DISEASE MANAGEMENT
Volume 11, Number 1, 2008
© Mary Ann Liebert, Inc.
DOI: 10.1089/dis.2008.111723
Guided Care: Cost and Utilization Outcomes in a
Pilot Study
MARTHA L. SYLVIA, R.N., M.S.N., M.B.A.,1 MICHAEL GRISWOLD, Ph.D.,2
LINDA DUNBAR, Ph.D.,1 CYNTHIA M. BOYD, M.D., M.P.H.,2,3
MARGARET PARK, M.P.H.,2 and CHAD BOULT, M.D., M.P.H., M.B.A.2,3
ABSTRACT
Guided Care (GC) is an enhancement to primary care that incorporates the operative principles of disease management and chronic care innovations. In a 6-month quasi-experimental
study, we compared the cost and utilization patterns of patients assigned to GC and Usual
Care (UC). The setting was a community-based general internal medicine practice. The participants were patients of 4 general internists. They were older, chronically ill, communitydwelling patients, members of a capitated health plan, and identified as high risk. Using the
Adjusted Clinical Groups Predictive Model® (ACG-PM), we identified those at highest risk
of future health care utilization. We selected the 75 highest-risk older patients of 2 internists
at a primary care practice to receive GC and the 75 highest-risk older patients of 2 other internists in the same practice to receive UC. Insurance data were used to describe the groups’
demographics, chronic conditions, insurance expenditures, and utilization. Among our results, at baseline, the GC (all targeted patients) and UC groups were similar in demographics and prevalence of chronic conditions, but the GC group had a higher mean ACG-PM risk
score (0.34 vs. 0.20, p 0.0001). During the following 6 months, the GC group had lower unadjusted mean insurance expenditures, hospital admissions, hospital days, and emergency
department visits (p 0.05). There were larger differences in insurance expenditures between
the GC and UC groups at lower risk levels (at ACG-PM 0.10, mean difference $4340; at
ACG-PM 0.6, mean difference $1304). Thirty-one of the 75 patients assigned to receive
GC actually enrolled in the intervention. These results suggest that GC may reduce insurance
expenditures for high-risk older adults. If these results are confirmed in larger, randomized
studies, GC may help to increase the efficiency of health care for the aging American population. (Disease Management. 2008;11:29–36)
BACKGROUND
ciaries have at least 1 chronic condition, 63%
have 2 or more, and 20% have 5 or more.2 The
United States currently spends more than 75%
of its $1.4 trillion health care budget on chronic
disease.3 A strong association exists between
the number of chronic conditions and the like-
B
Y 2010, 141 MILLION AMERICANS are expected
to be living with at least 1 chronic condition, and 70 million will have multiple chronic
conditions.1 Already, 78% of Medicare benefi1Johns
Hopkins HealthCare, Glen Burnie, Maryland.
Hopkins University Bloomberg School of Public Health, Baltimore, Maryland.
3Johns Hopkins University School of Medicine, Baltimore, Maryland.
2Johns
29
30
lihood of hospitalization and overall health
care utilization. The number of chronic conditions is related to per capita annual medical
costs, increasing from $163 for those with no
chronic conditions to $13,730 for those with 5
or more conditions.2 Currently, 24% of
Medicare beneficiaries account for 80% of
Medicare’s total health care spending.2 Not
only are individuals with 4 or more chronic
conditions more likely to be hospitalized, they
are 99 times more likely to have potentially preventable hospitalizations.4
Today’s acute-care-focused health care delivery system, which emphasizes the treatment
of episodic events, is not designed to deliver
the type of chronic care needed by the majority of older Americans who have complex
chronic conditions.5,6,7,8 With this system, people with chronic illnesses often find it challenging to get the necessary support to manage
their conditions, obtain treatment, prevent
complications, and participate in their care
plans.9
To improve quality of life for older adults
with multimorbidity and complex care needs
and promote the efficient use of resources, we
designed an enhancement to primary care.
Guided Care (GC) is primary health care infused with the operative principles of recent innovations to ensure optimal outcomes for patients with chronic conditions and complex
health care needs. A registered nurse who has
completed a supplemental educational curriculum works in a practice with several primary care physicians (PCPs) to provide costeffective chronic care to 50 to 60 multimorbid
patients. Using a Web-accessible electronic
health record (EHR), the guided care nurse
(GCN) collaborates with the patient’s PCP in
conducting 8 clinical processes: assessing the
patient and primary caregiver at home, creating an evidence-based care plan, promoting patient self-management, monitoring the patient’s conditions monthly, coaching the
patient to practice healthy behaviors, coordinating the patient’s transitions between sites
and providers of care, educating and supporting the caregiver, and facilitating access to
community resources. A more detailed description of GC has been published previously.10
SYLVIA ET AL.
In 2003–2004 we conducted a pilot study of
a partial version of GC, in which the chronic
disease self-management classes and the caregiver education and support components were
omitted because of budgetary constraints. We
sought to compare the cost and utilization patterns of patients assigned to receive GC and
those assigned to receive Usual Care (UC).
METHODS
Study design
We conducted a quasi-experimental study of
the partial version of GC. Four general internists at an urban community practice were
assigned to provide either GC or UC to their
highest risk patients aged 65 years or older who
were enrolled in a capitated health plan with
benefits similar to Medicare.
We used insurance information and a predictive modeling algorithm, the Adjusted Clinical Groups Predictive Model® (ACG-PM), part
of the Johns Hopkins ACG® Case Mix System,
to identify these physicians’ high-risk patients
for the study.11 ACG-PM software analyzes
health insurance enrollment records and diagnostic information from claims to compute the
probability that patients will generate insurance expenditures that will rank within the
highest 5% of the population during the following year.
Values of R2 in ACG performance studies
measuring costs or utilization as the outcome
variable vary between 0.07 and 0.23, which
is consistent with the performance of other
similar predictive models.12,13,14,15,16,17,18,19,20,21
C-statistic results for the ACG system in predicting utilization outcomes vary between 0.69
and 0.73 depending on the outcome measured.18,21,22
Through insurance records, we observed
each participating patient’s use and cost of
health services for 6 months. The Johns Hopkins Bloomberg School of Public Health Institutional Review Board approved the study.
Intervention
One GCN was assigned to work with 2 of the
physicians and their high-risk older patients.
GUIDED CARE OUTCOMES
The GCN completed a 25-module curriculum
covering chronic disease management (ie,
depression, falls, congestive heart failure, diabetes, osteoarthritis, chronic obstructive pulmonary disease, angina, dementia, osteoporosis, and constipation), patient preferences, case
management, geriatric assessment and care
planning, transitional care, information technology, patient education, ethnogeriatrics,
community resources, communication with
physicians, and insurance benefits. The GCN
then underwent a 3-month integration into the
practice by working with the 2 GC physicians
and their office staff.
After the integration phase, the GCN began
contacting the high-risk patients of the 2 physicians by phone to schedule times to conduct
initial assessments in the home. When caregivers were identified, they were encouraged
to be present. During October 2003–March
2004, the GCN visited each patient’s home to
assess the patient’s medical, functional, cognitive, affective, psychosocial, environmental,
and nutritional characteristics. The GCN enrolled 31 of the 75 identified high-risk patients
into her caseload during this period. Reasons
for nonenrollment included death, moving out
of the area, changing PCP, and already being
enrolled in a case or disease management program.
The GCN and the PCP worked collaboratively with the patient and caregiver to develop
a care guide that incorporated relevant “best
practice” guidelines for each of the patient’s
conditions. The GCN, who was based in the
primary care practice, then provided regular
monitoring and coaching, transitional care,
care coordination, and access to community resources in collaboration with the 2 PCPs.
Measures
From the health plan’s enrollment data and
insurance claims, we obtained information
about participating enrollees’ baseline characteristics and about the health plan’s subsequent
expenditures for their health care services. The
utilization and expenditure data were not adjusted for inflation. Total expenditures included all costs incurred by the health insurer
for all fee-for-service care. Primary care, labo-
31
ratory, and radiology services, which were capitated by the health plan, were not included.
Pharmacy costs also were excluded because
health plan enrollees used multiple pharmacy
benefit providers within and outside of the
health plan. Utilization outcome measures included hospital admissions, hospital days, and
emergency department (ED) visits (except
those that resulted in a hospital admission). We
used the ACG-PM software to categorize International Classification of Diseases, 9th Revision (ICD-9) diagnostic codes submitted with
the insurance claims to estimate the prevalence
of the chronic conditions shown in Table 1.
Statistical analysis
We analyzed the data using an “intent to
treat” (ITT) framework, thus including all patients assigned to receive GC or UC. We used
the ITT approach as a primary method to compare cost savings to account for those unmeasurable factors, such as adherence and compliance, that made GC-assigned patients likely
to take up the intervention. ITT is considered
one of the best ways to ensure that evaluation
results are unbiased by these unmeasurable
factors.23
For demographic comparisons of the GC and
UC groups, we used histograms, boxplots,
t-tests, and chi-square statistics to evaluate
ACG-PM scores, gender, age, and disease
prevalence. Insurance expenditures, hospital
admissions, hospital days, and ED visits were
compared between the groups. Nonparametric
smooth mean response curves were used to explore the relationships between costs and continuous variables, and a regression plateau
model was developed accordingly to quantify
the trends in expenditure differences between
groups across increasing ACG-PM risk levels,
adjusting for age and gender.24 The regression
plateau model is a special case of a linear spline
regression model, which parameterizes a leveling off of the mean response past a given cut
point after adjusting for other confounding
variables.25 Here, the empirical “leveling off”
observed in the exploratory smoothed response
curves captures a biological effect of a similarity in illness burden across the range of higher
ACG-PM scores.
32
SYLVIA ET AL.
TABLE 1.
BASELINE CHARACTERISTICS
GC (n 62)
UC (n 65)
p-value*
Demographics
Age, mean
Female
ACG-PM, mean
74.1
60.3%
0.34
75.8
47.7%
0.20
.7620
.1241
.0001
Health status
Ischemic heart disease
Congestive heart failure
Hypertension
Diabetes
Osteoarthritis
Parkinson’s Disease
Dementia
Depression
COPD
Chronic conditions, mean (total of 9)
51.6%
30.7%
88.7%
29.0%
48.4%
1.6%
8.1%
12.9%
19.4%
2.9
49.2%
21.5%
86.2%
20.0%
46.2%
7.7%
13.8%
18.5%
21.5%
2.9
.7884
.2421
.6644
.2362
.8011
.1065
.2984
.3900
.7605
.8204
*Chi Square for categorical variables, Student’s t-test for continuous variables.
GC, guided care; UC, usual care; ACG-PM, Adjusted Clinical Groups Predictive Model;
COPD, chronic obstructive pulmonary disease.
RESULTS
Sixty-two GC-assigned and 65 UC-assigned
patients remained enrolled in the health plan
for 6 months and were included in this analysis. At baseline, the GC- and UC-assigned patients were similar in age, gender, number of
chronic conditions, and chronic condition
prevalence (Table 1). GC-assigned patients,
however, had a significantly higher mean
TABLE 2.
ACG-PM risk score than UC-assigned patients
(0.34 vs. 0.20, p 0.0001).
Unadjusted and adjusted expenditure differences between GC and UC groups are reported.
Table 2 shows the unadjusted mean insurance
expenditures and utilization of health care services by the GC and UC groups. Despite the
higher baseline risk of utilization, the GC-assigned patients had lower insurance expenditures; fewer hospital admissions, hospital days,
UNADJUSTED MEAN TOTAL 6-MONTH EXPENDITURES
AND
UTILIZATION
Mean
(95% CI)
Total insurance expenditures
Hospital admissions
Hospital days
Emergency department visits
Primary care visits
GC (n 62)
UC (n 65)
p-value*
$4586
($2678, $6493)
$5964
($3759, $8171)
0.347
0.24
(0.09, 0.39)
0.43
(0.19, 0.67)
0.185
0.82
(0.15, 1.49)
2.45
(0.60, 4.30)
0.103
0.15
(0.00, 0.32)
0.31
(0.12, 0.49)
0.200
5.2
(4.01, 6.37)
4.9
(3.93, 4.88)
0.676
*Student’s t-test
CI, confidence interval; GC, guided care, UC usual care.
GUIDED CARE OUTCOMES
33
Six-month Insurance Expenditures
40000
30000
UC
GC
20000
UC (N⫽65)
GC (N⫽62)
10000
0
0.0
0.1
0.2
0.3
0.4
0.5
Baseline ACG-PM
0.6
0.7
0.8
0.9
FIG. 1. Estimates from the regression plateau model, comparing 6-month insurance expenditures for the usual care
(UC) and guided care (GC) recipients with different ACG-PM scores.
Mean Difference Between Six-month Insurance
Expenditures for UC and GC groups
and ED visits; and more primary care visits
than UC-assigned patients. However, these results were not statistically significant.
The relationship between baseline ACG-PM
levels and subsequent expenditures is shown
in Fig. 1. Smoothed regression techniques suggested a linear relationship in both groups between baseline ACG-PM and total expenditures. The UC-patients had higher costs than
the GC-patients when ACG-PM scores were
0.6. This relationship was difficult to evaluate at ACG-PM scores greater than 0.6 because
of the small number of patients. Fig. 2 shows
that the mean expenditure differences between
the GC and UC groups were greater at lower
baseline ACG-PM scores, after adjusting for
age and gender. These differences were statistically significant at lower ACG-PM scores.
$10,000
$8,000
$6,000
$4340
$3733
$1910
$2518
$3126
$1304
$4,000
$2,000
$0
P⫽0.04
P⫽0.02
-$2,000
P⫽0.06
P⫽0.27
-$4,000
P⫽0.54
-$6,000
P⫽0.75
-$8,000
0.0
0.1
0.2
0.3
0.4
Baseline ACG-PM
0.5
0.6
0.7
FIG. 2. Mean differences (with 95% confidence interval) in 6-month insurance expenditures between guided care
and usual care groups at increasing baseline ACG-PM scores, adjusted for age and gender.
34
SYLVIA ET AL.
Six-month Insurance Expenditures
Cost
40000
30000
UC
GC-received
GC-assignedbut-not-received
20000
GC-received (n⫽31)
10000
UC (n⫽65)
GC-assignedbut-notreceived (n⫽31)
0
0.0
0.1
0.2
0.3
0.4
0.5
Baseline ACG-PM
0.6
0.7
0.8
0.9
FIG. 3. Nonparametric smooth response curves showing usual care (UC), guided care (GC)-received, and guided
care (GC)-assigned-but-not-received expenditures with different baseline ACG-PM scores.
Until this point, cost and utilization differences have been reported for all “targeted” patients; however, of the patients assigned to the
GC group, 31 actually received GC. Therefore,
we computed a second regression plateau
model to compare the expenditures for the GCreceived and the GC-assigned-but-not-received, and the UC group (Fig. 3). The GC-received group had lower expenditures than the
UC group at baseline risk levels of 0.10-0.40.
ITT is a standard method to evaluate methods from a randomized trial. We have provided
both ITT and per protocol results to report a
range of estimated magnitudes of cost differences. What we have observed in this study is
that the GC-assigned-but-not-received group
had the lowest expenditures at all risk levels
and increases the magnitude of cost savings in
the ITT analysis. Importantly, both figures
show lower costs for all GC patients throughout lower baseline risk levels.
DISCUSSION
Administrative costs of providing the GC pilot can be estimated from the additional resources required during the 6-month period.
These included salary and benefits for a full-
time nurse, travel and other expense reimbursement, database development, laptop
computer, printer, and office supplies. These
costs totaled approximately $70,000.
The dramatic increase in the number of older
adults and the prevalence of chronic disease
will contribute significantly to the future demand for health care resources and the rise in
health care expenditures in the United States.
More attention is being placed on the need to
improve the quality of chronic care and the efficiency of health care expenditures for people
with multiple chronic conditions. The Centers
for Medicare and Medicaid Services is attempting to address these challenges by conducting several national demonstration projects.26
The results of this pilot study suggest that
GC may reduce health care utilization and total insurance expenditures for high-risk older
adults. Although the small sample size urges
caution and reduces the precision of the estimated cost savings, the apparent trend toward
lower insurance expenditures for persons receiving GC is encouraging and warrants further investigation. The observed higher savings
at lower ACG-PM levels suggest the greatest
differences may be attained among older people at less extreme levels of risk for health care
GUIDED CARE OUTCOMES
utilization. This relationship also warrants further study.
This small pilot study has several significant
limitations. The time period for evaluation was
short, especially considering that aspects of the
intervention require a change in health behaviors on the part of the patient in which the desired outcomes usually take longer than 6
months to realize. Of the 62 high-risk patients
assigned to the GC group, 31 did not receive
GC. Assignment to GC or UC was not random,
and baseline differences that we were unable
to measure, such as the propensity of GC-assigned patients to participate in the intervention, could have affected the outcomes we observed. In addition, unmeasured differences in
the care provided by the GC and UC physicians
could have affected the results. Further, the
measurement of baseline characteristics and
the identification of high-risk patients occurred
several months before the study observation
period began, so changes in patients’ status
during this interval could have affected our results. Finally, we did not measure the costs of
providing GC in detail.
In summary, this pilot study suggests that
GC, a nurse-led, patient-centered, primary-carebased, comprehensive approach to chronic care
may reduce insurance expenditures for highrisk older adults. A multicenter, randomized
controlled trial of the complete model of GC
(including chronic disease self-management
and caregiver education and support) is now
addressing many of the limitations of this pilot study. This trial will provide more definitive information regarding the potential impact
of GC on the quality and cost of health care for
multimorbid older Americans.
ACKNOWLEDGMENT
Funding for the development and pilot-testing of Guided Care: Johns Hopkins HealthCare contributed the funding and administrative support for the guided care nurse, the
development of the EHR, the analysis of
claims data and supported Martha Sylvia’s
time. Johns Hopkins Community Physicians
contributed access to patients, the efforts of
the PCPs, and office space and equipment at
35
the community-based primary care practice.
The Lipitz Center for Integrated Health Care
contributed seed funding, administrative support, and support for data analysis. Dr. Boyd
was supported by the Johns Hopkins Bayview
Scholars at the Center for Innovative Medicine
at the Johns Hopkins Bayview Medical Center.
This abstract was presented in poster form at
the Academy Health Annual Research Meeting, Seattle, WA, 2006; HMO Research Network Annual Conference, Portland, Oregon,
2007.
REFERENCES
1. Weinstock M. Chronic care: an acute problem. Hosp
Health Netw. 2004;78:40-48, 2.
2. Berenson RA, Horvath J. The clinical characteristics
of Medicare beneficiaries and implications for
Medicare reform. Paper presented at: Medicare Coordinated Care Conference; March 21–22, 2002; Washington, DC.
3. Centers for Disease Control, National Center for
Chronic Disease Prevention and Health Promotion.
Chronic Disease Overview. 2006. Available at: www.
cdc.gov/nccdphp/overview.htm. Last accessed on
December 14, 2007.
4. Wolff JL, Starfield B, Anderson G. Prevalence, expenditures, and complications of multiple chronic
conditions in the elderly. Arch Intern Med. 2002;162:
2269-2276.
5. Anderson G, Horvath J. Chronic Conditions: Making the
Case for Ongoing Care. Baltimore, MD: Partnership for
Solutions; 2002. Available at: www.partnershipforsolutions.org/DMS/files/chronicbook2002.pdf.
Last accessed on December 14, 2007.
6. Gruenberg L, Tompkins C, Porell F. The health status
and utilization patterns of the elderly: implications for
setting Medicare payments to HMOs. Adv Health Econ
Health Serv Res. 1989;10:41-73.
7. Institute of Medicine. Crossing the Quality Chasm: A
New Health System for the 21st Century. Washington,
DC: National Academy Press; 2001.
8. McGlynn EA, Asch SM, Adams J, et al. The quality of
health care delivered to adults in the United States. N
Engl J Med. 2003;348:2635-2645.
9. Wagner EH, Austin BT, Davis C, Hindmarsh M,
Schaefer J, Bonomi A. Improving chronic illness care:
translating evidence into action. Health Aff (Millwood).
2001;20:64-78.
10. Boyd CM, Boult C, Shadmi E, et al. Guided care for
multi-morbid older adults. The Gerontologist. In press.
11. Weiner JP, Abrams C, Bodycombe D. The Johns Hopkins ACG Case-Mix System Version 6 Release Notes. Section 2. The ACG Predictive Model: Helping to Manage
36
12.
13.
14.
15.
16.
17.
18.
19.
SYLVIA ET AL.
Persons at Risk for High Future Costs. Baltimore, MD:
Johns Hopkins Bloomberg School of Public Health;
2003.
Cumming RB, Knutson D, Cameron BA, Derrick B. A
Comparative Analysis of Claims-based Methods of Health
Risk Assessment for Commercial Populations. Minneapolis, MN: Society of Actuaries; Park Nicollet Institute Health Research Center; 2002.
Dudley RA, Medlin CA, Hammann LB, et al. The best
of both worlds? Potential of hybrid prospective/concurrent risk adjustment. Med Care. 2003;41:56-69.
Fishman PA, Goodman MJ, Hornbrook MC, Meenan
RT, Bachman DJ, O’Keeffe Rosetti MC. Risk adjustment using automated ambulatory pharmacy data:
the RxRisk model. Med Care. 2003;41:84-99.
Hu G, Root M. Accuracy of prediction models in the
context of disease management. Dis Manag. 2005;8:4247.
Juncosa S, Bolibar B, Roset M, Tomas R. Performance
of an ambulatory casemix measurement system in primary care in Spain. European Journal of Public Health.
1999;9:27-35.
Liu CF, Sales AE, Sharp ND, et al. Case-mix adjusting performance measures in a veteran population:
pharmacy- and diagnosis-based approaches. Health
Serv Res. 2003;38:1319-1337.
Perkins AJ, Kroenke K, Unutzer J, et al. Common comorbidity scales were similar in their ability to predict health care costs and mortality. J Clin Epidemiol.
2004;57:1040-1048.
Pietz K, Ashton CM, McDonell M, Wray NP. Predicting healthcare costs in a population of veterans
affairs beneficiaries using diagnosis-based risk ad-
20.
21.
22.
23.
24.
25.
26.
justment and self-reported health status. Med Care.
2004;42:1027-1035.
Reid RJ, MacWilliam L, Verhulst L, Roos N, Atkinson
M. Performance of the ACG case-mix system in two
Canadian provinces. Med Care. 2001;39:86-99.
Wahls TL, Barnett MJ, Rosenthal GE. Predicting resource utilization in a Veterans Health Administration primary care population: comparison of methods
based on diagnoses and medications. Med Care.
2004;42:123-128.
Petersen LA, Pietz K, Woodard LD, Byrne M. Comparison of the predictive validity of diagnosis-based
risk adjusters for clinical outcomes. Med Care. 2005;43:
61-67.
Lachin JM. Statistical considerations in the intent-totreat principle. Control Clin Trials. 2000;21:167-189.
Hastie TJ, Tibshirani RJ. Generalized Additive Models.
News York: Chapman & Hall CRC; 1990.
Hansen MH, Kooperberg C. Spline adaptation in extended linear models. Statistical Science. 2002;17:2-20.
Wolff JL, Boult C. Moving beyond round pegs and
square holes: restructuring Medicare to improve
chronic care. Ann Intern Med. 2005;143:439-445.
Address reprint requests to:
Martha L. Sylvia, R.N., M.S.N., M.B.A.
Johns Hopkins HealthCare
6704 Curtis Court
Glen Burnie, MD 21060
E-mail: msylvia@jhhc.com