Age/Sex Adjusted Standardized Morbidity Ratios
Step 1: Form the sub-groups; obtain denominators for prevalence rates.
Identify the subgroups based on an appropriate identifier such as a medical group,
product-line, employer group, or primary-care practitioner identification number. Obtain
a count of the number of in-scope persons (e.g., selecting only those enrolled for 6+
months) within each subgroup. These counts will serve as the denominators for the
subgroup prevalence rates.
Step 2: Calculate observed prevalence rates for the sub-groups.
Obtain a count of the number of persons with a given EDC within the subgroup of
interest. Divide this numerator by the denominator obtained in Step 1, and multiply the
quotient by 1,000. Repeat this procedure for each EDC and subgroup in the analysis.
These calculations are the observed prevalence rates per 1,000 population.
Step 3: Obtain age-sex specific EDC prevalence for the total study population.
The goal of this step is to calculate the average prevalence of each EDC for each age-sex
cohort within the entire population (e.g., all subgroups combined). These rates will be
used in subsequent steps. We recommend dividing the population into the following agesex groups:
Males: 0-4, 5-11, 12-17, 18-34, 35-44, 45-54, 55-64, 65+
Females: 0-4, 5-11, 12-17, 18-34, 35-44, 45-54, 55-64, 65+
Calculate EDC-specific prevalence rates for each age-sex group. The numerator for these
calculations will be the number of persons in the age/sex group who have the EDC; the
denominator is total number of persons in the age/sex group. These rates will be used in
Step 4. (These age-breaks assume a working age population; if a full age-range
population or elderly population is being assessed, we suggest expanding the 65+ age
category as follows: 65-74, 75-84, 85+.)
Step 4: Calculate expected prevalence rates.
By comparing prevalence rates that have been age/sex standardized, an analyst can rule
out the possibility that any differences in disease rates between two populations are due
solely to differences in age/ sex distribution. In this step, we outline how to calculate an
expected prevalence rate that is adjusted for the unique age-sex distribution of each subgroup.
Each member of the subgroup of interest should be categorized into one of the age-sex
cells listed in Step 3 above. The EDC-specific expected prevalence rate for each
demographic cell is calculated as follows:
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Health Status Monitoring
2-11
1. Obtain a count of persons within the subgroup falling into each age-sex cell.
2. For the EDC of interest, multiply these observed counts by the age-sex specific
prevalence rates obtained from the total study population in Step 3 to get expected
counts.
3. Sum the cell-specific expected counts across all age/sex cells.
4. Divide this sum by the total number of persons in the sub-population being studied.
5. Multiply this quotient by 1,000 to get an expected prevalence rate per 1,000
population for that particular EDC.
6. Repeat this process for each EDC of interest.
This expected rate tells you how common the EDC would be if the sub-group had the
exact age-sex disease rates as the total population. In epidemiologic terms, this expected
rate is obtained by indirect demographic adjustment. (For further information on this
adjustment process, readers are referred to epidemiology texts listed in the endnotes.) 1 2, 3
F
F
F
Step 5: Obtain a Standardized Morbidity Ratio (SMR) for each EDC.
To assess whether the observed (i.e., actual) EDC prevalence rates are different from the
age-sex expected disease rates, a type of observed to expected ratio known as a
standardized morbidity ratio or SMR is calculated. To obtain the SMR for each EDC,
divide the observed prevalence rate for the entire subgroup (reference Step 2) by the
expected prevalence rate (reference Step 4). An SMR above 1.0 indicates that the
subgroup has a higher rate of disease than the total population, even after controlling for
age and sex. An SMR below 1.0 suggests a lower than expected rate. For example, an
SMR of 1.2 indicates that the EDC was 20% more common than expected; an SMR of
0.8 indicates that the EDC was 20% less common, and so on.
Step 6: Calculate Confidence Intervals for the SMRs.
Given that subgroup size will affect the statistical stability of the SMRs, this step
suggests how a confidence interval can be calculated. A 95% confidence interval (CI)
provides a statistical measure of how reliable the SMR is. If the CI includes 1.0 in the
range, then we are 95% certain that the observed subgroup prevalence is not different
from the total population prevalence. Thus, greatest attention should be paid to those
EDCs where the SMR confidence interval does not cross 1.0.
1
Kahn HA, Sempos CT. (1989) Statistical Methods in Epidemiology, New York: Oxford University Press.
2
MacMahon B, Trichopoulos D. (1996) Epidemiology: Principles and Methods, 2nd ed., Boston: Little &
Brown.
3
Szklo M, Nieto FJ. (2000) Epidemiology: Beyond the Basics, Gaithersburg, Maryland: Aspen.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
2-12
Health Status Monitoring
Use the following calculations to determine the 95% CI for each EDC.
1. Calculate the standard error (SE) of the SMRi for the ith EDC as:
SE = sqrt (SMRi / count of expected cases for the ith EDC)
2. Using the resulting SE, calculate a 95% confidence interval around SMRi as:
CI = SMRi +/- 1.96 (SE(SMRi))
Again, if the CI crosses 1.0, then we are 95% certain that the subgroup’s SMR for that
EDC does not differ from the overall population prevalence.
Using a Combination of EDCs and ACGs to Support Case
Management and Disease Management
In addition to being used independently, there are potentially many applications where
the disease/condition specific EDCs can be effectively integrated with the ACG (and
ADG) System. EDCs are useful for identifying persons or cohorts each with a single
disease of interest (e.g., diabetes), while ACGs can be used to control for co-morbidities.
We anticipate that this method will be most relevant to users involved in case
management or disease management activities intended to improve the quality and
efficiency of care to a selected group of patients.
As a stand-alone tool, EDCs can be effectively used to identify persons with specific
conditions or explicit combinations of common conditions: for example, the number of
persons in a population with diabetes only, or diabetes and ischemic heart disease.
However, EDCs, on their own, do not provide any information about which patients with
diabetes are most costly and presumably the best candidates for disease/case
management, nor do EDCs consider the full range and extent of co-morbidities among
patients with diabetes.
As discussed in considerable detail elsewhere in this manual, the ACG Software assigns
all persons to a single ACG category. ACGs represent actuarial cells composed of
persons with similar types of co-morbidity and expected healthcare costs. In the
application described here, for the sake of parsimony, we recommend combining ACG
categories into so-called resource utilization bands (RUBs). RUBs are aggregations of
ACGs with similar expected resource use.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Health Status Monitoring
2-13
Table 7: Distribution of RUB Co-Morbidity Levels Within Selected
EDC Disease Categories and Relative Resource Use Morbidity Ratios
for Each EDC/RUB Category
Percent Distribution of
Each Co-Morbidity
Level
EDC
Prevalence
Per 1,000
Population
EDC Group
Low
(b)
Mid
High
(c)
(d)
Resource Based
Morbidity Ratio by
Co-Morbidity Level
Low
Mid
High
(e)
(f)
(g)
Total
Row
(a)
Total Population
--
49.0
27.5
4.0
0.33
1.64
9.80
1.25
Asthma (ALL02)
19.1
24.0
63.8
12.2
0.44
1.76
10.50
2.80
Hypertension (CAR02)
44.1
20.7
65.4
13.9
0.34
1.85
11.60
3.10
Ischemic heart disease (CAR03)
9.2
3.9
49.0
47.1
0.58
2.20
12.19
8.00
Congestive heart failure (CAR05)
1.5
2.6
35.1
62.3
0.58
2.33
16.47
11.70
Disorders of the lipid metabolism
(CAR11)
27.0
17.6
69.9
12.5
0.35
1.87
11.33
3.30
Diabetes mellitus (END01)
15.2
13.9
63.2
22.9
0.39
1.92
11.75
4.40
Applications Guide
The Johns Hopkins ACG System, Version 9.0
2-14
Health Status Monitoring
Table 7, Distribution of RUB Co-Morbidity Levels Within Selected EDC Disease
Categories and Relative Resource Use Morbidity Ratios for Each EDC/RUB Category,
provides a good example of the use of EDC groups and RUBs. Focusing on the total
population line, column b indicates that roughly half of the population (49 percent) falls
into a low resource category (column b), 27.5 percent fall into a medium resource group
(column c) while 4.0 percent fall into a high resource group (column d). (Note that
because non-users are excluded from this analysis that the total may not sum to 100
percent). The columns in the far right of the table, e through g, provide estimates of
resource use for these same population sub-groups. For the total population row,
individuals assigned to a low resource group consume less than half (0.33) of the
“average” or mean user while those in a medium resource group consume 1.64 times the
average. The 4.0 percent of users assigned to a high resource group, consume 9.8 times
the average. Subsequent rows of the table provide similar population distribution and
estimates of resource use for selected EDC categories. Additionally, the EDC prevalence
rate is included (column a). Such reports are useful for understanding differences across
population groupings and can help to aid in the understanding of the “why” of outlier
status. Additionally, such reports may be especially relevant to case-managers. The table
clearly demonstrates variability of costs within disease category. For example, focusing
on the EDC for hypertension, CAR02, it is clear that not all individuals with hypertension
are expensive. Rather, the small sub component (13.9 percent) who are assigned to a high
co-morbidity level are expected to have average resource use 11.6 times average. If one
is interested in maximizing efficacy of a hypertension disease management program and
maximizing ROI, it is on this group, the small sub-component with high co-morbidity
consuming significant resources, on which one should probably focus the most attention.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-i
3 Performance Assessment
Introduction ............................................................................................................... 3-1
H
H
Goals and Objectives ................................................................................................. 3-1
H
H
Theory and Background ........................................................................................... 3-1
H
H
Software-Produced Weights and Their Uses .......................................................... 3-2
H
H
Table 1: Risk Weights and Scores .......................................................................... 3-3
H
H
Concurrent ACG-Weights ........................................................................................ 3-4
H
H
Customizing Risk Scores Using Local Cost Data ................................................... 3-5
H
H
Including Part-Year Enrollees ................................................................................. 3-5
H
H
Table 2: Comparison of PMPM and PMPY Average Costs by Months
Enrolled Within an HMO Population ...................................................................... 3-6
H
H
Table 3: Comparison of Actual and ACG Expected Costs: Months of
Member Enrollment (PMPM) versus (PMPY) Weight Calculation Approaches.... 3-8
H
H
Table 4: Effect of Enrollment Period on Selected ACG-Specific
Weights .................................................................................................................. 3-10
H
H
Addressing the Impact of Age on the Calculation of ACG-Weights .................... 3-11
H
H
How to Take a Population-Based Approach to Practitioner Profiling............... 3-12
H
H
Preparatory Steps................................................................................................... 3-13
H
H
Define a Patient Panel............................................................................................ 3-13
H
H
Calculate Expected Values for the Patient Panel............................................... 3-13
Table 5: Example Calculation of Expected Values ............................................... 3-14
H
H
H
H
Calculate a Morbidity Ratio for the Patient Panel............................................. 3-15
Calculate an O/E Ratio for the Patient Panel..................................................... 3-15
Analysis of the O/E Ratios .................................................................................... 3-15
H
H
H
H
H
H
Figure 1: Comparison of ACG and Age/Gender-based O/E Ratios—
Practices of all BC Physicians ............................................................................... 3-16
H
H
Calculate Various ‘Expected’ Levels of Resource Use..................................... 3-17
Comparison of Actual to Morbidity Expected to Create a Morbidity Ratio ......... 3-17
H
H
H
H
Using Various O/E Ratios ..................................................................................... 3-17
H
H
Examples: Profiling Primary Care Physicians .................................................... 3-18
H
H
Table 6: Comparison of Patient Populations and Payments for Two General
Practitioners Identified as High-Cost Outliers, 1999............................................. 3-19
H
H
Table 7: Comparison of Case-Mix Adjusted Practice Profiles for Two General
Practitioners Identified as High-Cost Outliers in Unadjusted Analyses................ 3-20
H
H
Introducing Primary Responsible Physician (PRP)............................................. 3-21
H
Applications Guide
H
The Johns Hopkins ACG System, Version 9.0
3-ii
Performance Assessment
Comparing Specialists to Specialists – Intra-specialty Expected Level
of Service and Costs................................................................................................. 3-23
H
H
Table 8: Example Internist – Global Expenditures on the Patient Panel,
by Category of Service........................................................................................... 3-23
H
H
Using EDCs in the Context of Practitioner Profiles ............................................. 3-24
H
H
Table 9: Example of an EDC Report for an Internist ........................................... 3-25
H
H
Table 10: Example of an EDC Report for a General Practitioner ........................ 3-26
H
H
Evaluating Productivity and Distributing Workload ............................................. 3-26
H
H
Table 11: Comparison of Characteristics Affecting Physician Productivity........ 3-27
H
H
Quality of Care Assessment .................................................................................. 3-27
H
H
Figure 2: Percentage of Patients with Selected Outcomes by ACG PM
Risk Group ............................................................................................................. 3-28
H
H
Summary .................................................................................................................. 3-29
H
H
The ACG System Team is grateful to Lorne Verhulst, MD, MPA of the British Columbia
Ministry of Health in Vancouver, Canada, for his contribution to this chapter. Dr.
Verhulst joined the ACG Team as a Visiting Scholar and Postdoctoral Fellow for 20042005.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-1
Introduction
Profiling is a technique for comparing the activities of one or more health care
providers. Typically, profiling involves examination of resource utilization: dollars
spent on overall patient care or discrete services such as laboratory, pharmacy or
inpatient care. In profiling, the principal underlying question is "How does a provider's
pattern of practice compare to that of other providers once case-mix is accounted for?"
By taking into account the differences in illness burden among different providers'
patient populations, ACGs allow one to determine whether variations in practice are a
result of providers having sicker patient populations or whether these variations are
potentially attributable to differences in the way providers practice medicine.
Goals and Objectives
Practitioner profiling with the ACG System allows the user to perform the following
functions:
• Compare the patient panels of multiple physicians, controlling for:
- Case-mix and overall morbidity within the patient panel
- The practitioner’s special interests and, therefore, differences in disease prevalence
within the patient panel.
• Identify practitioners who may be contributing to excessive costs or services for their
patient panel
• Identify patient panels who may have a relative lack of access to overall physician
services or certain types of physician services
Theory and Background
Practitioner profiling has historically focused on individual providers and their patterns or
styles of practice, with a view to determining whether the practitioner is providing more
or less service (and by implication, higher cost or lower access) than he or she “should”
provide. A simple approach often used is to sum up the charges submitted by the
physician, divide by the number of patients, and calculate a crude cost per patient. This
cost per patient is then compared to an average cost per patient based upon the experience
of a health system or other level of analysis. Due to the increasing sophistication of the
health care industry, this simple practitioner profiling of expenditures per patient is no
longer sufficient.
Patient factors account for most of the cost variation and need to be taken into account
before one can make any inference about the contribution of the provider’s style to the
cost outcome. Age and gender has been used extensively to adjust for patient need for
service. However, age and gender as independent variables only explain 5 to 10% of
variation in expenditures. Continuity of care, or how much of the care is actually
Applications Guide
The Johns Hopkins ACG System, Version 9.0
3-2
Performance Assessment
provided by the individual physician, must also be taken into account. This is especially
important in an open system where patients have the discretion to seek services from
multiple physicians. A physician who provides a greater proportion of an individual’s
care will appear to be more expensive, even though the patient’s overall use of resources
may be well within expected limits. The challenge is to integrate a population-based
approach with individual provider assessment (e.g., appropriate comparisons of overall
averages to individual practitioner performance metrics while at the same time
accounting for case-mix or severity of illness burden of the practitioner’s patient panel).
To use a clinical analogy, it is important to stress that an ACG-based analysis might be
considered a screening rather than diagnostic test applied to a practitioner’s utilization
statistics. It should not be considered definitive, but rather an approach that helps an
organization to direct its resources to educational detailing, peer review or audit
programs.
The ACG System provides a means for controlling for case-mix and provides data views
and extracts that will facilitate profiling efforts. The following text will provide some
specific examples of how the ACG System can be used to support profiling. It is
recommended that each organization develop a locally-tailored profiling system suitable
to its own needs and resources and that integrates risk adjustment into the information
processing. The typical approach taken to develop a profiling system is to process the
claims and encounter data through the ACG System and then to attach the ACG-derived
patient attributes to each patient/health plan member, and then analyze that data, along
with patient- and provider- specific utilization and service data.
The examples used in this chapter apply to a retrospective analysis and are taken from
actual reports generated by the Ministry of Health Services in the Province of British
Columbia, Canada. Identifiers have been removed to protect privacy. In order to
reproduce reports as shown here, claims or encounter records must capture all of the
following fields in order to emulate the examples provided in the balance of this chapter:
Software-Produced Weights and Their Uses
In this chapter the term “weight” is used to represent a relative value for resource use
with respect to some population average and is generally expressed as a numeric value
with a mean of 1.0 (i.e., where the resource use is the same as that of the reference
population). Relative weights can be applied to mean resource use for a population to
arrive at expected resource use. Weights can be generated concurrently (i.e., for the
current period) or prospectively.
Table 1 provides a summary of the concurrent risk weights and scores produced by the
software and briefly summarizes their potential application. The remainder of this
chapter discusses the use of concurrent weights in performance assessment applications.
Table 1 begins on the next page.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-3
Table 1: Risk Weights and Scores
Metric
Description
Unadjusted Weights
Reference
Unscaled Weight
An estimate of concurrent resource use
associated with a given ACG based on a
reference database and expressed as a
relative value. Each patient is assigned a
weight based on his or her ACG.
Separate weights for non-elderly and
elderly eligible populations will be
applied depending on the Risk
Assessment Variable selected by the user.
Reference
Rescaled Weight
Local Weight
Reference weights that are rescaled so
that the mean across the population is 1.0.
A concurrent weight assigned to this
patient based upon their ACG Cd using
local cost data. The weight for each ACG
is calculated as the simple average total
cost of all individuals assigned to each
category. Local weights are calibrated to
reflect the unique properties of your
population and do not make use of
national norms.
Use
Useful in drawing external
comparisons between your population
morbidity burden and that of the
reference database. Generally, scores
greater than 1.0 indicated the case-mix
or predicted risk of your population is
sicker than the reference population
while scores less than 1.0 indicate they
are healthier.
Adjusted Weights
Applications Guide
Rescaling facilitates internal
comparisons of morbidity burden,
based on reference population, between
different subpopulations.
The Johns Hopkins ACG System, Version 9.0
3-4
Performance Assessment
Concurrent ACG-Weights
A fixed set of concurrent ACG-weights based upon the Risk Assessment Variables
selection is available as part of the software output file (see the chapter entitled,
“Installing and Using ACG Software,” in the Installation and Usage Guide for
instructions on activating this option). Separate sets of weights exist for under age 65
working age populations and for over 65 Medicare eligible populations. Which set of
weights is applied is dependent upon the user-specified options selected about which
population the user is working on (i.e., under or 65 and over). The weights produced by
the software are relative weights, i.e., relative to a population mean, and are standardized
to a mean of 1.0. An individual weight is associated with each ACG. The softwaresupplied weights may be considered a national reference or benchmark for comparisons
with locally calibrated ACG-weights. In some instances (e.g., for those with limited or
no cost data), these weights may also be used as a reasonable proxy for local cost data.
(See the following discussion regarding the importance of rescaling so that dollars are not
over predicted or under predicted.)
The software-supplied reference ACG-weights are supplied in two forms: unadjusted and
adjusted. Unadjusted ACG-weights are simply the values of the reference ACG-weights
applied to a population of interest. The mean value of the unadjusted ACG-weights
provides a rudimentary profiling statistic. If the mean of the unadjusted ACG-weight is
greater than 1.0 it indicates the rating population (the population to which the weights are
being applied) is sicker than the reference population (the national reference database). If
the mean is less than 1.0, it indicates the rating population is healthier. To ensure that
dollars in the system are not over or under-estimated, we have also made available an
adjusted or standardized ACG-weight that mathematically manipulates the unadjusted
ACG-weight to have a mean of 1.0 in the local population. The steps for performing this
manually are discussed in more detail subsequently.
Our experience indicates that concurrent (also referred to as retrospective) ACG-weights,
especially when expressed as relative values, have remarkable stability. Where
differences in ACG-weights across plans are present, it is almost universally attributable
to differences in covered services reflected by different benefit levels. The softwareprovided concurrent weights associated with the US Non-elderly Risk Assessment
Variables which were developed from a nationally representative database comprising
approximately 4.74 million lives with comprehensive benefit coverage.
If local cost data are available, the ACG Software also calculates local ACG-weights.
These local weights more accurately reflect local benefit levels and area practice patterns.
In general it is recommended that the reference population (on which the weights are
developed) should be as similar as possible to the assessment population to which the
weights are applied. However in the absence of local cost data, the reference weights
may prove useful for calculating reasonably representative profiling statistics.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-5
Customizing Risk Scores Using Local Cost Data
Two approaches for calculating ACG weights from local data are:
• PMPM (per member per month)
• PMPY (per member per year or other extended period of time)
The calculations for these two approaches are:
•
PMPM (ACG) = R (ACG) / Months (ACG)
(per member per month)
•
PMPY (ACG) = R (ACG) / N (ACG)
(per member or other extended period of time)
Where R (ACG) is calculated as the sum of resource use across all members assigned to a
particular ACG and Months (ACG) is calculated as the total number of member months of
eligibility for this cohort. N (ACG) is the number of individuals in this cohort. Weights are
calculated separately for each ACG category. The primary difference between these two
methodologies hinges on whether or not costs are annualized to account for part-year
enrollment (more on this issue later in the chapter).
The default calculation for local calibration of ACG-weights within the software is the
PMPY approach. Compared to the more widely-used PMPM, the PMPY approach
represents a new way of actuarial thinking, which is only feasible because of the use of
ICD-based adjusters such as ACGs. (Note: The per-member per-year notation or PMPY
will be used generally to reflect a per member per period approach where the extended
period may be other than a 12 month year (e.g., 10 months or 18 months)). Since PMPY
can be considered a paradigm shift in the manner by which such expected values are
usually calculated, we have attempted to provide extensive background information on
why the PMPY is preferred over the traditional PMPM approach for many risk
adjustment applications.
Including Part-Year Enrollees
The primary reason PMPY is preferred for risk adjustment is because of the way it
handles part-year enrollees.
Past work using data from multiple sites has demonstrated that persons who are enrolled
for fewer than 12 months in a health plan during a given year tend to use more resources
on a PMPM or annualized basis than those who are continuously enrolled for the entire
period. New, previously uninsured enrollees may have higher costs as a result of
previously unmet needs or they could be switching plans in the midst of a special
healthcare episode (e.g., they could be responding to a newly diagnosed condition).
Shorter-term enrollees as a group also exhibit higher costs in part because they include
those who leave a plan either because they have special medical circumstances or, at the
extreme, die. In addition to these circumstances, as the following tables will illustrate,
shorter-term enrollees have seemingly higher PMPM costs in large part because the
denominator of the PMPM calculation is relatively smaller for those enrollees. By
Applications Guide
The Johns Hopkins ACG System, Version 9.0
3-6
Performance Assessment
contrast, the average cost of 12-month enrollees tends to be more stable. The following
analysis illustrates the implications of this within the context of diagnosis-based risk
adjustment such as ACGs.
Table 2 presents a side-by-side comparison of the PMPM and PMPY costs of enrollee
sub-groups defined in terms of months enrolled during a given recent year at a large
commercial HMO. The table is limited to those who used services because retrospective
analyses (e.g. provider profiling) are typically limited to those who actually used
services. The average PMPM costs for the enrollee cohorts decrease as the length of
enrollment increases. Those who were enrolled for 12 months used $86.95 PMPM while
those enrolled for only one month used $768.92 PMPM, illustrating almost a nine-fold
difference between twelve-month and one-month enrollees. Viewed from this
perspective, it would appear that it is important to account for months enrolled when
examining the pattern of costs over a given time period. In contrast, there is less than a
two-fold difference between those enrolled for 1 and 12 months on a (non-annualized)
PMPY basis. As would be expected, those enrolled for very few months tend to have
lower within-plan annual average costs, but this effect is less marked than the differential
found when PMPM values are compared.
Table 2: Comparison of PMPM and PMPY Average Costs by Months
Enrolled Within an HMO Population
Months
Enrolled
1
2
3
4
5
6
7
8
9
10
11
12
Persons
488
934
1,517
1,411
1,601
1,701
2,027
1,550
1,781
1,941
1,355
70,786
87,092
Months
488
1,868
4,551
5,644
8,005
10,206
14,189
12,400
16,029
19,410
14,905
849,432
957,127
% Months
0.1
0.2
0.5
0.6
0.8
1.1
1.5
1.3
1.7
2.0
1.6
88.7
100
$ PMPM
$ PMPY
768.92
438.65
212.53
198.55
157.91
144.00
136.47
140.35
125.45
105.65
105.22
86.95
93.18
768.92
877.29
637.59
794.21
789.55
863.99
955.27
1,122.80
1,129.09
1,056.46
1,157.43
1,043.40
1,023.99
Total
Notes:
Cost includes total paid claims truncated at $35,000.
The population was limited to service users in a large commercial HMO population for 1996.
PMPM = Per member per month
PMPY = Per member per year. (Note: Although 12 months were used here, other extended periods could
also be used to calculate per-member-per-period weights.)
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-7
When diagnoses are assigned on a concurrent basis and partial year enrollees are included
in the analysis, the denominator in the PMPM calculation tends to skew the relationship
between actual and expected costs, particularly when performing retrospective analyses
such as provider performance profiles. As previously described, PMPM ACG weights are
calculated by determining the costs associated with each ACG divided by the total
member months associated with that ACG. The total expected costs associated with any
given individual, in this case, would be the PMPM ACG weight times the number of
months enrolled. Alternately, ACG weights derived on a PMPY basis are calculated as
the costs associated with each ACG for the analysis period divided by the number of
persons associated with that ACG. Therefore, total expected costs associated with any
given individual would be independent of the time enrolled during the analysis period.
Based on total paid costs truncated at $35,000 (to mimic stop-loss reinsurance levels in
this plan), ACG weights were calculated using both the PMPM and PMPY alternative
approaches for the population shown in Table 2. Based on each of these approaches,
actual costs were compared to expected ACG costs within that population. Sections A
and B of Table 3 present a series of measures comparing actual to expected costs for
cohorts of enrollees defined in terms of the months they were enrolled during a 12-month
period. This table, as does the previous one, represents a retrospective cohort analysis of
users as appropriate for a provider profiling assessment.
Section A of Table 3 presents the results using a PMPM calculation. The column labeled
“% deviation” reflects expected costs divided by actual costs minus one. For persons
enrolled for one month, the (85.1) figure indicates that when the actual (1996) costs of
these 488 single month enrollees are compared to their ACG expected costs (calculated
on a PMPM basis), the cohort would have been underpaid by 85.1 percent, on average. In
contrast, persons who were enrolled for the full 12 months of the year were overpaid, on
average, by 5.3 percent. The “% deviation” column is expressed in absolute dollars in the
column labeled over (under) $000. Section A of Table 4 illustrates a shift of expected
dollars from part-year enrollees to 12-month enrollees. The net result of this for profiling
applications is that subpopulations that include a disproportionate number of shorter-term
enrollees will look inefficient because the associated expected dollars calculated on a
PMPM basis will tend to be lower than their actual costs. Conversely, a population
comprised exclusively of 12-month enrollees will be overpaid and appear to be efficient
because of the shift of expected dollars embedded in the PMPM calculation.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
3-8
Health Status Monitoring
Table 3: Comparison of Actual and ACG Expected Costs: Months of Member Enrollment (PMPM)
versus (PMPY) Weight Calculation Approaches
Months
Enrolled
1
2
3
4
5
6
7
8
9
10
11
12
Months
(A) Using A PMPM Calculation
(B) Using a PMPY Calculation
Over (Under)
$000
Over (Under)
$000
% Deviation
488
1,868
4,551
5,644
8,005
10,206
14,189
12,400
16,029
19,410
14,905
849,432
957,127
(85.1)
(73.5)
(57.5)
(52.5)
(39.1)
(33.9)
(27.7)
(18.1)
(12.2)
(0.1)
13.7
5.3
(0.0)
(319)
(603)
(556)
(589)
(495)
(498)
(537)
(314)
(245)
(3)
214
3,943
(0)
Adjusted
R-squared
0.013
0.109
0.156
0.226
0.326
0.375
0.312
0.446
0.382
0.371
0.465
0.380
0.338
% Deviation
8.6
(0.5)
14.3
0.9
8.0
0.5
(3.0)
(3.8)
(3.2)
(0.5)
3.1
(0.2)
(0.0)
32
(4)
139
10
102
8
(59)
(67)
(64)
(11)
48
(134)
(0)
Adjusted
R-squared
0.327
0.408
0.369
0.386
0.442
0.509
0.392
0.545
0.411
0.385
0.553
0.385
0.395
Total
Notes:
Costs include total paid claims truncated at $35,000.
The population was limited to service users in a large commercial HMO population for 1996.
Total absolute error was $8.3 million using a PMPM calculation and $677,000 PMPY calculation. See text for a description of these calculations.
PMPM = Per member per month
PMPY = Per member per year. (Note: Although 12 months were used here, other extended periods could also be used to calculate per-member-per-period
weights.)
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-9
Section B of Table 3 shows the results using a PMPY calculation. While there is a slight
overpayment associated with shorter-term enrollees (e.g., one month enrollees are
overpaid by 8.6 percent on average), the extent of the deviation between actual and
expected costs is markedly lower for each subgroup (i.e., each row) as a result of using
the PMPY orientation. The sum of the absolute error of each enrollment cohort reflected
in section B of the table is less than $700,000 while the comparable figure is $8.3 million
reflected in section A.
R-squared (R2) is a measure of the extent to which expected values explain variation in
actual costs. The R2 for the population as a whole using a PMPM calculation is .338
(shown in the row labeled Total in Section A of the table), and this measure decreases
with shorter-term enrollment, particularly for those with less than five months of
enrollment. The R2 is higher using a PMPY calculation (.395 in section B of the table)
and remains largely stable regardless of the length of time a patient has been enrolled.
The modest tendency of the PMPY approach to overpay or inflate expected costs
associated with very short-term eligibility (e.g., one to three months of enrollment)
reaffirms that time has some effect on the calculation of diagnosis specific expected
values.
To examine the nature of this effect in more detail within this case-study population,
Table 4 presents the average costs per-person and the number of persons by three-month
enrollment windows for selected ACGs. Some ACGs have relatively low mean costs
given shorter-term enrollment, as opposed to costs for all cases during the full period (a
year). At the same time, many ACGs are quite stable regardless of time enrolled,
particularly for persons enrolled more than three months. The highest morbidity/highest
cost ACGs (e.g. ACGs 4940-5070) tend to be uncommon for those enrolled for the
shortest periods, but nonetheless are fairly consistent (in terms of average costs per
period) across the enrollment windows, even given the small numbers of cases for shorter
periods of time. Generally, much of the variability in average costs probably can be
attributed to the very small sample size in the shorter enrollment columns. Again, while
enrollment time has an influence on costs associated with some ACGs, the general
consistency of costs across the columns in Table 4 and the relatively limited number of
persons with less than 12 months enrollment tend to limit the overall plan-wide effect of
time on risk adjusted concurrent analyses. However, analyses where some sub-cohorts
include a disproportionate number of short-term enrollees are likely to undervalue
expected costs for those groups. In any event, such analyses should be approached
cautiously because of the instability associated with the shorter-term enrollment.
In summary, when performing concurrent (or retrospective) risk-based adjustment, a
PMPM calculation of ACG weights for a population that includes some number of parttime enrollees tends to over-represent the expected costs associated with 12 month
enrollees and under-represent the expected costs associated with shorter-term enrollees.
A PMPY calculation of concurrent ACG weights appears to provide a more accurate
measure of the expected weight. As noted earlier, we believe this empirical observation
represents a relatively new paradigm, and we encourage analysts performing profiling
Technical User Guide
The Johns Hopkins ACG System, Version 9.0
3-10
Performance Assessment
and other concurrent analyses to test whether and how such an approach could replace the
PMPM approach within their organization. The Johns Hopkins ACG Development Team
expects to continue providing empirical findings and support material regarding this
innovation.
Table 4: Effect of Enrollment Period on Selected ACG-Specific
Weights
ACG
All
1-3 Months
Avg$ Cases
736
2,939
4-6 Months
Avg$ Cases
818
4,713
7-9 Months
Avg$ Cases
1,062 5,358
10-12 Months
Avg$ Cases
1,046 74,082
All Enrollees
(users)
Avg$ Cases
1,024 87,092
200
66
62
111
95
115
71
153
969
143
1,197
400
275
163
300
192
287
202
353
2,222
340
2,779
500
137
264
131
335
169
316
182
3,743
175
4,658
800
510
27
322
15
973
18
785
166
736
226
1300
173
58
217
65
232
57
265
599
252
779
1600
97
272
110
395
119
382
119
4,195
117
5,244
1711
3,186
12
3,412
27
3,791
22
4,155
193
3,998
254
1712
241
35
390
35
890
26
782
149
660
245
1752
422
2
1,129
7
4,212
13
3,552
95
3,427
117
1800
316
106
498
207
654
225
584
3,417
576
3,955
2400
267
15
225
46
206
55
223
1,268
223
1,384
2500
268
20
259
40
256
35
402
571
381
666
3200
865
35
858
106
1,012
141
1,028
2,300
1,018
2,582
3500
493
10
390
27
607
29
793
686
767
752
3600
2,111
17
1,656
29
1,406
66
1,876
1,506
1,855
1,618
3900
702
43
457
63
474
86
590
803
577
995
4100
610
116
838
206
702
228
692
4,986
696
5,536
4220
1,796
3
1,344
28
1,017
21
1,328
553
1,320
605
4320
1,498
23
2,274
54
1,811
82
1,709
1,192
1,735
1,351
4330
5,787
8
5,360
7
1,754
19
2,515
252
2,625
286
4410
553
7
742
30
1,450
37
1,037
1,476
1,039
1,550
4420
1,805
13
1,535
24
2,485
35
1,741
1,108
1,760
1,180
4430
12,039
6
10,454
8
7,145
16
5,803
260
6,134
290
4510
297
1
666
1
1,600
15
1,818
186
1,789
203
4910
6,071
4
1,938
42
2,795
77
2,372
2,824
2,382
2,947
4940
18,946
4
19,979
5
25,181
5
16,363
60
17,343
74
5030
0
0
0
0
0
0
13,554
41
13,554
41
5040
0
0
1,234
2
4,317
11
4,165
336
4,153
349
5050
0
0
5,430
2
7,330
5
7,245
261
7,218
268
5060
0
0
11,243
4
16,426
4
11,887
222
11,954
230
5070
0
0
24,892
5
27,790
11
20,766
140
21,393
156
5110
64
67
40
53
54
33
46
541
48
694
5310
1,195
413
1,253
483
1,563
369
1,563
200
1,357
1,465
5320
4,416
70
5,553
40
5,036
41
5,811
18
4,984
169
5340
11,121
12
12,454
29
9,936
40
8,316
39
10,136
120
Notes:
Average mean costs include total 1996 paid claims truncated at $35,000 for users in a large commercial HMO
population.
These figures reflect a retrospective/concurrent analysis.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-11
Addressing the Impact of Age on the Calculation of ACG-Weights
Age is incorporated as a control variable in the sorting algorithm that determines final
ACG assignment. At the same time, there are some ACGs that include both pediatric and
adult populations because splitting on age was not consistently found to contribute to
variation explained within those categories. Despite this, pediatric populations (those
younger than 18) tend to generate fewer costs than adult populations within broadly
defined commercial populations.
Where ACG-based applications are stratified by pediatric versus adult populations, riskadjusted resource weights derived from the population as a whole may over- or underrepresent expected values associated with these groups. For example, in profiling primary
care providers, weights derived from a broadly defined population may over-represent
expected values for physicians whose practice is limited to pediatric cases. Those
providers will, on average, tend to look more efficient than providers for the health plan
as a whole.
One common way to address this issue is to calculate ACG weights separately for
pediatric and adult cohorts within a health plan. For example, two weights could be
calculated for ACG0500, Likely to Recur, without Allergies. One ACG weight would be
based on the resource used by adults who were assigned to ACG0500. The second ACG
weight would be based on similar data but restricted to those under age 18. Note: Only
those ACGs not automatically split by age are affected.
Technical User Guide
The Johns Hopkins ACG System, Version 9.0
3-12
Performance Assessment
How to Take a Population-Based Approach to Practitioner
Profiling
Necessary Data Elements
1. Unique provider ID
2. Provider specialty
3. Unique patient ID
4. Individual patient age
5. Individual patient gender
6. Referring physician on claim for consultation and or diagnostic test
7. ICD code for the principle reason for the visit or service
8. ICD codes for the secondary and even tertiary reasons for the visit or service
(optional but highly recommended)
9. ACG assignment for each patient
10. EDC assignments for each patient
11. Service ID – a consistent and systematic service nomenclature, such as the
Common Procedural Terminology (CPT). Some health plans (e.g., each Canadian
province) may have their own payment schedule codes, and these could also be
used
12. Assign each CPT or Payment Schedule code a service specialty type. Examples:
a. General practice “type” of service
b. General internist’s “type” of service
c. Thoracic surgeon’s service
Note: The type of service is irrespective of the specialty of the physician
providing the service. For instance, an appendectomy is a general surgical
“type of service” irrespective of whether it was done by a general practitioner,
a gynecologist or a general surgeon.
13. Service category type. Each of the above specific “types” can be grouped into
service categories. Examples:
a. Specialists consultations
b. Diagnostic services
i. Imaging
ii. Blood chemistry
iii. Anatomic pathology
iv. Other
c. Major surgical procedure
d. Minor surgical procedure
14. Date of service
15. Value of the service
Notes:
• If services are not all reimbursed as a fee-for-service item, an encounter record and
a measure of the service’s relative value and a conversion to a monetary unit (e.g.,
Dollars, Euros, etc) are needed.
• Incentive payments attached to services, such as “pay for performance,”
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-13
reimbursement for after hours or emergency call-outs, or subsidies paid for rural
and remote practices should be treated as a separate category and not considered in
the value of the service itself.
Preparatory Steps
1. Having run the ACG Software ON THE GLOBAL MEMBERSHIP in your health
plan, capture the ACG-derived attributes of each patient (their ACG and EDCs)and
link this information (relate or merge) to the unique patient ID.
2. Attach a type of practice marker to every CPT code.
3. To develop physician peer groups, assign physicians to type of practice groups based
on functional specialty – that is the type of practice that you assign to the CPT or
Payment Schedule codes that represent the largest proportion of their activity.
4. Group patients by whether or not they attended a physician in the peer group.
Patients can be assigned to multiple types of practice peer groups.
5. For each set of patients (attending physicians of a given peer group) re-calculate their
mean costs per age/gender group and per ACG group in each of the service
categories. These new means will differ from the ACG means and age/gender means
that were calculated using the entire population. These new values will be used to
calculate expected costs for the physician patient panels. This step does not
necessarily require re-running the ACG Software.
Define a Patient Panel
The sub-population of interest will be every member of your health plan who had a
contact with a given individual provider. A given member may thus be on several panels.
In a gatekeeper model, patients may be assigned to a single PCP. However, if the
member elects to change their PCP in the middle of the reporting period, or if the model
is an open choice model, some assignment of provider will need to be made. The
objective is to assess resources used (claims experience) by the panel and in that context,
what contribution or effect the individual (subject) practitioners had on the overall costs.
Summarize the costs incurred by patients in the panel by service category.
Calculate Expected Values for the Patient Panel
1. Age gender expected
2. ACG expected
To get the age gender expected, use the new means you calculated in preparatory step 5
for each age gender group, multiply by the number of patients in the practitioner’s panel
in each age gender group, and then sum them. Repeat this procedure for calculating the
ACG expected values. Repeat these steps for each category of service.
This procedure mathematically tells you what the expenditure for each category of
service would have been had each member of the panel incurred expenditure that was
Technical User Guide
The Johns Hopkins ACG System, Version 9.0
3-14
Performance Assessment
average for his or her age gender group and ACG group, respectively. The procedure is
known as “indirect standardization” and is best illustrated by the simplified example
shown below.
Table 5: Example Calculation of Expected Values
Total Expenditure
Average Expenditure
Incurred by Members of
Per Person in Risk
the Risk Group
Group
1
10
440
44.00
2
44
550
12.50
3
23
660
28.70
4
16
770
48.13
5
99
880
8.89
6
65
990
15.23
7
41
800
19.51
8
11
700
63.64
9
5
600
120.00
10
2
400
200.00
316
6,790
Total
Call this group of 316 persons the “reference population.” In actual practice, the reference population
will be much larger, usually the entire population of a health plan, or a larger jurisdiction if data is
available.
Crude Average per capita expenditure = 6,790/316:
21.49
Apply the reference population average per risk group to the risk groups of the subset of the population,
accepting that this subset has a different distribution among risk groups. The arrow indicates that the
averages derived from the reference population are then placed in the calculation of expected costs in
the subpopulation of interest.
Risk Group
ID Number
Number of Persons
in Risk Group
Total “Expected”
Expenditure for
Members of Risk Group
1
5
220.00
2
10
125.00
3
15
430.43
4
20
962.50
5
25
222.22
6
30
456.92
7
21
409.76
8
14
890.91
9
6
720.00
10
2
400.00
148
4,837.75
Total
Contrast this to the “expected” based on the crude per capita average expenditure. The total is different
because the risk structure is different.
Persons
Crude Per Capita Average
Total Expected
148
21.49
3,180.13
Total
The same logic is applied whether the risk group is age/gender groups, ACG groups, or any other
characteristic.
Risk Group
ID Number
Number of Persons
in Risk Group
The Johns Hopkins ACG System, Version 9.0
Reference Average
Expenditure Per Person in
the Risk Group
44.00
12.50
28.70
48.13
8.89
15.23
19.51
63.64
120.00
200.00
Applications Guide
Performance Assessment
3-15
Calculate a Morbidity Ratio for the Patient Panel
Divide the ACG expected cost by the age/gender expected cost to obtain a ratio. This
ratio tells you how much more or less expensive the patients are expected to be as
compared to the average for their age and sex, as a result of their ACG assignments.
Calculate an O/E Ratio for the Patient Panel
Sum the observed (actual) costs in each cost category for the patients on the physician’s
panel. Divide the sum of actual cost by the sum of expected cost for each cost category
to obtain an O/E ratio. (Refer to Table 5).
Analysis of the O/E Ratios
You may apply your favorite statistical package to analyze the distribution of these O/E
ratios in order to define a statistical exception limit (2 or more Standard Deviations is
often used). Check to determine whether the distribution is normal or skewed. Even in
skewed distributions one may calculate a standard deviation, but the usual assumptions,
such as the rule that 95% of the data are within +/- 1.96 SDs of the mean, will not apply.
Another measure of an outlier value is to use percentiles, such as quartiles. The third
quartile (75th percentile) + 1.5 times the interquartile range (IQR) (the difference between
the 75th percentile and the 25th percentile, that being the range that surrounds the middle
50% of values) is a commonly accepted definition of an outlier in a right-skewed
distribution. Cost data are often right skewed. However, the distribution of the ACG
based O/E ratios ought to resemble more closely a normal distribution. See Figure 1 for
an example.
Technical User Guide
The Johns Hopkins ACG System, Version 9.0
3-16
Performance Assessment
Figure 1: Comparison of ACG and Age/Gender-based O/E Ratios—
Practices of all BC Physicians
Source:
Verhulst L, Reid RJ, Forrest CB. Hold It - My Patients are Sicker! The Importance of
Case Mix Adjustment to Practitioner Profiles in British Columbia.@ BC Med Journal.
July/August 2001; 43(6):328-333.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-17
Calculate Various ‘Expected’ Levels of Resource Use
1. Crude expected – based on mean per capita cost without adjustment.
2. Age sex adjusted ‘expected’. Using indirect age/gender standardization, determine
what the patient panel experience would have been if they had average resource use
for their age sex relative to your standard population. In general the standard
population is all the members of your entire health plan. However, other standard
populations may be used if you have the data. Table 5 provides an example of how
this is done.
3. ACG expected. Using indirect standardization, using ACG categories rather than age
gender categories, determine what the patient panel experience would have been if
they had average resource use for their ACG group relative to your reference
population.
Comparison of Actual to Morbidity Expected to Create a Morbidity
Ratio
The Morbidity Ratio (ACG expected/age sex expected) tells you in relative terms the
contribution of ACG categories (co-morbidity) to the panel’s expected experience after
accounting for age and sex. Simply put, the following interpretations arise from
morbidity ratios:
• A morbidity ratio = 1.0 occurs when patients in the panel, on average, fall into ACG
co morbidity categories that are average for their age sex.
• A morbidity ratio <1.0 occurs when patients in the panel, on average, fall into ACG co
morbidity categories that are less expensive (lower burden of disease) than average for
their age sex.
• A morbidity ratio >1.0 occurs when patients in the panel, on average, fall into ACG co
morbidity categories (higher burden of disease) than average for their age sex.
Using Various O/E Ratios
Compare the actual (observed) experience to the expected levels. These comparisons are
referred to as Observed to Expected or (O/E) ratios with the various E’s:
a. O/E with mean per capita cost
b. O/E with age-sex means
c. O/E with ACG means
Why do we offer three O/E ratios? Mean expected and age sex adjusted expected costs
are easy to do and easy to understand and communicate. They also can be used to test the
face validity of the conclusions you make about morbidity at the panel level. Obviously,
age-sex expected costs will be higher than crude per capita expected for panels with older
than average age. Note: The relationship between the age-sex-based expected and the
ACG-based expected is characterized by the Morbidity Ratio. For example, if the
patients are “healthy elderly” they will have lower morbidity ratios. Age sex adjustment
is unequivocal in terms of patient assignment to categories. In general, age adjustment
Technical User Guide
The Johns Hopkins ACG System, Version 9.0
3-18
Performance Assessment
and ACG adjustment should move in the same direction relative to the crude per capita
estimates in the vast majority of patient panels. ACG assignment modifies how much the
expected varies from the crude unadjusted estimate.
Examples: Profiling Primary Care Physicians
One of the reports generated for a patient panel is a global or summary view of their total
claims experience. The practitioner’s contribution to that total is a subset or subtotal of
that experience. This is done in order to gain an insight into the practitioner’s
contribution to the total outcome. In certain venues an individual practitioner may
provide a high proportion of the overall service.
\ Example: The lone cardiologist in a small city that serves as a regional resource may
provide the bulk of cardiology care. Compare that to the large metropolitan area,
where multiple cardiologists may have participated in the care of the patient panel. In
more traditional methods of profiling, the lone cardiologist may have appeared
relatively more expensive. What is important is whether the patients received a
reasonable total level of service overall, rather than how much of the expenditure was
related to the services of the lone cardiologist in this example.
In terms of primary care, a rural practitioner may similarly have little sharing of patients,
whereas an urban practitioner’s patients may have multiple sources of primary care. To
illustrate this point, Tables 6 and 7 provide a true to life example of a report on total
physician claims experience of the patient panel of two general practitioners. These
tables were originally published in the British Columbia Medical Journal. 1
F
1
Verhulst L, Reid RJ, Forrest CB, Hold it—my patients are sicker! The importance of case-mix adjustment
to practitioner profiles in British Columbia BC Medical Journal Volume 43, Number 6, July/August 2001,
pages 328-333
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-19
Table 6: Comparison of Patient Populations and Payments for Two
General Practitioners Identified as High-Cost Outliers, 1999
Table 6 shows that these two practitioners had costs per patient that were approximately
twice their peer group average before any adjustments. The percentile rank of this
observation was very high (above the 98th percentile). Looking only at the patients who
received the majority of their care from these two physicians, the cost per patient is still
approximately 1.5 times the peer group average, and still ranks very high (above the 95th
percentile).
Practitioner 1
Practitioner 2
A. All patients seen by practitioner
Total number of patients
Total MSP payments made to practitioner for GP
services*
Average payments per patient
Average payments per patient in peer group†
Practitioner's cost per patient rank in peer group
(percentile)
1360
$252,259
755
$150,443
$185
$99
$199
$99
98.1
99
B. Patients for whom practitioner is the "Primary Responsible Physician" (PRP)‡
Number of PRP patients
790
Percentage of total patients
58.1%
Total MSP payments to all MDs for GP services for
$228,715
practitioner's PRP patients
Percentage of total GP service payments to practitioner
for PRP patients
451
59.7%
$158,424
93.7%
85.0%
Average payments for PRP patients
$271
$298
Average payments for PRP patients in peer group
$192
$192
Practitioner's cost per patient rank in peer group
(percentile)
96.3
98.2
Technical User Guide
The Johns Hopkins ACG System, Version 9.0
3-20
Performance Assessment
Table 7: Comparison of Case-Mix Adjusted Practice Profiles for Two
General Practitioners Identified as High-Cost Outliers in Unadjusted
Analyses
Practitioner 1
Number of patients where practitioner is the primary responsible
physician (PRP)*
Practitioner 2
790
451
A. Observed payments for PRP patients
Total MSP payments to all MDs for GP services†
$228,715
$158,424
Percentage of total MSP payments to practitioner
93.7%
85.0%
96.2
98.2
$234,080
134.0%
$102,847
91.5%
99.7
19.6
97.7%
53.5
154.0%
99.0
Practitioners rank in "observed" per-patient costs in peer group‡
(percentile)
B. Expected payments for PRP patients based on ACGs§
Total expected MSP payments for GP services
Ratio of "ACG expected" to "age/gender expected" for all GPs
(morbidity ratio**)
Practitioner's rank in morbidity ratio in peer group (percentile)
C. Practitioner's O:E ratio (efficiency ratio)¶
Observed to expected cost ratio (O:E ratio)
Practitioner's rank of O:E ratio in peer group (percentile)
Notes to tables:
“* “Primary Responsible Physician patients” is defined as those for whom the
practitioner provides the majority of the GP services. In other words, these are the
patients for whom the practitioner is most responsible.
† Includes all services grouped as general practice services in fee-for-service payment
schedule.
‡ The practitioners' peer group is defined as practitioners who derive the majority of their
revenue on general practice “owned” fee items in the MSP Payment Schedule
§
Expected costs are defined as the total cost of services if all patients consumed an
average amount for their ACG.
** The ACG morbidity ratio is defined as the average expected per-person cost for the
practitioner's panel of patients based on ACGs divided by the average expected perperson cost based on age/gender for all practitioners. It can be interpreted as how sick the
practitioner's panel of patients is compared to the average.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-21
The observed-to-expected cost ratio (or efficiency ratio) is defined as the total observed
costs for a practitioner's panel of patients divided by the expected costs if all patients were
average for their ACG. It can be interpreted as the costliness of a provider's practice after
removing the effects of morbidity.” 2
F
What these examples illustrate is that high cost per patient may be largely explained by
the burden of illness within the patient panel. The two physicians’ practices in this
example have very different O/E ratios, and very different Morbidity Indices. Table 7
and the underlying analysis focused on the patients who received the majority of their
primary care services from the subject physician. In the case of physician 2, patients’
morbidity appeared to be close to average for the peer group. Expected costs, calculated
as the sum of the ACG average cost for each respective patient, were $102,847, while the
observed cost was $158,424, or 58% more than expected. In contrast, physician one had
patients whose ACG expected cost greatly exceeded age and gender expected cost
(134.0%). The morbidity ratio was 134%, which ranked very high (>99th percentile) in
keeping with the higher ACG expected costs. Observed cost as a percentage of expected
cost was 97.7%, and as a result, the physician who has sicker patients no longer appears
as a statistical outlier.
Introducing Primary Responsible Physician (PRP)
The preceding example introduced the concept of Primary Responsible Physician (PRP).
Determining a primary responsible physician may be of special significance to physicians
or health care organizations that are subject to profiling. They may prefer to be evaluated
only on the basis of patients for which they had implicit or explicit responsibility for their
care (care gatekeepers would have explicit responsibility). While the following text offers
a method for determining a PRP relationship between a patient and a physician, this is not
the only way this problem can be approached. Given that care is often supplied by teams
of clinicians, one could form panels of patients who had ANY contact with a particular
provider. Indeed, an approach based on PRP and an approach based on any contact both
provide information and could be used in tandem for profiling. For instance, much could
be learned by comparing a PRP profile for a general practitioner with a profile that
included their referrals.
In determining the PRP, first decide, on the basis of services, what constitutes a primary
care service. Assign a primary care marker to the relevant codes, such as the CPT or
Payment Schedule code. Distinguish between Evaluation and Management (E&M) items
done upon consultation/referral versus those that are patient self-referred. That indicates
whether the specialist was acting in a primary care role or a consulting role at the time of
the service.
2
Verhulst L, Reid RJ, Forrest C, ibid.
Technical User Guide
The Johns Hopkins ACG System, Version 9.0
3-22
Performance Assessment
Classify a member to a PRP only if he or she has 3 or more visits. Low users during the
period of analysis (typically one year) are not assigned a PRP physician. A person who
has only one visit in a year will have only one physician. Persons with one or two visits
are equally likely to have one or two primary care physicians. 3 For the patients with
3 or more primary care services (visits) calculate a simple ratio for each patient-doctor
combination.
F
Ratio (patient 1 & Dr. A) = [visits to Dr. A / total visits by patient] (in the primary
care service range)
The sum of an individual patient's ratios = 1. Only if the ratio for the nth doctor-patient
combination is ≥ 0.5 does the patient qualify as a PRP patient of that doctor. You may
wish to create a separate category for patients with a ratio of 0 up to 0.49. The largest
ratio <0.50 is called the “largest non-Primary Responsible Physician.”
The connection between this doctor and patient is weaker than the PRP connection, but
stronger than all other connections. This is especially true when the individual is a
relatively higher user of services.
Tie-breaker rules may be needed. One, if the service counts are equal, assign the patient
to the physician with the highest charges. The rationale is that a more comprehensive or
time-consuming service implies a stronger relationship or likelihood of ongoing contact.
Two: all else being equal, assign the patient to the most recent physician contacted. This
adjusts for people recently changing physician or relocating and the likelihood of ongoing
contact.
The pattern that emerges can be used to assess the style of the physician's practice. It
distinguishes practices with a high level of continuity of care (high proportion of PRP
patients) to less continuity of care (low proportion PRP patients.) Table 8 shows how the
proportion of PRP patients can be tabulated. The distribution of the PRP proportion and
the physician's rank on that proportion within his or her peer group can also be shown.
The proportion of patients who are PRP patients serves as an indicator, or index of
continuity of care.
The continuity of care index is highly variable. It will be high in rural practices and
traditional family practice venues with a small practice group. It will be low if the
practitioner is very itinerant as, for example, one who does frequent locum tenens.
3
Unpublished data from the Ministry of Health, British Columbia.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-23
Comparing Specialists to Specialists – Intra-specialty
Expected Level of Service and Costs
Within a case mix category a selection of sicker patients may take place through the
process of referral from a primary care practitioner to a specialist. Some profiling
applications use an intra-specialty measure of expected. To do this, one calculates the
expected values based not on the overall plan mean for all patients in the respective ACG,
but rather, a mean based on the subset of patients (members), matched by ACG and who
were seen by specialists within the practitioners’ own specialty. For example, patients
with chronic renal failure who are referred to nephrologists are typically sicker than
patients with the same diagnosis who are cared for only by the PCP. To assess the
efficiency of nephrologists, it is important to compare their observed cost compared to
expected, after case-mix adjustment, with the patients of other nephrologists.
The following example is taken from an actual physician profile from calendar year 2003,
provided courtesy of the Ministry of Health Services in British Columbia, Canada. The
examples have been severed of personal identifiers. The first example (Table 8) is that of
an Internist. It is intended to demonstrate the value of placing the practitioner’s claims
experience in the context of the global claims experience of all the patients that he saw.
Table 8: Example Internist – Global Expenditures on the Patient
Panel, by Category of Service
General
Practice
Service
Patient count (%) for
whom this practitioner
provided a service
Patient count total,
including service provided
by other physicians (i.e.,
upon referral by the
subject physician and
“intersection” with other
panels)
Cost (payments) to subject
physician
Referred services costs
where the subject
physician made the
referral
Costs (payments to) all
other physicians
Total observed costs
ACG expected costs
Total Observed to
Expected (ACG-based
Technical User Guide
Specialist
Consultation
Diagnostic Procedure
Total
Test
and Other Physician
3 (<1)
907(91)
0
585(58)
942(94)
942(94)
965(96)
995(99)
807(80)
1,000(100)
$666
$183,990
$49,803
$234,450
N/A
$26,291
$203,131
N/A
$229,442
$277,267
$139,109
$345,082
$348,997
$1,110,455
$279,574
$349,979
$548,559
$399,260
$1,577,372
$355,234
$207,407
$533,339
$520,145
$1,616,127
78.7%
168.7%
102.9%
76.8%
97.6%
The Johns Hopkins ACG System, Version 9.0
3-24
Performance Assessment
General
Practice
Service
Specialist
Consultation
Diagnostic Procedure
Total
Test
and Other Physician
expected)
Interpretation notes using the preceding example
Including patients not seen personally, but who had a referred service upon referral by
this physician, the physician had 1000 patients. The health plan total physician claims
experience ($1.58 Million) for this group was very close to ACG expected overall.
However, the distribution of those costs was somewhat unusual, with less than expected
general practice costs (79%), procedure and other costs (77%), close to the expected
amount of diagnostic tests costs (103%) and much more than expected specialist
consultation costs (169%). Under the specialist consultation category $183,900 out of a
total of $349,979 was paid to this particular physician. Subsequent audit could be
directed at the question as to whether this physician had assumed primary care
responsibility for the patients and perhaps was charging follow-up visits as if they were
referred for consultation. Although, the mix of services may have been unusual, the
overall expenditure was not unusual.
Using EDCs in the Context of Practitioner Profiles
The following table (Table 9) shows an excerpt from the EDC section of the physician’s
practice in the preceding example. The EDCs are sorted in descending order of their
prevalence within the 1000 patients that comprise this physician’s panel. The prevalence
of these conditions or disease groupings can be compared to the average prevalence in the
panels of physicians in his peer group. A co-morbidity score is also given. Taking the
698 diabetic patients in this example, this is simply the ACG expected costs for these 698
patients, divided by the ACG average costs for all diabetic patients seen by internists
multiplied by 698.
If patients have more ADGs (the building blocks of ACGs) and more major ADGs, they
will have higher ACG expected costs. This example supports a hypothesis that the
diabetic patients of this physician may have less co-morbidity than average diabetic
patients who visited internists.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-25
Table 9: Example of an EDC Report for an Internist
Rank/
MEDC
MEDC/EDC
1 END
Endocrine
END01-Diabetes Mellitus
END04-Thyroid Disease
2 CAR
Cardiology
CAR02-Hypertension
CAR01-Cardiovascular
Signs and Symptoms
Patient
Count
MEDC/EDC
As % Of
This Practice
MEDC/EDC
% Of
Practice In
Peer Group
CoMorbidity
Score
731
73.1
23.2
67.4
698
41
69.9
4.1
15.4
6.0
65.9
80.4
628
62.8
58.5
75.4
463
130
46.3
13.0
30.7
23.4
72.7
87
Notes:
1.
The prevalence of diseases/symptoms is shown whether the subject physician or other physicians
made the diagnosis.
2.
The co-morbidity score is based on the mean ACG expected costs of patients in MEDC or EDC who
have been seen by a member of the subject physician’s peer group.
3.
This example demonstrates an internist with a higher prevalence of diabetes in his practice (73.1%)
than the average prevalence among his peers (23.2%).
4.
The co-morbidity score indicates that the diabetic patients in this panel are “less sick” than the
average diabetic patients seen by internists, in that their ACG expected costs are 32.6% less (100 –
67.4) than the average ACG expects costs of diabetic patients seen by internists.
Technical User Guide
The Johns Hopkins ACG System, Version 9.0
3-26
Performance Assessment
The following table (Table 10) is another example of an EDC report, this time for a
general practitioner. This is intended to illustrate how the EDC report can be used to get
a better understanding of the characteristics of the patients and the special interest of the
physician.
Table 10: Example of an EDC Report for a General Practitioner
Rank/
MEDC
MEDC/EDC
1 ALL
Allergy
ALL03-Allergic Rhinitis
ALL02-Asthma
2 MUS
Musculoskeletal
MUS13-Low back pain
MUS03-Degenerative
Joint Disease
3 EAR
Ear, Nose and Throat
EAR11-Acute Upper
Respiratory Infection
EAR10-Otitis Media
Patient
Count
MEDC/EDC
As % Of
This Practice
MEDC/EDC
% Of
Practice In
Peer Group
(Average)
CoMorbidity
Score
1,395
58.7
8.1
83.1
1,086
522
45.7
22.0
3.5
4.3
86.4
90.7
944
41.8
30.4
95
559
273
23.5
11.5
10.8
3.8
85.6
81.1
920
38.7
30.2
105.6
663
27.9
23.6
110.6
295
12.4
5.1
114.4
This table shows that the prevalence of allergic disorders and asthma is much more
prevalent in this patient panel (59%) than the average prevalence among general
practitioners’ patient panels (8%). This is most likely indicative of a special interest of
this practitioner.
Evaluating Productivity and Distributing Workload
In addition to efficiency assessment, case-mix adjustment is vital to the evaluation of
physician productivity. Physicians may be under pressure to reduce the duration of visits
in order to increase the number of daily visits performed. This can be counter-productive
when the physician’s panel is more complex. Communication with the patient about
primary and secondary prevention, medication adherence and treatment decisions are key
to the successful management of a patient with multiple co-morbid conditions. Time and
discussion with the patient is needed to identify a patient’s psychosocial problems or a
lack of support at home. Additional time with a patient can also improve patient
satisfaction and may even reduce utilization of laboratory tests, consultations and
medications. Case-mix adjustment is key to understanding the differences in physician
productivity.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-27
Table 11: Comparison of Characteristics Affecting Physician
Productivity
Average Patient Age
% Female
Average Case-Mix
% patients with ≥1 hospital dominant condition
% patients with ≥3 chronic conditions
% patients with frailty condition
% patients with >2 major ADGs
% patients with psycho-social condition
Average # EDCs
Average # Rx-MGs
Average visit length
Panel 1
Panel 2
36
39.6%
0.86
1.0%
7.3%
1.3%
1.6%
11.5%
5.3
2.5
13.6 min
36
77.0%
1.23
1.9%
30.7%
2.5%
2.3%
21.7%
6.5
3.3
20.4 min
Quality of Care Assessment
Case-mix adjustment is relevant in population-based assessments of provider clinical
performance where there is a plausible basis for results to vary among patients with
different levels of morbidity burden. Many long-standing performance assessment
programs, such as those promulgated by the National Committee on Quality Assurance
and the Joint Commission on the Accreditation of Healthcare Organizations, have long
focused on process metrics only because there is little basis to believe that the provision
of specific services should differ in populations that differ by case-mix. The steady rise
in pay-for-performance initiatives and balanced scorecards for health care providers has
been accompanied by the steady expansion of performance assessments to include
outcome metrics. There is a strong basis of evidence that health outcomes do vary by
case-mix and that these metrics need some form of case-mix adjustment to ensure
appropriate comparisons between health care providers. When performance assessment
is focused on specific diseases there is a tendency to look for case-mix or severity
adjustment that is tailored to the specific disease. There are numerous risks to such a
disease-oriented performance assessment strategy, not the least of which is that there are
often insufficient numbers of cases for an accurate assessment and that such a disease
orientation will encourage care practices that are not holistic. Some pay for performance
programs have chosen to roll up disease-specific metrics into an overall summary
measure that is less prone to the problem of small numbers and also broadens the quality
focus. In such cases, ACGs used as RUBs or Dx-PM risk scores will work quite
effectively as case-mix adjusters. Indeed, prior work has shown that ACGs do an
excellent job of adjusting for differences in case-mix for commonly used outcome
indicators such as re-hospitalizations and even mortality. Table 12 shows how outcomes
can vary dramatically between groups characterized as low or high risk based upon DxPM risk score.
Technical User Guide
The Johns Hopkins ACG System, Version 9.0
3-28
Performance Assessment
Figure 2: Percentage of Patients with Selected Outcomes by ACG PM
Risk Group
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
Secondary Care
Healthy Patients
No of unique
radiology
No of unique
lab tests
Polypharmacy
(Avg no of
Polypharmacy
(5+ unique
Home visits
No of GP
visits
No of
referrals
Outpatient
unique
Outpatient
visits
Length of
stay
Inpatient
unique
A&E
Inpatient
Inpatient
Admissions
0.0
Primary Care
Population Mean
Very High Risk Patients
F ROM P ILOTING AND E VALUATING C ASE -M IX AND P REDICTIVE M ODELLING
M EASURES W ITHIN T HE B RITISH P RIMARY C ARE S ECTOR , F E B 2007
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Performance Assessment
3-29
Summary
There may be substantive differences in the distribution of risk characteristics in the
relatively small subsets of a population that have a contact with a single physician. In
order to systematically examine the pattern of practice or expenditure related to that
group, adjustment must be made to allow for the patients’ characteristics to be taken into
account. ACGs provide a measure that has greater predictive value than age and gender.
EDCs provide an insight into the special interest or diagnostic mix of patients seen by a
physician. A measure of co-morbidity within EDCS can help discern whether patients
with that diagnosis have more or fewer co-morbidities than patients with the same
diagnosis seen by the physician’s peers. Rather than looking only at charges that the
physician incurred, a global look at the overall utilization of service incurred by the
practitioner’s panel gives a perspective on that physician’s contribution to the whole.
Risk adjustment of profiles should not be considered definitive, but should be used to
focus scarce resources allocated to education, peer review or audit. ACG risk adjustment
in a whole patient approach may avoid false positive analytic findings that wrongly
suggest that certain physicians’ practice styles are too expensive. Figure 1 (above)
provides a graphic representation of this property.
Technical User Guide
The Johns Hopkins ACG System, Version 9.0
3-30
Performance Assessment
This page was left blank intentionally.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Clinical Screening by Care and Disease Managers
4-i
4 Clinical Screening by Care and Disease
Managers
Introduction................................................................................................... 4-1
H
H
High-Risk Case Identification for Case Management............................... 4-1
H
H
Table 1: Amount of Data and Its Impact on Model Performance............. 4-2
Figure 1: Percent Correctly Identified as High Cost; Comparing
One-Month of Rx to 12-Months of Dx+Rx ............................................... 4-3
Figure 2: Percent of Patients Identified by ICD or NDC or Both............. 4-4
Figure 3: Combining Rx and Dx Predictive Modeling Scores for
Targeted Intervention ................................................................................. 4-5
Table 2: Number of Cases and the Johns Hopkins ACG Dx-PM
Predicted Relative Resource Use by Risk Probability Thresholds
for Selected Chronic Conditions ................................................................ 4-6
H
H
H
H
H
H
H
H
H
H
The ACG Predictive Model’s Probability Score ........................................ 4-7
H
H
Table 3: Care Management Listing........................................................... 4-8
Figure 4: Hospital Prediction .................................................................... 4-9
Figure 5: Pharmacy Adherence............................................................... 4-10
Figure 6: Comprehensive Patient Clinical Profile .................................. 4-11
H
H
H
H
H
H
H
H
Risk Stratification ....................................................................................... 4-12
H
H
Table 4: Percentage Distribution of Each Co-Morbidity Level
Within an EDC (Samples)........................................................................ 4-13
Table 5: Estimated Concurrent Resource Use by RUB by EDC
(Samples).................................................................................................. 4-14
H
H
H
H
Disease Management Candidates .............................................................. 4-14
H
H
Figure 7: Cost Predictions by Select Conditions ..................................... 4-15
H
H
Case Mix Control ........................................................................................ 4-16
H
H
Table 6: Measuring Return on Investment.............................................. 4-16
H
H
Technical Considerations ........................................................................... 4-17
H
H
Prospective Risk Scores........................................................................... 4-17
Converting Scores to Dollars ................................................................... 4-17
How to Rescale and Assign Dollar Values .............................................. 4-17
Table 7: Estimating Costs in a Sample of Cases..................................... 4-18
Adjustments for Inflation ......................................................................... 4-18
Local Calibration of ACG Predictive Modeling Scores .......................... 4-19
Applications Guide
H
H
H
H
H
H
H
H
H
H
H
H
The Johns Hopkins ACG System, Version 9.0
4-ii
Clinical Screening by Care and Disease Managers
This page was intentionally left blank.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Clinical Screening by Care and Disease Managers
4-1
Introduction
The ACG System provides a robust set of clinical markers that help Care and Disease
Managers to:
•
Proactively identify patients with expectations of high future costs
•
Stratify a population based on risk for program intensity
•
Select candidates for Disease Management programs
•
Calculate return on investment controlling for case-mix
•
Review a comprehensive risk profile of the member including likelihood of
hospitalization, risk for poorly coordinated care, and possession of chronic
medications.
High-Risk Case Identification for Case Management
The suite of ACG Predictive Models, includes the Dx-PM (based on diagnosis codes), the
Rx-PM (based on drug codes), and the combined DxRx-PM (which uses both diagnostic
and medication information). These represent a real advance if you want to establish or
augment care management programs within your organization. Existing ACG measures
have many applications in this domain as well.
There are a great number of variants within the ACG predictive models. You can select a
model based on data source (diagnosis, pharmacy or both), calibration data (elderly or
non-elderly), and prior cost (total cost, pharmacy cost or no prior cost). In general, the
accuracy of the predictive model will increase as more information is made available.
Therefore, a model that uses diagnosis, pharmacy and prior cost will be more predictive
than a model based only on pharmacy claims without prior cost. There is still good
reason to implement the pharmacy only model. Pharmacy data is fairly complete after 90
days and there is generally minimal lag. As new enrollees are brought on to the plan,
rapid risk assessment can be performed on these members using Rx-PM. The minor
differences in predictive accuracy are compensated for by the gains in time for
intervention. The ACG predictive modeling suite provides choices that allow you to
select the model that best fits your application.
Using just a single month of claim’s data, Table 1 demonstrates the benefit of the ACG
Rx-PM model.
Applications Guide
The Johns Hopkins ACG Case-Mix System, Version 9.0
4-2
Clinical Screening by Care and Disease Managers
Table 1: Amount of Data and Its Impact on Model Performance
Data and Model
C-Statistic
1 Month Rx
0.774
3 Months Rx
0.784
6 Months Rx
0.784
12 Months Rx
0.782
12 Months Rx+Dx+Prior Cost
0.831
There are many ways to adapt the ACG predictive models in the pursuit of improved
patient care. This section provides a summary and overview of some of the recommended
approaches that an organization may wish to consider in the care-management and
quality improvement (QI) domains.
ACG predictive modeling provides information at the individual patient level to help
identify persons who potentially would be well served by special attention from the
organization’s care management infrastructure. This high-risk case identification process
could be used to target a person for interventions such as a referral to a case-manager,
special communication with the patient’s physician, structured disease management
programs, or educational outreach. There are several benefits to this approach to case
selection:
•
The various clinical categories and markers from the system provide a comprehensive
patient profile that can improve the productivity of the screener.
•
A rapid assessment can be performed on the whole population, not just those being
referred through other programs.
•
Predictive modeling helps to identify a unique population of members at risk
−
By identifying members that are complex and co-morbid, but not necessarily
currently high cost, you identify a population that is more open to care
management services and therefore, higher case open rates are seen using ACG
predictive models as a referral tool. This is a productivity improvement for the
care management staff as well.
−
Approximately 25% of the members correctly identified as high risk by an ACG
predictive model were not previously high cost. This percentage seems to hold
regardless of the model – Dx-PM, Rx-PM or DxRx-PM. When using Rx-PM, this
percentage holds true with as little as 1 month of data.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Clinical Screening by Care and Disease Managers
−
4-3
Figure 1 illustrates two pie charts providing a comparison by percentage of high
cost members correctly identified using prior cost, Dx-PM and DxRx-PM models.
The two charts contrast the difference between making predictions using just one
month of pharmacy data versus making predictions using twelve months of
diagnosis+pharmacy data. While the Rx-PM model works well on as little as one
month of data, the accuracy of predictive modeling improves as the quality of the
underlying data (as measured by diagnoses and pharmacy data) improves. Using
Dx-PM and Rx-PM as independent assessments of risk can yield even more
information for a care manager.
Figure 1: Percent Correctly Identified as High Cost; Comparing OneMonth of Rx to 12-Months of Dx+Rx
36%
25%
50%
39%
25%
−
25%
Rx-PM (1 month data)
DxRx-PM (12 months data)
Prior Cost
Prior Cost
Both
Both
The Rx-MGs can supplement the EDCs in describing the clinical conditions of the
patient. Depression and hypertension, in particular, may not be part of the diagnoses,
but will be captured in the prescriptions. If these patients are tracked over time and
there is a pattern of prescriptions without visits, communication with the member and
provider may be helpful.
Applications Guide
The Johns Hopkins ACG Case-Mix System, Version 9.0
4-4
Clinical Screening by Care and Disease Managers
\ Pharmacy identifies additional members with specific conditions as compared to
diagnosis alone as demonstrated in Figure 2.
Figure 2: Percent of Patients Identified by ICD or NDC or Both
Depression
Hypertension
CHF
14%
28%
27%
46%
54%
59%
32%
14%
26%
Rx
ICD
Both
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Clinical Screening by Care and Disease Managers
4-5
Figure 3 shows the value of evaluating members with discordant scores based on
diagnosis and pharmacy. Both the Dx-PM and Rx-PM scores were grouped into
percentiles to indicate high, medium and low risk. Those members with high risk as
defined by Dx-PM were more likely to be hospitalized, especially when they were low
risk as defined by Rx-PM. The combination of scores may provide insight into the
under-treatment or non-compliance of particular populations.
Figure 3: Combining Rx and Dx Predictive Modeling Scores for
Targeted Intervention
% hospitalized
with MI
10
8
6
4
2
0
90-99
50-89
<50
<50
50-89
Rx-PM
(NDC) Risk
Percentile
90-99
Dx-PM (ICD) Risk Percentile
The ACG predictive models include reports providing disease-specific (based on selected
individual and aggregated EDCs and/or pharmacy-based morbidity categories (Rx-MGs))
distributions of risk probability scores and average expected resource use for different
risk cohorts. An example of such a report for The Johns Hopkins ACG Dx-PM model,
shown as Table 2, will be useful in helping to frame a strategy for targeting various risk
cohorts within disease management programs.
Applications Guide
The Johns Hopkins ACG Case-Mix System, Version 9.0
4-6
Clinical Screening by Care and Disease Managers
Table 2: Number of Cases and the Johns Hopkins ACG Dx-PM
Predicted Relative Resource Use by Risk Probability Thresholds for
Selected Chronic Conditions
Predicted Relative
Resource Use
Number of Cases
Disease Category
(EDC)
Arthritis
Asthma
Diabetes
Hypertension
Ischemic Heart Disease
Congestive Heart Failure
Disorders of Lipid
Metabolism
Low Back Pain
Depression
Chronic Renal Failure
COPD
Probability Score
Category
Probability Score Category
Total
≥0.4
≥0.6
≥0.8
<0.4
≥0.4
≥0.6
≥0.8
17,679
27,863
16,991
50,122
9,330
1,634
940
764
1,307
2,064
971
460
463
386
716
1,011
514
292
172
136
345
457
242
184
2.18
1.43
2.67
2.06
3.27
5.17
6.82
6.75
7.59
7.25
7.40
8.81
9.31
9.29
10.62
10.27
10.35
12.26
15.71
14.85
17.36
17.57
17.33
19.61
31,240
61,980
10,190
742
6,204
1,170
1,493
599
308
545
529
723
298
253
301
186
279
113
183
147
1.97
1.76
2.09
13.11
2.58
7.13
6.53
6.63
16.48
7.71
9.49
8.77
9.03
19.40
10.24
15.46
14.27
14.30
25.21
16.68
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Clinical Screening by Care and Disease Managers
4-7
The ACG Predictive Model’s Probability Score
The ACG predictive model probability score (used in Table 2) identifies persons in your
organization who would be likely to benefit from special attention. To capitalize on this
method, you will want to develop periodic reports of members with high PM scores who
also meet other organizational criteria such as:
•
Enrolling with certain providers
•
Falling into certain eligibility categories
•
Residing in certain geographic areas
•
Meeting previous patterns of utilization
After these other stratifiers are taken into consideration as appropriate, a case finding
report should list all in-scope individuals arrayed from highest to lowest, based upon the
overall PM high-risk probability score within your organization. Table 3 provides an
example of a case finding report.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
4-8
Clinical Screening by Care and Disease Managers
Sex
Total Cost
Rescaled
Total Cost
Resource
Index
Probability
High Total
Cost
Hospital
Dominant
Count
Chronic
Condition
Count
Frailty Flag
Depression
Diabetes
Disorders of
Lipid
Metabolism
Hypertension
Persistent
Asthma
Rheumatoid
Arthritis
A
71
M
$ 7,127
29.15
0.95
3
6
N
NP
NP
NP
NP
NP
NP
NP
Rx
B
51
M
$ 7,304
23.39
0.95
2
6
Y
NP
NP
NP
NP
NP
NP
NP
NP
C
47
M
$ 8,082
21.12
0.88
2
3
N
NP
NP
NP
NP
Rx
NP
Rx
Rx
D
49
F
$ 7,861
18.33
0.88
0
7
N
BTH
Rx
NP
NP
NP
NP
NP
BTH
E
50
M
$ 5,375
18.44
0.88
1
7
N
NP
NP
NP
BTH
NP
ICD
BTH
BTH
F
63
M
$ 8,306
20.58
0.88
1
7
N
NP
NP
NP
ICD
NP
BTH
BTH
BTH
G
42
F
$ 4,757
17.07
0.88
1
6
N
NP
NP
NP
ICD
Rx
NP
NP
NP
H
76
F
$ 6,276
20.16
0.88
1
3
N
NP
NP
NP
NP
Rx
BTH
NP
Rx
I
65
M
$ 8,004
19.16
0.88
2
8
N
NP
Rx
BTH
ICD
NP
NP
NP
BTH
J
85
M
$ 7,466
17.36
0.88
3
10
Y
ICD
NP
ICD
NP
NP
BTH
ICD
BTH
Patient Id
Age
Congestive
Heart Failure
Chronic
Renal Failure
Table 3: Care Management Listing
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Clinical Screening by Care and Disease Managers
4-9
An alternative strategy for identifying members may be to focus on members at high risk
for hospitalization. The ACG System includes a series of probability scores to identify
members with increased risk of a variety of hospitalization events. The goal in providing
multiple hospitalization scores was to improve the sensitivity of identifying members
with particular characteristics, such as intensive care requirements or extended stays in
the hospital. Figure 4 shows the “Hospital Predictions for Select Major Conditions”
analysis. This analysis quantifies the conditions associated with highest probability of
future hospitalization.
Figure 4: Hospital Prediction
Care managers may also find it useful to understand the use of chronic medications to
manage specific disease states. Figure 5 identifies conditions for which medication
possession (MPR) is low. For a full description of the pharmacy adherence methodology,
see the Technical Reference Guide, Chapter 11.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
4-10
Clinical Screening by Care and Disease Managers
Figure 5: Pharmacy Adherence
Additional context is provided for clinical screeners by combining all of the risk factor
and descriptive information from the system. The Comprehensive Patient Clinical Profile
(see Figure 6) combines all of the ACG system markers into a single report view that can
support the clinical screening process.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Clinical Screening by Care and Disease Managers
4-11
Figure 6: Comprehensive Patient Clinical Profile
Applications Guide
The Johns Hopkins ACG System, Version 9.0
4-12
Clinical Screening by Care and Disease Managers
In addition to running the report automatically generated by the software, you are
encouraged to develop your own individual risk summary reports on each potential case
over a certain threshold (for instance the top 1% of individuals). This target group can be
separated further by case managers on the basis of various sources of information
available from the ACG Software and elsewhere. These additional data might include
primary care provider information, service history, history of prior inclusion in care
management programs, and results from any ongoing surveys (such as health-risk
appraisals).
Risk Stratification
In care management program development, it is useful to vary the resource intensity of
the program offering based upon the needs of the patients. The ACG System can help
identify patients with complex disease profiles who will benefit from more intensive
programs and personal contact with a nurse case manager.
Concurrent ACG/RUB morbidity information can be combined with EDCs to control for
morbidity differences across a given disease-specific group of interest (e.g., diabetics
enrolled in a disease management program). EDCs are useful in portraying the disease
characteristics of a population of interest. Within disease management programs, if
significant differences in expected resource consumption exist across the morbidity
subclasses, this analytic approach is useful for better targeting interventions towards
subgroups at higher risk.
The ACG Software produces tables in which each row represents persons falling into
EDC (or MEDC) disease-specific categories; the columns array these individuals into
RUB co-morbidity categories according to their ACG assignment. Table 4 presents the
percentage distribution for a series of selected EDCs across the five RUB categories.
Table 5 presents the expected relative resource use within each RUB and illustrates comorbidity’s profound influence on resource use within individual disease groups. The
ACG-based RUBs do a very good job of explaining variations in resource use within
specific diseases. For additional detail on interpreting or building similar tables please
refer to the chapter entitled “Expanded Diagnosis Clusters (EDCs)” in the Technical
Reference Guide.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Clinical Screening by Care and Disease Managers
4-13
Table 4: Percentage Distribution of Each Co-Morbidity Level Within
an EDC (Samples)
EDC
ADM02
ADM03
ALL01
ALL03
ALL04
ALL05
ALL06
CAR04
CAR05
CAR06
CAR07
CAR08
CAR09
CAR10
CAR11
CAR12
CAR13
Description
Surgical aftercare
Transplant status
Allergic reactions
Allergic rhinitis
Asthma, w/o status
asthmaticus
Asthma, with status
asthmaticus
Disorders of the
immune system
Congenital heart
disease
Congestive heart
failure
Cardiac valve
disorders
Cardiomyopathy
Heart murmur
Cardiac arrhythmia
Generalized
atherosclerosis
Disorders of lipoid
metabolism
Acute myocardial
infarction
Cardiac arrest, shock
RUB-1
Very
Low
RUB-2
Low
RUB-3
Average
RUB-4
High
RUB-5
Very
High
4.7
3.8
0.0
0.0
19.3
7.7
36.2
34.5
46.6
32.9
53.6
56.0
18.9
26.6
8.5
8.2
10.4
29.1
1.6
1.3
0.0
23.6
63.2
10.7
2.5
0.0
20.9
58.0
15.6
5.4
0.0
6.5
47.6
25.5
20.4
0.0
17.4
45.9
23.9
12.4
0.0
0.4
36.6
31.1
31.9
0.0
7.6
59.1
22.2
11.1
0.0
12.3
0.0
2.2
25.8
3.7
43.8
44.5
58.4
30.1
11.9
24.5
23.9
5.4
13.3
0.0
7.0
43.7
25.4
23.9
0.0
17.3
68.0
10.4
4.2
0.0
0.2
21.3
39.3
39.2
0.0
5.4
19.2
31.2
44.2
You can develop your own reports, and the EDCs that define the rows in Tables 4 and 5
could be replaced by episodes of illness categories that an organization may obtain from
other sources. ACG-based RUBs are equally effective in explaining variations in
resource use within episodes of care.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
4-14
Clinical Screening by Care and Disease Managers
Table 5: Estimated Concurrent Resource Use by RUB by EDC
(Samples)
EDC
ADM02
ADM03
ALL01
ALL03
ALL04
ALL05
ALL06
CAR04
CAR05
CAR06
CAR07
CAR08
CAR09
CAR10
CAR11
CAR12
CAR13
Description
Surgical aftercare
Transplant status
Allergic reactions
Allergic rhinitis
Asthma, w/o status
asthmaticus
Asthma, with status
asthmaticus
Disorders of the
immune system
Congenital heart
disease
Congestive heart
failure
Cardiac valve
disorders
Cardiomyopathy
Heart murmur
Cardiac arrhythmia
Generalized
atherosclerosis
Disorders of lipoid
metabolism
Acute myocardial
infarction
Cardiac arrest, shock
RUB-1
Very
Low
RUB-2
Low
RUB-3
Average
RUB-4
High
RUB-5
Very
High
0.20
0.20
0.00
0.00
0.63
0.65
0.54
0.54
2.31
2.39
2.07
2.13
7.94
8.23
7.49
7.43
27.30
29.89
25.41
25.40
0.00
0.62
2.03
7.43
26.10
0.00
0.62
2.13
7.50
28.23
0.00
0.74
2.39
7.71
29.63
0.00
0.73
2.20
7.11
25.56
0.00
0.81
2.62
8.30
28.83
0.00
0.56
2.42
7.86
27.10
0.00
0.21
0.17
0.73
0.64
0.61
2.37
2.22
2.37
8.23
7.20
8.07
28.69
23.05
25.82
0.00
0.46
2.47
8.23
27.06
0.00
0.49
2.29
8.17
25.14
0.00
0.82
1.85
7.87
26.28
0.00
0.62
2.12
7.74
27.84
Disease Management Candidates
The EDCs allow care managers to associate members with specific conditions to
facilitate the identification of candidates for disease management programs. When
processing your ACG data file, the use of the “stringent” option for diagnostic certainty is
recommended to limit the identification of members based on a single instance of a
potentially provisionally recorded diagnosis code. For a select list of common disease
management programs, the ACG System also provides condition markers (See the
Technical Reference Guide, Chapter 6 for definitions). These markers are used in the
“Cost Predictions by Select Conditions” analysis shown in Figure 7. This analysis
allows care managers to size programs and estimate predicted costs for each condition.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Clinical Screening by Care and Disease Managers
4-15
Figure 7: Cost Predictions by Select Conditions
Individual members can be identified with each of these conditions and the evidence
(pharmacy, diagnosis or both). For a subset of conditions that warrant chronic
medications, care managers are also able to determine if members are treated or untreated
with chronic medications.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
4-16
Clinical Screening by Care and Disease Managers
Case Mix Control
When calculating return on investment or other outcomes attributed to care management
programs, it is essential to control for the case-mix of study populations. Candidates that
opt-in to programs may be selectively more or less risky than all eligible candidates.
Further, care management candidates often have prior utilization that would reduce
without intervention given a tendency to the mean. Table 6 provides an example which
controls for case-mix differences between patients receiving interventions and those not
receiving interventions. In all cases, there was a reduction of costs year over year.
Without risk adjustment, one might incorrectly conclude that the 43% reduction in the no
intervention group surpassed the 34% reduction in the intervention group. However, cost
measurements are broken into high and low risk populations, you see that high risk
members were preferentially enrolled in the program and had reductions that far
exceeded the high risk no intervention group.
Table 6: Measuring Return on Investment
Risk Level
Intervention
Level
Number Year 1 Costs Year 2 Costs
of
(pre
(post
Patients intervention) intervention)
% Change
Overall
No Intervention
7661
738
420
-43%
Intervention
172
2852
1881
-34%
No Intervention
7573
682
368
-46%
Intervention
105
654
347
-47%
No Intervention
88
$5571
$4932
-11%
Intervention
67
$6296
$4285
-32%
Low
High
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Clinical Screening by Care and Disease Managers
4-17
Technical Considerations
Prospective Risk Scores
With the advent of the ACG PM, it is possible to generate prospective risk scores within
the ACG Software. This prospective risk score or “weight” is called the Predictive
Resource Index, or PRI. Unlike the concurrent ACG-weights which are linked to specific
ACGs, the PRI is individualized and thus, conceivably, every member could have a
distinct PRI score. Two PRI scores are produced--one for total cost and one for pharmacy
cost. The PRI is interpreted in the same manner as a concurrent ACG weight, i.e., as a
relative value. The software produces both an unadjusted and adjusted form of the PRI.
The adjustment process is identical to that used to produce the adjusted concurrent
weights.
Converting Scores to Dollars
As noted above, both the ACG-weights and the ACG PM’s PRI are expressed as relative
values, where the mean is centered at 1.0 (assuming the scores have been appropriately
rescaled). The interpretation then is that individuals with scores higher than 1.0 are more
expensive than average, whereas those with scores less than 1.0 are less expensive than
average. Such relative indices can easily be converted to dollar amounts by multiplying
by the underlying mean of the population to which the risk adjustment values will be
applied. These dollars can be used as the expected cost values for profiling and other risk
adjustment applications.
Before converting scores to dollar amounts, it is important to rescale the data (one option
is to just use the “adjusted” weights described above) to account for differences between
the reference population (in this case, the US Non-Elderly Risk Assessment Variables
from Johns Hopkins nationally representative database) and the population to which the
weights are applied (e.g., your population of interest). Rescaling is necessary to assure
that the underlying mean of the weights is 1.0. A similar process is undertaken when you
use your own reference population and it has somewhat different characteristics (e.g., it is
from a previous time period, or benefit coverage is somewhat different). Unless rescaling
is done, resource use (or payments) may be over or under-predicted. Table 7 and the
accompanying discussion provide a simplified example for a population with only twelve
members.
How to Rescale and Assign Dollar Values
The rescaling process consists of the following steps:
Step 1: Compute population mean weight. Compute a separate grand mean for each of
the weights (either concurrent ACG weights or the ACG PM PRI) generated for your
population (the observations represent individuals). The mean for this example is shown
in Table 7 at the bottom of Column B.
Step 2: Apply weighting factor. Divide each individual weight by the rescaling factor
(i.e., the mean) that you computed in Step 1. The result is the rescaled relative weight
(Column C).
Applications Guide
The Johns Hopkins ACG System, Version 9.0
4-18
Clinical Screening by Care and Disease Managers
Step 3: Compute population mean cost. For the same population on which the weights
were based, compute the mean cost for the current data year. For this example, the mean
cost was $1,265.11.
Step 4: Compute cost. Multiply the rescaled relative weights generated for each
member of the population (Column C) by the average population cost generated from
Step 3 to calculate an estimated individual cost (Column D).
Table 7: Estimating Costs in a Sample of Cases
A
Member
B
Relative Weight
1
2
3
4
5
6
7
8
9
10
11
12
Mean
0.185
0.291
0.387
0.457
0.541
0.609
0.696
0.842
1.025
1.293
1.892
4.783
1.083
C
Rescaled Weight
0.171
0.268
0.357
0.422
0.499
0.562
0.642
0.777
0.946
1.194
1.746
4.415
1.000
D
Estimated Cost
$216.36
$339.61
$451.64
$533.33
$631.33
$711.58
$812.58
$982.84
$1,196.68
$1,510.19
$2,209.38
$5,585.78
$1,265.11
The rescaling factor functions as a summary case-mix index for understanding how the
rating population (e.g., your local population) compares to the development data (the US
Non-Elderly Risk Assessment Variables from JHU’s nationally representative database).
The interpretation of this factor is analogous to how one interprets both relative weights
and profiling indicators. If the rescaling factor is greater than 1.0 (as it was in the
example), then your population is sicker; if the factor is less than 1.0, then your
population is healthier than the reference population.
Adjustments for Inflation
If you are going to use the scores for predicting future expenditures it may be appropriate
to inflation-adjust these values. Based on Bureau of Labor Statistics results for the
calendar year 2004, medical care costs rose by approximately 5% over the previous year
(see http://data.bls.gov). In the preceding example, if you were going to apply this
inflation adjustment, you would multiply the mean cost computed in Step 3 by 1.05 to
reflect inflation. For this example, the inflation-adjusted mean cost for the next year
would have been $1,328.37 instead of $1,265.11. Depending on the local situation, it
may also be appropriate to modify future cost expectations for other actuarial factors such
as changes in benefit structure of cost-sharing provisions.
H
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Clinical Screening by Care and Disease Managers
4-19
Note: The above discussion was meant to offer general instructional guidance on the
rescaling of relative weights and inflation adjustment. Given that no two analytic or
actuarial applications are exactly alike, and given the potentially major impact that such a
process may have on the management or financial applications within your organization,
it is essential that you seek and follow advice from experienced statistical or actuarial
specialists before finalizing the general processes described above.
Local Calibration of ACG Predictive Modeling Scores
The prospective scores provided in the Dx-PM, Rx-PM and DxRx-PM are based upon
multivariate linear regression models. To develop a locally-based PRI score would
involve fitting a regression to local data using the variables included within the ACG
predictive models. A listing of the predictor variables (the “independent” variables) is
provided as Appendix B to the Technical Reference Guide. Using these variables and
local cost data, an experienced analyst could develop a new set of PRI scores that are
customized for the local enrollee population. Custom models should be based on
populations of no fewer than 100,000 individuals.
\ Tip: In the Export ACG Data Menu there is a Model Markers file that contains two
columns, a member ID and a string of Boolean (0/1) flags representing the right-hand
side of the regression equation. Local calibration can be performed by merging this
file with cost information. We strongly recommend you talk to your ACG support
analyst for technical support in implementing this application, at least the first time.
The Model Marker file contains all necessary flags for the DxRx-PM and
hospitalization prediction models.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
4-20
Clinical Screening by Care and Disease Managers
This page was left blank intentionally.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Managing Financial Risk for Pharmacy Benefits
5-i
5 Managing Financial Risk for Pharmacy Benefits
Managing Financial Risk for Pharmacy Benefits ...................................... 5-1
H
H
Examining Differences in Prescribing Risk............................................... 5-1
Table 1: Standardized Morbidity Ratio by Rx-MG ................................... 5-2
Predicting Pharmacy Use ........................................................................... 5-5
Table 2: Cost Predictions for Selected Rx-MGs ....................................... 5-6
Medication Therapy Management Program (MTMP) Candidate
Selection ..................................................................................................... 5-9
Figure 1: Care Management List............................................................. 5-10
Figure 2: Comprehensive Patient Clinical Profile .................................. 5-11
H
H
H
H
H
H
H
H
H
H
Applications Guide
H
H
H
H
The Johns Hopkins ACG System, Version 9.0
5-ii
Managing Financial Risk for Pharmacy Benefits
This page was left blank intentionally.
The Johns Hopkins ACG System, Version 9.0
Managing Financial Risk for Pharmacy Benefits
5-1
Managing Financial Risk for Pharmacy Benefits
Prescription Drug Plans (PDPs) have unique challenges. The organizations are at
financial risk yet have access to very limited data to manage that risk. The ACG Rx-PM
and the pharmacy based morbidity groups Rx-MGs provide a unique opportunity to
leverage this information for comparing population health, predicting resource needs and
providing useful and relevant information to care managers.
Examining Differences in Prescribing Risk
While medical claims provide explicit diagnostic information compared to prescriptions,
diagnostic coding is subject to the challenges of medical chart documentation and
abstraction. Prescription information is more readily available, with claims processed at
the point of service, and prescribing events often occur with more frequency than office
visit events for capturing risk information. This makes prescription information ideal for
assessing population health and aggregate population risk.
Rx-Morbidity Groups (Rx-MGs) provide a mechanism to describe population health
solely based on prescription data. Rx-MGs are a limited number of morbidity markers
that may be applied similarly to Expanded Diagnosis Clusters (EDCs). The calculation
of the Standardized Morbidity Ratio in Table 1 is described in the Application Guide,
Chapter 2. For a given population, e.g., a clinic, region, network or benefit plan, the
standardized morbidity ratio shows differences in disease prevalence (imputed from
prescriptions) when compared to the health plan average.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
5-2
Managing Financial Risk for Pharmacy Benefits
Table 1: Standardized Morbidity Ratio by Rx-MG
Rx-MG Name
Allergy/Immunology / Acute Minor
Allergy/Immunology / Chronic
Inflammatory
Allergy/Immunology / Immune Disorders
Cardiovascular / Chronic Medical
Cardiovascular / Congestive Heart
Failure
Cardiovascular / High Blood Pressure
Cardiovascular / Disorders of Lipid
Metabolism
Cardiovascular / Vascular Disorders
Ears, Nose, Throat / Acute Minor
Endocrine / Bone Disorders
Endocrine / Chronic Medical
Endocrine / Diabetes With Insulin
Endocrine / Diabetes Without Insulin
Endocrine / Thyroid Disorders
Endocrine / Weight Control
Eye / Acute Minor: Curative
Eye / Acute Minor: Palliative
Eye / Glaucoma
Female Reproductive / Hormone
Regulation
Female Reproductive / Pregnancy and
Delivery
Gastrointestinal/Hepatic / Acute Minor
Gastrointestinal/Hepatic / Chronic Liver
Disease
The Johns Hopkins ACG System, Version 9.0
Patient Observed/
Count
1000
Age/Sex
Expected/
1000
SMR
95%
Confidence
Low
95%
Confidence
High
Significance
173
130.27
97.29
1.34
1.14
1.54
+
139
2
46
104.67
1.51
34.64
83.32
0.44
31.25
1.26
3.45
1.11
1.05
0.00
0.79
1.47
8.24
1.43
+
22
270
16.57
203.31
12.68
186.58
1.31
1.09
0.76
0.96
1.85
1.22
191
39
36
27
55
26
96
80
18
61
26
10
143.83
29.37
27.11
20.33
41.42
19.58
72.29
60.24
13.55
45.93
19.58
7.53
128.47
22.81
24.38
20.21
37.65
17.43
48.92
46.08
10.41
41.93
16.21
8.74
1.12
1.29
1.11
1.01
1.10
1.12
1.48
1.31
1.30
1.10
1.21
0.86
0.96
0.88
0.75
0.63
0.81
0.69
1.18
1.02
0.70
0.82
0.74
0.33
1.28
1.69
1.48
1.39
1.39
1.55
1.77
1.59
1.90
1.37
1.67
1.40
87
65.51
68.53
0.96
0.76
1.16
23
80
17.32
60.24
15.35
46.65
1.13
1.29
0.67
1.01
1.59
1.57
1
0.75
1.00
0.75
0.00
2.23
+
+
+
Applications Guide
Managing Financial Risk for Pharmacy Benefits
Rx-MG Name
Gastrointestinal/Hepatic / Inflammatory
Bowel Disease
Gastrointestinal/Hepatic / Peptic Disease
General Signs and Symptoms / Nausea
and Vomiting
General Signs and Symptoms / Pain
General Signs and Symptoms / Pain and
Inflammation
General Signs and Symptoms / Severe
Pain
Genito-Urinary / Acute Minor
Infections / Acute Major
Infections / Acute Minor
Malignancies
Musculoskeletal / Gout
Musculoskeletal / Inflammatory
Conditions
Neurologic / Alzheimers Disease
Neurologic / Chronic Medical
Neurologic / Migraine Headache
Neurologic / Parkinsons Disease
Neurologic / Seizure Disorder
Psychosocial / Attention Deficit
Hyperactivity Disorder
Psychosocial / Addiction
Psychosocial / Anxiety
Psychosocial / Depression
Psychosocial / Acute Minor
Psychosocial / Chronic Unstable
Respiratory / Acute Minor
Applications Guide
5-3
Patient Observed/
Count
1000
Age/Sex
Expected/
1000
SMR
95%
Confidence
Low
95%
Confidence
High
Significance
5
144
3.77
108.43
4.36
85.34
0.86
1.27
0.11
1.06
1.62
1.48
+
51
344
38.40
259.04
39.78
227.22
0.97
1.14
0.70
1.02
1.23
1.26
+
227
170.93
147.19
1.16
1.01
1.31
+
26
85
4
637
9
12
19.58
64.01
3.01
479.67
6.78
9.04
14.96
49.64
2.82
432.87
6.05
7.34
1.31
1.29
1.07
1.11
1.12
1.23
0.81
1.02
0.02
1.02
0.39
0.53
1.81
1.56
2.12
1.19
1.85
1.93
10
1
15
37
12
71
7.53
0.75
11.30
27.86
9.04
53.46
7.80
1.85
5.69
21.56
4.73
43.16
0.97
0.41
1.98
1.29
1.91
1.24
0.37
0.00
0.98
0.88
0.83
0.95
1.56
1.20
2.99
1.71
2.99
1.53
36
2
71
229
53
19
211
27.11
1.51
53.46
172.44
39.91
14.31
158.89
23.09
2.12
49.88
146.69
33.30
11.70
124.83
1.17
0.71
1.07
1.18
1.20
1.22
1.27
0.79
0.00
0.82
1.02
0.88
0.67
1.10
1.56
1.69
1.32
1.33
1.52
1.77
1.44
+
+
+
+
The Johns Hopkins ACG System, Version 9.0
5-4
Managing Financial Risk for Pharmacy Benefits
Rx-MG Name
Respiratory / Chronic Medical
Respiratory / Airway Hyperactivity
Skin / Acne
Skin / Acute and Recurrent
Skin / Chronic Medical
Other and Non-Specific Medications
Patient Observed/
Count
1000
14
144
27
172
2
50
10.54
108.43
20.33
129.52
1.51
37.65
Age/Sex
Expected/
1000
12.20
94.08
21.52
116.50
4.87
30.32
SMR
0.86
1.15
0.94
1.11
0.31
1.24
95%
Confidence
Low
95%
Confidence
High
0.41
0.96
0.59
0.95
0.00
0.90
1.32
1.34
1.30
1.28
0.74
1.59
Significance
-
The subpopulation shown in Table 1 has a sicker population than average demonstrated through higher disease prevalence across
most conditions. Significantly higher rates chronic illness, such as Diabetes identified by non-insulin diabetic drugs and Depression,
are likely to have impacts on future pharmacy costs.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Managing Financial Risk for Pharmacy Benefits
5-5
Predicting Pharmacy Use
To date, commercial risk adjustment systems have focused on predicting total patient
costs. This creates challenges for the actuarial bid process that specifically carves out a
pharmacy benefit.
ACG-PM includes predictions of pharmacy cost to support actuarial analysis. There are
substantial differences in predicted pharmacy cost for members with a probability of high
pharmacy use exceeding 40%. In Table 2, members with probability scores <0.4 are
predicted to spend 0.63 of average pharmacy cost in the next period. Members with
probability scores ≥0.4 are predicted to spend 8.82 times average, a 14-fold difference.
This difference is more distinct within specific conditions. For example, members with
growth problems are predicted to spend more than 27 times the average.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
5-6
Managing Financial Risk for Pharmacy Benefits
Table 2: Cost Predictions for Selected Rx-MGs
Rx-Morbidity Groups
ALL CASES
Allergy/Immunology / Acute Minor
Allergy/Immunology / Chronic Inflammatory
Allergy/Immunology / Immune Disorders
Allergy/Immunology / Transplant
Cardiovascular / Chronic Medical
Cardiovascular / Congestive Heart Failure
Cardiovascular / Disorders of Lipid Metabolism
Cardiovascular / High Blood Pressure
Cardiovascular / Vascular Disorders
Ears, Nose, Throat / Acute Minor
Endocrine / Bone Disorders
Endocrine / Chronic Medical
Endocrine / Diabetes With Insulin
Endocrine / Diabetes Without Insulin
Endocrine / Growth Problems
Endocrine / Thyroid Disorders
Endocrine / Weight Control
Eye / Acute Minor: Curative
Eye / Acute Minor: Palliative
Eye / Glaucoma
Female Reproductive / Hormone Regulation
Female Reproductive / Pregnancy and Delivery
Gastrointestinal/Hepatic / Acute Minor
Gastrointestinal/Hepatic / Chronic Liver
Disease
The Johns Hopkins ACG System, Version 9.0
Total
Cases
Avg. Pred
Resource Use
Avg. Pred.
Resource Use
Prob <0.4
Avg. Pred.
Resource Use
Prob >=0.4
Cases Prob
<0.4
Cases Prob
>=0.4
11101
1096
934
4
18
343
142
1438
2104
248
260
231
456
199
540
3
539
118
457
187
94
817
184
526
10603
955
806
1
3
233
86
1141
1778
156
236
178
392
114
388
0
455
103
425
156
74
795
182
439
498
141
128
3
15
110
56
297
326
92
24
53
64
85
152
3
84
15
32
31
20
22
2
87
1.00
2.33
2.42
14.30
11.83
4.39
4.82
3.30
2.76
4.71
1.54
3.67
2.76
5.23
3.91
27.02
2.64
2.14
1.49
2.83
3.35
0.94
0.72
2.69
0.63
1.40
1.26
4.46
2.93
2.16
1.94
1.98
1.68
2.15
0.81
2.04
1.62
2.64
2.17
0.00
1.40
1.29
0.89
1.36
1.73
0.74
0.54
1.41
8.82
8.66
9.75
17.58
13.61
9.11
9.24
8.38
8.63
9.06
8.64
9.16
9.74
8.69
8.34
27.02
9.33
7.99
9.50
10.19
9.37
8.36
17.05
9.13
13
6
7
5.87
1.38
9.73
Applications Guide
Managing Financial Risk for Pharmacy Benefits
Rx-Morbidity Groups
Gastrointestinal/Hepatic / Chronic Stable
Gastrointestinal/Hepatic / Inflammatory Bowel
Disease
Gastrointestinal/Hepatic / Pancreatic Disorder
Gastrointestinal/Hepatic / Peptic Disease
General Signs and Symptoms / Nausea and
Vomiting
General Signs and Symptoms / Pain
General Signs and Symptoms / Pain and
Inflammation
General Signs and Symptoms / Severe Pain
Genito-Urinary / Acute Minor
Genito-Urinary / Chronic Renal Failure
Hematologic / Coagulation Disorders
Infections / Acute Major
Infections / Acute Minor
Infections / HIV/AIDS
Infections / Severe Acute Major
Infections / Tuberculosis
Malignancies
Musculoskeletal / Gout
Musculoskeletal / Inflammatory Conditions
Neurologic / Alzheimers Disease
Neurologic / Chronic Medical
Neurologic / Migraine Headache
Neurologic / Parkinsons Disease
Neurologic / Seizure Disorder
Other and Non-Specific Medications
Psychosocial / Acute Minor
Applications Guide
5-7
Total
Cases
Avg. Pred
Resource Use
Avg. Pred.
Resource Use
Prob <0.4
Avg. Pred.
Resource Use
Prob >=0.4
Cases Prob
<0.4
Cases Prob
>=0.4
5
3
2
6.34
2.97
11.39
51
5
966
34
2
743
17
3
223
4.48
6.35
3.44
2.17
3.70
1.84
9.10
8.13
8.79
454
2580
366
2266
88
314
2.91
2.10
1.24
1.13
9.83
9.11
1688
178
542
5
4
31
4807
8
2
5
71
82
91
19
66
253
56
491
331
374
1480
99
426
0
0
23
4434
0
1
2
50
61
53
8
8
212
37
342
233
278
208
79
116
5
4
8
373
8
1
3
21
21
38
11
58
41
19
149
98
96
2.15
5.94
3.34
14.09
24.80
3.60
1.53
15.40
10.29
7.55
4.63
3.64
6.05
6.27
10.90
2.83
5.39
4.22
4.05
3.67
1.23
1.80
1.77
0.00
0.00
1.23
0.90
0.00
0.87
2.72
2.42
1.71
2.94
2.70
3.58
1.60
2.23
1.95
1.80
1.67
8.73
11.12
9.12
14.09
24.80
10.42
8.96
15.40
19.71
10.77
9.89
9.22
10.39
8.86
11.91
9.18
11.53
9.41
9.40
9.47
The Johns Hopkins ACG System, Version 9.0
5-8
Managing Financial Risk for Pharmacy Benefits
Rx-Morbidity Groups
Psychosocial / Addiction
Psychosocial / Anxiety
Psychosocial / Attention Deficit Hyperactivity
Disorder
Psychosocial / Chronic Unstable
Psychosocial / Depression
Respiratory / Acute Minor
Respiratory / Airway Hyperactivity
Respiratory / Chronic Medical
Respiratory / Cystic Fibrosis
Skin / Acne
Skin / Acute and Recurrent
Skin / Chronic Medical
Toxic Effects/Adverse Effects / Acute Major
Total
Cases
Avg. Pred
Resource Use
Avg. Pred.
Resource Use
Prob <0.4
Avg. Pred.
Resource Use
Prob >=0.4
Cases Prob
<0.4
Cases Prob
>=0.4
24
577
12
466
12
111
5.88
2.94
2.39
1.58
9.36
8.63
240
129
1703
1387
1023
128
8
230
1305
51
1
208
84
1435
1248
900
91
0
224
1181
43
0
32
45
268
139
123
37
8
6
124
8
1
2.12
4.44
2.65
1.80
2.10
4.13
12.79
1.02
1.90
2.84
4.85
1.17
2.01
1.48
1.07
1.23
2.56
0.00
0.83
1.09
1.30
0.00
8.27
8.97
8.94
8.42
8.43
8.01
12.79
8.36
9.58
11.13
4.85
The Pharmacy Predicted Resource Index (PRI) scores that come out of the system can be converted to the predicted pharmacy dollar
amounts by multiplying the PRI by the average plan pharmacy cost.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Managing Financial Risk for Pharmacy Benefits
5-9
Medication Therapy Management Program (MTMP) Candidate
Selection
Medicare PDPs have unique challenges in that one of the regulatory requirements of
PDPs is that they implement Medication Therapy Management Programs (MTMPs).
MTMPs are designed to improve medication adherence, patient safety and quality. The
programs typically focus on promoting beneficiary education and counseling, increasing
enrollee adherence to prescription medication regimens and of detecting adverse drug
events and patterns of over-use and under- use of prescription drugs. These outreach
programs should reach individuals with multiple chronic diseases, such as, but not limited
to, diabetes, asthma, hypertension, disorders of lipid metabolism, and congestive heart
failure who are taking multiple covered Part D Drugs and who are identified as likely to
incur annual costs for covered Part D drugs that exceed the level specified by the
Secretary of Health and Human Services. Since PDPs have access only to prescription
history under their program, meeting this criteria can be a challenge. Rx-PM and the RxMorbidity groups provide an excellent means of finding the population of individuals
defined in the regulations. The Rx-MGs identify members being treated for particular
conditions while the Rx-PM predicted resource index, calibrated for an elderly
population, can be used to calculate an individual cost forecast. Using these tools for the
identification of candidates for MTMPs allows a PDP to screen the whole population
with an objective and reproducible method.
The Care Management List (Figure 1) allows the user to create and store filters based
upon any field provided on the patient file or any field generated by the ACG System.
The example in Figure 1 filtered patients with 3 or more chronic conditions and a Total
Cost PRI greater than 5 to approximate the MTMP criteria described above.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
5-10
Managing Financial Risk for Pharmacy Benefits
Figure 1: Care Management List
Another strategy for MTMP is to identify candidates with multiple gaps in medication
adherence. The ACG System evaluates 17 conditions that warrant ongoing medication
therapy. Patients that frequently exhaust their drug supply before refilling prescriptions
will have lower Medication Possession Ratios (MPR). If the patient delays refill past a
grace period, a gap will be identified. Patients with multiple gaps are showing a pattern
of poor pharmacy adherence and may be good candidates to be offered medication
therapy management. The Care Management List (Figure 1) can be used to identify
members with multiple gaps. Once identified, the Comprehensive Patient Clinical Profile
(Figure 2) provides detail about the cost and risk profile of the member, including any
gaps in chronic medication use.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Managing Financial Risk for Pharmacy Benefits
5-11
Figure 2: Comprehensive Patient Clinical Profile
Applications Guide
The Johns Hopkins ACG System, Version 9.0
5-12
Managing Financial Risk for Pharmacy Benefits
This page was left blank intentionally.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Capitation and Rate Setting
6-i
6 Capitation and Rate Setting
Capitation and Rate Setting......................................................................... 6-1
H
H
ACGs in Multivariate Models.................................................................... 6-2
Predictive Model Predicted Resource Index (the PM PRI Score) ............. 6-2
Table 1: Predictive Ratios by Quintile for The Johns Hopkins
ACG Dx-PM Applied to Commercial and Medicare Populations ............. 6-3
Underwriting .............................................................................................. 6-4
Table 2: Actuarial Cost Projections .......................................................... 6-6
H
H
H
H
H
H
Applications Guide
H
H
H
H
The Johns Hopkins ACG System, Version 8.0
6-ii
Capitation and Rate Setting
This page was left blank intentionally.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Capitation and Rate Setting
6-1
Capitation and Rate Setting
The ACG System has made it possible to accomplish risk adjustment with fairly simple
and straightforward analytic strategies and the ACG actuarial cells have long been the
primary actuarial method for capitation and rate setting. Actuarial cells represent a fixed
number of discrete categories into which individuals are placed based on their expected
use of resources.
There are a number of advantages associated with using an actuarial cell-based approach
to risk adjustment for capitation and underwriting, which include:
Simplicity. Once the population has been classified into around 100 ACG cells, it is
possible to risk-adjust the population by using a spreadsheet. Some users have chosen to
simplify this approach even further by collapsing the ACGs into smaller homogeneous
groupings called resource utilization bands (RUBs). Even when grouped into RUBs,
studies indicate that ACGs retain much of their explanatory power.
Less prone to manipulation. Particularly in applications involving rate setting, there
could be incentives to manipulate risk-adjustment strategies to increase payment. Unlike
some other disease-specific risk adjusters, aggressive efforts to capture additional
diagnostic codes on the part of providers will have a more limited impact on ACG
assignments. Where “code creep” associated with general increases in completeness and
accuracy of coding exists, the simplicity of the ACG System makes it very easy to
identify this trend and to implement appropriate action, such as recalibration of the
underlying cost weights.
Stability. The conceptual elegance and underlying simplicity of ACGs have made the
system very stable over long periods. The underlying clinical truth captured by ACGs
does not change dramatically with each new data set and each new application.
Ease of making local calibrations. It is very easy to recalibrate ACG-based actuarial
cells to reflect local differences in patterns of practice, benefit structure, and provider
fees. Especially for capitation and rate-setting tasks, we encourage you to calibrate the
ACG output to reflect the unique nature of the local cost structure. The same simplicity
that makes it possible to risk-adjust using a spreadsheet makes it equally possible to
accomplish recalibration using the same types of simple tools.
The ultimate testimony to the value of ACGs used as the basis of actuarial cells is the fact
that for over a decade they have been used to facilitate the exchange of many billions of
dollars within numerous private and public health plans in both the United States and
Canada.
\ Example: For a simple case study illustrating the use of ACG actuarial cells for
prospective payment see “The Development of Risk-Adjusted Capitation Payment
System For Medicaid MCOs: The Maryland Model”, Weiner et al, Journal of
Ambulatory Care Management, January, 1998.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
6-2
Capitation and Rate Setting
ACGs in Multivariate Models
Multivariate regression for risk adjustment has been used for many years by some of the
more sophisticated users of the ACG System. If additional risk descriptors are available
beyond diagnosis, age, and sex, this approach has the potential for improved predictive
models that have both actuarial and payment applications.
The strength of regression-based strategies is the ease with which additional risk factor
information can be incorporated and thereby introduce better control for the effects of
case-mix. If you have access to additional well-validated risk factor data and if you have
previous experience using regression models within your organization, then you should
consider using regression. In regression strategies, ACGs, ADGs, and EDCs remain
valuable as distinct risk factors to be supplemented by additional data. NOTE: Although
EDCs are useful for identifying individuals with specific high impact diseases, it is
important to note that they do not account for burden of co-morbidity as do ACGs.
Therefore, we do not generally recommend that EDCs be used as the only means of
controlling for case-mix in regression analysis.
However, there is also a potential drawback since regression may introduce some
assumptions and statistical pitfalls that can be troublesome without seasoned analytical
support. Their inherent complexity makes them difficult to calibrate to local cost
patterns, and regression models are also potentially easier to game because more factors
can be manipulated. Finally, while it is possible to introduce a wide range of variables
that improve the model’s explanatory power, this explanatory power is often confined to
the data set and time period on which the model is based. The model’s results may end
up differing significantly from year to year depending on the inter-relations of the myriad
risk factors that have been included, a phenomenon referred to as over-fitting.
Predictive Model Predicted Resource Index (the PM PRI Score)
To address some of the analytic challenges inherent in regression-based approaches, the
ACG Predictive Model provides a ready-made solution and assigns a relative value that
can be readily converted to dollars. Termed the Predicted Resource Index (or PRI for
short), this output is most relevant for prospective financial applications. Table 1
presents Predictive Ratios by Quintile for the diagnosis based, Dx-PM, applied to
commercial and Medicare populations.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Capitation and Rate Setting
6-3
Table 1: Predictive Ratios by Quintile for The Johns Hopkins ACG DxPM Applied to Commercial and Medicare Populations
Predictive Ratio
Lowest Quintile Total Spending
Year 1
2nd Quintile Total Spending
Year 1
3rd Quintile Total Spending
Year 1
4th Quintile Total Spending
Year 1
Highest Quintile Total Spending
Year 1
Commercial
Medicare
1.29
1.08
1.10
1.13
1.13
1.07
1.04
0.98
0.88
0.93
Ratios reflect actual year-2 costs for each year-1 “quintile” cohort divided by their
predicted costs.
One important caveat is worth noting here. Though not included in the results presented
in Table 1, prior cost is available as an optional risk factor in Dx-PM. Although
inclusion of pharmacy cost information improves model performance, we do NOT
recommend that models using the optional pharmacy cost predictor be applied to
capitation rate setting. Instead, we suggest that the Dx-PM model, relying only on ICD
input variables, be used for such a purpose.
We take this position for the same reason we believe that episode groupers that rely on
procedure codes (such as CPT) and Rx-groupers based on use of specific medications (as
defined by NDC codes) should not be used for rate-setting purposes or efficiency
profiles. Risk factor variables of this type, which are directly defined by the providers’
clinical practices, are potentially intertwined with patterns of over use or under use.
Risk-adjusted rates based on these factors may, in a circular manner, lead to setting rates
that are inappropriate--either too high or too low. Moreover, when risk factors are
determined by such drug use (or procedural) delivery patterns, providers who practice
efficiently could potentially be penalized for their efficiency. This circularity issue is not
a major concern when only diagnostic information (not linked to specific types or settings
of service) is used as the main source of information on risk factors.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
6-4
Capitation and Rate Setting
Underwriting
The ACG predictive models, calibrated for high-risk case-identification, provide
underwriters with a suite of tools to estimate future resource use based on the case-mix of
the enrolled population, which offers an improvement over more traditional prior
utilization models. For example, in addition to just estimating future resource use, the
models can also be used to help identify persons expected to convert from relatively low
to relatively high resource use. This not only improves the quality and accuracy of
underwriting, but also provides opportunities for reducing costs for employers by getting
at-risk employees enrolled in timely case management interventions to help reduce both
future medical expenses and illness-associated absenteeism.
The ACG predictive models are especially useful for small group underwriting because
the movement of one or two high- risk individuals into or out of a plan can have
potentially dramatic effects on costs for a small group. Small employer groups are
sensitive to price and have a tendency to shop for a new carrier at renewal time. The
initial rate process uses more data than is feasible during a typical renewal; therefore, the
initial rate process often produces the most competitive rates. Small groups exhibiting
low risk can often find rates lower than with their current provider; however, small
groups exhibiting a history of high expenditures may find going to a new insurer
prohibitively expensive. This type of selection bias can lead to a very high risk pool and
a future inability of a plan to offer attractive rates to retain the healthy groups. In order to
retain the best business, insurers are faced with the difficult task of offering competitive
pricing for these small groups by trying to accurately match premium revenue to
expected expense while complying with existing rating regulations. The Johns Hopkins
suite of Predictive Models provides, health plans the tools necessary to leverage existing
medical and pharmacy claims in order to better estimate risk and better set premiums for
small group renewal.
There are several benefits to using predictive modeling within the underwriting process:
•
There is greater efficiency. Predictive modeling can provide an automated risk
assessment on every member; thereby reducing the medical underwriting effort. This
reduction in effort, in turn, reduces the elapsed time needed for analysis and
consequently will reduce the lag between the experience period and the rating period.
Rx-PM can reduce this lag further. This leads to greater accuracy.
•
The ACG predictive models provide an objective, reproducible method which is
favored by regulators. It offers greater consistency among underwriters and is more
defensible to customers than manual approaches.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Capitation and Rate Setting
6-5
•
The various clinical groupings and markers from the system provide supporting detail
that can be used by sales and marketing. Discordant predictions based on Rx-PM and
Dx-PM can be used as a data quality check and prompt more targeted investigation by
medical underwriters.
•
Predictive modeling better matches premium to future costs allowing for more
competitive renewals and improved customer retention.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
6-6
Capitation and Rate Setting
Table 2: Actuarial Cost Projections
Age/Sex
Mean Mean %
#
Relative Observed/ National Local Total
Rx
High
%
%
%
%
%
Employer Cases
Risk
Expected CMI
CMI PRI
PRI Risk HOSDOM Frail Chronic Psychosocial Discretionary
33472*08
10
0.78
0.57
0.66
0.59
0.57
0.39
0.0
0.0
0.0
20.0
20.0
10.0
1214*37
11
0.74
2.19
0.61
0.52
0.80
1.74
0.0
0.0
0.0
27.3
18.2
9.1
1317*37
11
0.72
1.73
0.44
0.43
0.40
0.20
0.0
0.0
0.0
27.3
18.2
0.0
65466*93
11
1.02
0.54
1.27
1.21
0.98
0.98
0.0
9.1
18.2
36.4
18.2
9.1
4114253*37
12
0.85
0.35
0.52
0.51
0.39
0.27
0.0
0.0
0.0
25.0
16.7
0.0
34565*08
16
1.21
0.88
0.97
0.94
1.23
0.59
6.3
6.3
0.0
25.0
12.5
0.0
65215*16
19
1.15
0.72
1.34
1.17
0.86
0.47
0.0
0.0
0.0
21.1
21.1
10.5
1322*37
21
0.97
0.55
0.40
0.41
0.59
0.39
4.8
4.8
0.0
14.3
9.5
0.0
32316*08
22
0.89
0.47
0.65
0.56
0.80
1.14
0.0
0.0
0.0
27.3
18.2
4.5
74134*06
22
1.04
0.95
1.63
1.68
2.69
2.98
4.5
0.0
0.0
63.6
27.3
18.2
4112725*11
24
1.01
0.95
0.73
0.63
0.98
1.39
0.0
0.0
0.0
29.2
8.3
4.2
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Capitation and Rate Setting
6-7
The Actuarial Cost Report provided in Table 2 is a standard report produced by the
software and represents a summary of information relevant for actuarial purposes and for
differentiating groups as high medium and low risk. This analysis provides a number of
aggregate measures for both current and future costs expressed as a relative index (scores
equal to 1.0 indicate average morbidity or risk, greater than 1.0 indicate greater than
average morbidity burden or risk and less than 1.0 less than average). The National CMI
is a concurrent measure that compares the group case mix to a national benchmark based
on the mix of ACGs assigned to the members of the group. The Local CMI is a similar
measure but the comparison group is based on the population presented to the ACG
System. Mean Total PRI is a measure of prospective risk using the ACG predictive
model to forecast total cost relative to the plan average. Likewise, the Mean Rx PRI
measures the prospective risk of pharmacy cost relative to the plan average. These
resource indicators can be compared to the age-sex relative risk. When age-sex relative
risk is equal to the local CMI, the risk is driven by the age and sex of the group. When
age-sex relative risk is lower than the local CMI, the risk is driven by disease burden
more than the age-sex mix of the group. There is an additional index of the observed cost
to the expected cost (accounting for the local CMI) as a measure of how efficiently the
group utilizes services as compared to the population mean.
There are additional rate-based measures provided to describe the factors contributing to
group risk. Groups with higher disease burdens will also generally tend to have higher
prevalence rates of high risk members who are more likely to have chronic conditions,
higher rates of hospital dominant and frailty conditions, and higher rates of psychosocial
conditions. Comparisons can be made between the group and the population mean by
comparing the groups tab to the "overall" tab in the analysis window.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
6-8
Capitation and Rate Setting
This page was left blank intentionally.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Final Considerations
7-i
7 Final Considerations
Introduction................................................................................................... 7-1
H
H
Art of Risk Adjustment ................................................................................ 7-1
H
H
Figure 1: Risk Adjustment Pyramid.......................................................... 7-1
H
H
Time Frames and Basic Population Perspectives ...................................... 7-2
H
H
Figure 2: Typical Timeline for Risk Adjustment...................................... 7-3
H
H
Handling New or Part-Year Enrollees........................................................ 7-4
H
H
Non-Users Who are Eligible to Use Services ............................................ 7-5
H
H
Sample Size .................................................................................................... 7-5
H
H
Handling High Cost or Outlier Cases ......................................................... 7-6
H
H
Constructing Resource Consumption Measures........................................ 7-6
H
H
Summarizing Total or Ambulatory Charges.............................................. 7-6
Ambulatory Encounters ............................................................................. 7-7
Applications Guide
H
H
H
H
The Johns Hopkins ACG System, Version 9.0
7-ii
Final Considerations
This page was left blank intentionally.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Final Considerations
7-1
Introduction
The purpose of this chapter is to highlight and discuss some of the key analytical and
technical issues associated with the application of diagnosis-based risk adjustment in
populations. These issues affect both the framing and interpretation of analyses. Much
of this discussion relates to forming a population for risk adjustment, determining which
members to include and to exclude, and circumstances where sampling is appropriate.
Art of Risk Adjustment
Figure 1: Risk Adjustment Pyramid
While the essential methodological underpinnings of risk adjustment are straightforward,
technical challenges may be experienced when putting health-based risk adjustment in
place within an organization. Figure 1 is intended to help graphically illustrate the
variety of ways in which risk adjustment is most commonly applied within healthcare
organizations today. Some implementations, such as needs assessment or
payment/finance applications apply to the entire population base. Other implementations,
such as care-management or disease-management interventions, focus only on targeted
population subgroups. Depending on the application or the question being asked, it is
important to appropriately define the denominator or the population of interest. Another
key consideration is time frame—is the analysis retrospective or concurrent in nature
involving a comparison of morbidity across or between population subgroupings or is the
application prospective or predictive in nature? Each of these issues will be discussed in
more detail subsequently.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
7-2
Final Considerations
Time Frames and Basic Population Perspectives
For profiling, the population’s health characteristics (i.e., diagnoses used to adjust the
profiles) typically come from the same time period as the resource use being profiled.
Thus, the process is designated retrospective or concurrent. For example, to understand
the differences in per person pharmacy use across two provider panels in a given year,
you would assign risk assessment variables using diagnosis codes derived from patient
physician contacts during that same year.
In contrast, the most common approach for risk adjusting capitation payments is to
prospectively set rates in the following years for a cohort of enrollees based on the
diagnosis codes documented in data derived from the prior year(s). For administrative
reasons, there is usually a lag period (often of about three months’ duration) between the
risk assessment period and the target payment period. Additionally, some patients may be
enrolled during the first period but not the second, and vice versa. Others may be enrolled
during the entire period but use no services. Therefore, they do not have diagnosis
assignments during the first 12-month risk assessment period. These are a few of the
challenges that the prospective capitation process faces. The prototypical time line for
this process and the concurrent profiling process are outlined in Figure 2.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
Final Considerations
7-3
Figure 2: Typical Timeline for Risk Adjustment
12 Months
3 Months
Risk measurement period (also assessment Data lag period
period for retrospective profiling)
Applications Guide
3 Months
Analysis/rating
process
12 Months
Risk measurement period (also assessment
period for retrospective profiling)
The Johns Hopkins ACG System, Version 9.0
7-4
Final Considerations
There are numerous technical approaches for dealing with the data lag problem for
prospective applications. The simplest approach is to take the predictions provided by
the ACG-PM model. This, of course, means that the prediction is already aged by the
period of the lag. An alternative is to use an historical database to determine trended
resource use for successive years. For example, at Plan Z, by going back to a time period
24 months before the target year (the target year being months 25-36), it would be
possible to associate future resource use based on risk scores assigned during the
previous time period. In this simulation, months 1-12 would be used to predict months
13-24. Results from this model could then be applied to months 13-24 to yield
predictions for months 25-36. In essence, modeling would occur across the lag period.
These longer term models could serve as provisional models for a period of interest and
could be replaced once a potentially more predictive annual model becomes available.
Yet a third approach is that implemented by Minnesota Medicaid and the Buyers Health
Care Action Group (BHCAG) and several other tiered network applications where grouplevel predictions are based on historical group-level concurrent profiles with a trend
factor applied to generate an estimate of future resource expectations at the group level.
The assumption behind using group-level concurrent profiles to predict future costs is
that the case-mix of a group (at least of sufficient size) will not change much over time
and that projections based on concurrent profiles provide more accurate projections than
individual level predictions. In such an application the concurrent ACG-based profiles
are generally recalibrated approximately every three months and new “targets” are set,
thus mitigating the data lag problem.
Handling New or Part-Year Enrollees
Most ACG applications involve the analyst viewing a snapshot of the utilization history
of plan members during a particular period of time. If any members of the risk pool have
been eligible to use services for a period of time that is shorter than the in-scope period,
both their diagnosis history and their resource consumption profile may differ from
members who were enrolled for the entire period. For the most part, and so as long as
these new enrollees are randomly distributed across the population (and population subgroupings), their impact is minimal. If, however, large numbers of enrollees are
concentrated in one provider group being profiled or one employer group for which rates
are being set, concentration of new enrollees may bias results to make this group look
“healthier” than they otherwise might have if complete diagnoses and claims information
had been available for them.
In general, when including individuals who are not eligible for the entire enrollment
period, it is recommended that results be scrutinized closely. One approach would be to
compare results excluding and including these individuals to help assess whether their
inclusion has introduced any systematic bias. Another strategy for assessing their impact
would be to examine ACG distribution across the various units of analysis, such as by
provider. A disproportionate number of persons assigned to ACGs 5100 or 5110 and
5200 (i.e., no diagnoses and non-user ACGs) may indicate the enrollee cohort entered the
plan near the end of the analysis period and may lack sufficient contact with the provider
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Final Considerations
7-5
to allow accurate overall ACG assignment. Such groups can, and perhaps should, be
eliminated from the analysis or be reported with appropriate caveats. The specific
approach used will vary for each analysis/organization based on the quality of the
alternatives. Although new enrollees’ ICD codes may be incomplete, risk adjustment
based on a limited pool of diagnoses generally provides more accurate risk adjustment
than do alternative demographic adjustments.
Non-Users Who are Eligible to Use Services
Most grouping methods and case-mix measurement tools that focus on episodes of care
restrict their attention to the subset of a population that actually consumes resources (e.g.,
those visiting a provider or being admitted to the hospital). The most common
applications of these tools, provider profiling and other retrospective applications, are
concerned exclusively with users of services since only for these members can a
meaningful profile be developed. However, for capitation rate development and other
prospective applications, non-users are of great importance since many, if not most, of
the enrollees who do not use services in the current period will consume services, to at
least some degree, in the future period. Since capitation payments are made regardless of
whether the member interacts with the capitated provider, the characteristics of non-users
are important. For profiling, consideration of the percentage of enrollees assigned to a
physician who are non-users may provide information on access issues or illustrate
differences in provider practice patterns. In general, population-oriented analysis will
have more flexibility and be more comprehensive if both users and non-users are
included.
Sample Size
The question of what is an appropriate minimum enrollee/patient sample size arises at
many levels of the risk adjustment process. As a general rule, the larger the sample size,
the better. Ideally, the total population used to perform ACG-based analysis should be
larger than 20,000 individuals. Also, ideally, there should be a minimum of 30-50 cases
in each ACG cell. Smaller sample sizes may be applied but users should be cautious of
instability created by small cell size.
Sample size plays an important role in profiling provider practice patterns. Even when the
underlying ACG weights are calculated using a large reference population, providers
treating relatively few patients may be unfairly skewed simply because of the effects of
random error resulting from sample size.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
7-6
Final Considerations
Handling High Cost or Outlier Cases
How high cost or outlier cases are included affects many risk adjustment applications. If
untruncated cost weights of very high cost individuals are included in the calculation of
either concurrent or prospective risk scores, there will be a tendency for the variability of
all cost estimates or risk scores to increase. Similarly, high cost cases can create problems
for physician profiling analyses where the inclusion of one patient my falsely identify a
provider as an outlier physician. Yet, at the same time, it is these very high cost or
“outlier” patients that the ACG-PM high risk case identification tool is designed to
identify. Thus, the use of truncation depends upon the application. For applications that
relate to rate setting or profiling, a conservative strategy would be to top code (set a
ceiling) for per person costs to $50,000.
Constructing Resource Consumption Measures
Key to any ACG-based application for either physician profiling or capitation is
consideration of how the resource use measure is defined. Most analyses developed to
date have focused on visit rates, ambulatory charges, or total charges. However, more
recent work is being conducted to assess the ACG System as a means of evaluating
pharmacy use, understanding specialist use, and assessing quality of care.
Summarizing Total or Ambulatory Charges
Most plans retain the submitted charge, allowed or eligible amount, and paid amount for
healthcare services in their machine-readable claims files. The submitted charge refers to
the charge submitted on the provider’s claim. The allowed or eligible amount refers to the
amount the plan has determined it will pay for the covered service, after applying
reasonable and customary charge screens or a fee schedule. The paid charge is the
allowed amount reduced by any applicable copayments and deductibles required by the
subscriber.
\ Tip: Providing summarized total charges (including pharmacy cost) and/or a separate
summary pharmacy cost field on the patient input file will improve predictive model
performance.
Typically, it is recommended that users aggregate either the paid charge or the allowed
amount for each patient as the most appropriate measure of total and/or ambulatory
charges. Since the ACG System can be used to compare the consumption of resources
across groups, different copayment and deductible amounts, as well as different paid
charge amounts, may prevent accurate comparison of different subscriber groups.
Therefore, the allowed amount is typically used as the best measure of resource
consumption when comparing groups or profiling providers. In the case of capitation,
where the focus is in plan liability, paid amounts may be appropriate.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Final Considerations
7-7
Ambulatory Encounters
Some users, particularly those interested in ambulatory provider productivity, use the
ACG System to case-mix adjust profiles of provider-patient contacts. Users should
realize the potential difficulties associated with trying to define ambulatory encounters.
Physician visits are relatively straightforward mechanisms for estimating face-to-face
encounters; however, tabulating ancillary and surgical services into encounters is
problematic. This issue is a focus of much ongoing research and few workable solutions
currently exist. However, in the context of provider profiling, it is probably sufficient for
analysts to estimate ambulatory encounters in exactly the same way for each group to be
compared. Using this approach, even if the estimate of an ambulatory encounter is biased,
valid ACG-adjusted comparisons can still be performed. The notion of using compatible
techniques for estimating ambulatory encounters is especially important when the
comparison involves two different types of service delivery environments, such as
comparing a fully-capitated, at-risk independent practice association (IPA) and a staff
model HMO operating under a negotiated global budget.
Applications Guide
The Johns Hopkins ACG System, Version 9.0
7-8
Final Considerations
This page was left blank intentionally.
The Johns Hopkins ACG System, Version 9.0
Applications Guide
Index
IN-1
Index
A
D
ACG
ambulatory encounters, 7-7
capitation and rate setting, 6-1
concurrent weights, 3-4
multivariate models, 6-2
summarizing total or ambulatory charges, 7-6
ACG PM
local calibration of scores, 4-19
predictive model predicted resource index (PM PRI
score), 6-2
probability score, 4-7
Actual to morbidity expected to create a morbidity ratio,
comparison, 3-17
Addressing the impact of age on the calculation of ACG
weights, 3-11
Adjustments for inflation, 4-18
Ambulatory encounters, 7-7
Analysis of the O/E ratios, 3-15
Application guide
navigation, 1-1
Applications guide content, 1-2
Applications guide objective, 1-1
Art of risk adjustment, 7-1
Define a patient panel, 3-13
Disease management candidates, 4-14
C
Calculate
expected levels of resource use, 3-17
expected values for the patient panel, 3-13
O/E ratio for the patient panel, 3-15
Calculate
a morbidity ratio for the patient panel
performance assessment, 3-15
Capitation and rate setting, 6-1
underwriting, 6-4
Case mix control, 4-16
Clinical screening by care and disease management
adjustments for inflation, 4-18
converting scores to dollars, 4-17
how to rescale and assign dollar values, 4-17
local calibration of ACG PM scores, 4-19
prospective risk scores, 4-17
Clinical screening by care and disease managers, 4-1
Comparison
actual to morbidity expected to create a morbidity ratio,
3-17
specialists to specialists-intra-specialty expected levels
of service and costs, 3-23
Concurrent ACG weights, 3-4
Constructing resource consumption measures, 7-6
Converting scores to dollars, 4-17
Customer commitment and contact information, 1-5
Customizing risk scores using local cost data, 3-5
Applications Guide
E
EDC
context of practitioner profiles, 3-24
EDCs
age/sex adjusted comparison of disease distributions
across populations-standardized morbidity ratios
(SMRs), 2-9
calculating age/sex adjusted standardized morbidity
ratios, 2-10
comparing disease distributions across two or more
subpopulations, 2-5
epidemiology of diseases within a single population, 2-2
using a combination of EDCs and ACGs to support case
management and disease management, 2-12
Evaluating productivity and distributing workload, 3-26
Examining differences in prescribing risk, 5-1
Examples
profiling primary care physicians, 3-18
F
Figures
care management list, 5-10
combining Rx and Dx predictive modeling scores for
targeted intervention, 4-5
percent correctly identified as high cost, 4-3
comparison of ACG and age/gender-based O/E ratiospractices of all BC physicians, 3-16
comprehensive patient clinical profile, 4-11, 5-11
cost predictions by select conditions, 4-15
hospital prediction, 4-9
percent of patients identified by ICD or NDC or both, 44
percentage of patients with selected outcomes by ACG
PM risk group, 3-28
pharmacy adherence, 4-10
risk adjustment pyramid, 7-1
Final considerations
art of risk adjustment, 7-1
handling high cost or outlier cases, 7-6
handling new or part-year enrollees, 7-4
introduction, 7-1
non-users who are eligible to use services, 7-5
sample size, 7-5
time frames and basic population perspectives, 7-2
H
Handling high cost or outlier cases, 7-6
final considerations, 7-6
Handling new or part-year enrollees, 7-4
The Johns Hopkins ACG System, Version 9.0
IN-4
final considerations, 7-4
Health status monitoring, 2-1
Performance assessment
addressing the impact of age on the calculation of ACG
weights, 3-11
High-risk case identification for case management, 4-1
How to rescale and assign dollar values, 4-17
step 1, 4-17
step 2, 4-17
step 3, 4-18
step 4, 4-18
I
Including part-year enrollees, 3-5
Installation and usage guide content, 1-3
Introducing primary responsible phsyician (PRP), 3-21
Introduction
applications guide content, 1-2
applications guide objective, 1-1
customer commitment and contact information, 1-5
final considerations, 7-1
installation and usage guide content, 1-3
technical reference guide content, 1-4
the Johns Hopkins ACG® system, 1-1
L
Local calibration of ACG PM scores, 4-19
M
Managing financial risk for pharmacy benefits, 5-1
examining differences in prescribing risk, 5-1
medication therapy management program (MTMP)
candidate selection, 5-9
predicting pharmacy use, 5-5
Medication therapy management program (MTMP)
candicate selection, 5-9
Multivariate models, 6-2
N
Navigation
application guide, 1-1
Non-users who are eligible to use services, 7-5
final considerations, 7-5
O
O/E ratios in performance assessment, 3-17
P
Performance assessment
evaluating productivity and distributing workload, 3-26
Performance assessment
concurrent ACG, 3-4
customizing risk scores using local cost data, 3-5
software-produced weights and their uses, 3-2
Performance assessment
The Johns Hopkins ACG System, Version 9.0
Index
goals and objectives, 3-1
introduction, 3-1
theory and background, 3-1
Population-based approach to practitioner profiling
performance assessment, 3-12
Predicting pharmacy use, 5-5
Predictive model
predicted resource index (PM PRI score), 6-2
Preparatory steps
performance assessment, 3-13
Probability score, 4-7
Prospective risk scores, 4-17
Provider
assessment, 3-1
Performance assessment
calculate expected values for the patient panel, 3-13
calculate morbidity ratio for the patient panel, 3-15
define a patient panel, 3-13
population-based approach to practitioner profiling, 312
preparatory steps, 3-13
Performance assessment
analysis of the O/E ratios, 3-15
calculate O/E ratio for the patient panel, 3-15
comparing specialists to specialists-intra-specialty
expected levels of service and costs, 3-23
comparison of actual to morbidity expected to create a
morbidity ratio, 3-17
EDCs in the context of practitioner profiles, 3-24
examples of primary care physicians profile, 3-18
expected levels of resource use, 3-17
introducing primary responsible physician (PRP), 3-21
summary, 3-29
using various O/E ratios, 3-17
Q
Quality of care assessment, 3-27
R
Risk stratification, 4-12
S
Sample size, 7-5
final considerations, 7-5
Selecting the right tool
evaluating productivity and distributing workload, 3-26
Performance assessment
quality of care asssessment, 3-27
Software-produced weights and their uses, 3-2
Specialists to specialists-intra-specialty expected levels of
service and costs, comparison, 3-23
Summarizing total or ambulatory charges, 7-6
ACG, 7-6
T
Tables
Applications Guide
Index
IN-3
major EDC prevalence for four populations in three
exemplary health plans, 2-7
measuring return on investment, 4-16
member demographic and plan features for four
populations in three exemplary plans, 2-5
movers analysis-tracking morbidity burden over time,
amount of data and its impact on model performance, 4-
2
care management listing, 4-8
comparison of actual and ACG expected costs months
of member enrollment (PMPM) versus (PMPY)
weight calculation approaches, 3-8
comparison of case-mix adjusted practice profiles for
two general practitioners identified as high cost
outliers in unadjusted analyses, 3-20
comparison of characteristics affecting physician
productivity, 3-27
comparison of patient populations and payments for two
general practitioners identified as high cost outliers,
1999, 3-19
comparison of PMPM and PMPY average costs by
months enrolled within an HMO population, 3-6
cost predictions for selected Rx-MGs, 5-6
distribution of EDCs within a commercial HMO
population, 2-3
distribution of RUB co-morbidity levels within selected
EDC disease categories and relative resource use
morbidity ratios for each EDC/RUB category, 2-13
effect of enrollment period on selected ACG-specific
weights, 3-10
estimated concurrent resource use by RUB by MEDC
(samples), 4-14
estimating costs in a sample of cases, 4-18
example internist-global expenditures on the patient
panel, by category of service, 3-23
example of an EDC report for a general practitioner, 326
example of an EDC report for an internist, 3-25
Applications Guide
2-1
number of cases and the Johns Hopkins ACG Dx- PM
predicted relative resource use by risk probability
thresholds for selected chronic conditions, 4-6
observed to expected standardized morbidity ratio by
MEDC, 2-9
percentage distribution of each co-morbidity level
within an EDC (samples), 4-13
predictive ratios by quintile for the Johns Hopkins ACG
Dx-PM applied to commercial and Medicare
populations, 6-3
risk weights and scores, 3-3
example calculation of expected values, 3-14
standardized morbidity ratio by Rx-MG, 5-2
summary descriptive statistics for a commercial HMO’s
EDC distribution, 2-4
Technical reference guide content, 1-4
The Johns Hopkins ACG System, 1-1
Time frames and basic population perspectives, 7-2
final considerations, 7-2
U
Underwriting, 6-4
The Johns Hopkins ACG System, Version 9.0
IN-4
Index
This page was left blank intentionally.
The Johns Hopkins ACG System, Version 9.0
Applications Guide