Examples

advertisement
Practical Examples
Presenter disclosure information
Bradley G Hammill
Lesley H Curtis
Soko Setoguchi
Practical Examples
FINANCIAL DISCLOSURE:
None
UNLABELED/UNAPPROVED USES DISCLOSURE:
None
Example: Linked sample comparison
Representativeness of a National Heart Failure
Quality-of-Care Registry: Comparison of
OPTIMIZE-HF and Non-OPTIMIZE-HF Medicare
Patients
Lesley H. Curtis, Melissa A. Greiner, Bradley G. Hammill, Lisa
D. DiMartino, Alisa M. Shea, Adrian F. Hernandez and Gregg C.
Fonarow
Circ Cardiovasc Qual Outcomes 2009 2:377-384
Study objective
Objective: Compare patient characteristics and
health outcomes of Medicare beneficiaries enrolled
in OPTIMIZE-HF with those not enrolled who were
hospitalized for heart failure
Also, compare OPTIMIZE-HF hospitals to other
Medicare hospitals.
Analysis issues

Decisions to make
 Which records to include
 Comparisons of interest
 Comparison group selection
 Characteristics to compare
Which records to include

Patients potentially represented multiple times in
each database

Hospitalizations
 Take them all or one per patient?
 If one per patient, take first or random?
 Does it matter if patient has records in both
groups?
Comparison of interest

Within Medicare: OPTIMIZE-HF v. non-OPTIMIZEHF
 Among all sites?
 Among OPTIMIZE-HF sites? Define
participation period?

Within OPTIMIZE-HF: Medicare v. non-Medicare
 Among all sites?
 Among linked sites?
 Age restricted?
Possible comparisons
OPTIMIZE-HF
Linked
Medicare
OPTIMIZE
sites
<65y
Unlinked sites
Non-OPTIMIZE
sites
Comparison group selection

OPTIMIZE-HF = “New or worsening HF”

Medicare = ?
 HF diagnosis in any position on claim?
 HF primary diagnosis only? What if OPTIMIZEHF record is not primary?
Characteristics to compare

Within Medicare
 Require prior claims eligibility (12m)?
 Require follow-up period?
 Claims-based comorbidities? Outcomes?

Can we use OPTIMIZE-HF variables at all?
Study setup

Within Medicare comparison (all sites)

OPTIMIZE-HF / CMS-linked records
 Keep first per patient

Non-OPTIMIZE-HF records
 Eliminate OPTIMIZE-HF pts
 Take first hospitalization per patient in 2003-4
with primary diagnosis of HF

Compare claims-based comorbidities, mortality,
and readmission
Findings

Registry hospitals differed from non-registry
hospitals
 Higher volume, more cardiac services available,
more likely to be teaching hospitals

Patient demographic characteristics and comorbid
conditions were similar
Findings

Observed outcomes, registry v. non-registry



In-hospital mortality was not significantly different
(OPT=4.7% v Non-OPT=4.5%)
1-year mortality was slightly different
(OPT=37.2% v Non-OPT=35.7%)
1-year readmission was slightly different
(OPT=64.2% v Non-OPT=65.8%)
Example: Clinical effectiveness
Clinical Effectiveness of Implantable CardioverterDefibrillators Among Medicare Beneficiaries With
Heart Failure
Adrian F. Hernandez, Gregg C. Fonarow, Bradley G. Hammill,
Sana M. Al-Khatib, Clyde W. Yancy, Christopher M. O'Connor,
Kevin A. Schulman, Eric D. Peterson and Lesley H. Curtis
Circ Heart Fail 2010 3:7-13
Objective and analysis issues
Objective: Evaluate the long-term clinical
effectiveness of ICD therapy in older patients with
heart failure

Analysis issues
 Treatment and control group inclusion/exclusion
criteria
 Exposure definition
Inclusion/exclusion criteria

Indicated

Contraindicated

Include elective admits?

Age limit?
Exposure definition

Discharged with an ICD
 New only?
 Present at admission?

ICD planned after discharge
Study setup

Exclude contraindicated

Require EF  35%, exclude new onset HF

Exclude discharge to SNF, etc.

Exclude elective admits for lack of untreated
comparison group

Exclude very old for lack of treated comparison
group

New user design, exclude present at admission

Do not treat planned ICD as treated
Findings

Mortality was significantly lower among patients
who received an ICD compared with those who did
not
(38.1% v 52.3% at 3 years)

Adjusted hazard ratio of mortality over 3 years for
patients receiving an ICD was
0.71 (95% CI, 0.56 to 0.91)
Example: Clinical effectiveness
Improvements in long-term mortality after
myocardial infarction and increased use of
cardiovascular drugs after discharge: a 10-year
trend analysis
Soko Setoguchi, Robert J Glynn, Jerry Avorn, Murray A
Mittleman, Raisa Levin, Wolfgang C Winkelmayer
J Am Coll Cariolol. 2008 51:1255-7
Objective and analysis issues
Objective: Assess the relationship between
increasing use of cardiovascular medications and
trends in long-term prognosis after myocardial
infarction (MI) in the elderly

Design/analytic issues
 Defining ‘CV drug use’
 Start of follow-up
Avoid immortal person time bias
Potential explanations of improving survival over time
Potential Mediators of Changing Survival after MI
Trend in Characteristic of MI Patients
•Age, gender, and race
•Diagnosis of MI*
(Use and level of troponin for
diagnosis)
•Characteristics for MI*
(location, infarct size, affected
vessels)
•Complication of MI
•Comorbidity
Calendar
Year
Trend in In-hospital Management
•Thrombolytic therapy
•Antiplatelet agents and other drugs*
•Coronary angioplasty
•Surgery
Trend in Post-discharge Management
•Initiation and maintenance of drug
therapy
(aspirin*, BB, ACEI/ARB and statins)
•Life-style modification*
Effect on
Survival
Defining CV drug use

Started recommend meds during hospitalization

Filled prescription after discharge
 What timing?

Continued to take the medications for a certain
period
 What if some patients took it every day vs.
others skipped them once in a while?
Defining CV drug use

Dictate hypothesis clearly would help
 Increasing initiation of recommended CV meds
during acute hospitalization improved prognosis
in elderly patients after MI
 Increasing initiation of recommended CV meds
in outpatient setting ……
 Increasing ‘continued use’ of recommended CV
meds in outpatient setting ……..
Defining CV drug use

Things to consider in addition to choosing sound
hypothesis
 Availability of information
No inpatient drug use available
Aspirin use is not fully captured
 Sample size
Lose more patients as you assess drug use
over longer period
When to start the follow-up for an outcome?

Immortal person time bias
 Increasing initiation of recommended CV meds
during acute hospitalization improved prognosis
in elderly patients after MI
Immortal person-time bias

Comparing survival of responders vs. nonresponders to a chemotherapy

Usual method
 Categorize patients into responders vs. nonresponders based on tumor response
 Compare survival from the start of the treatment
 Length of survival affect the response
Anderson J Clin Onc 1983
Immortal person-time bias example

1st response evaluated at 2 months after
chemotherapy

All patients who died before the 1st evaluation
categorized as ‘non-responders’

Survival was from the time of chemo to 1 year.
 2 month ‘guarantee’ time for all responders
Anderson J Clin Onc 1983
Immortal person-time bias
Suissa PDS 2007
Landmark method (analysis)

Landmark Method (Analysis)
 ‘Select some fixed time after initiation of therapy
as a landmark for conducting analysis’
= starting follow-up after completion of
exposure assessment
 Limitations
Results may differ depending on which
landmark is chosen
Loss of power
Cannot observe the entire hazard function
Anderson J Clin Onc 1983
Study setup

All patients admitted to a hospital with MI (1995 -2004) using
algorithm previously shown to have high accuracy (PPV of
94%)

All study patients survived at least 30 days after discharge
from the index MI hospitalization

Long-term survival was observed from the 31st day after
discharge to the date of death

Assessed
 Trend in mortality
 Trend in CV drug use (filled prescription within 30 days
after discharge)
 Trend in PCI during MI hospitalization

Assessed contribution of increasing CV drug use by
sequentially including terms for the multivariate model
Time trends of treatment for MI
25
20
PCI
15
Surgery
10
Thrombolys
is
5
04
20
03
20
02
20
01
20
00
20
99
19
98
19
97
19
19
19
96
0
95
% Pateitns with Procedrues
30
Calendar Year
80
70
50
40
Statin
30
Beta blockers
20
ACEI/ARB
10
Non-aspirin
antiplatelet
Calendar Year
04
20
03
20
02
20
01
20
00
20
99
19
98
19
97
19
96
19
95
0
19
% Patients wtih Drugs
60
Of 21,484 MI patients, 12,142 died
during an average follow-up of
3.5 years.
A trend towards increasing age and
greater prevalence of
comorbidities such as
hypertension, peripheral vascular
diseases, cerebrovascular
diseases, diabetes, and chronic
kidney disease was observed
The use of percutaneous coronary
interventions increased over time,
whereas use of thrombolytic
therapy decreased (Top)
Use of all study drugs also increased
over time. (Bottom)
Potential explanations of improving survival over time
Potential Mediators of Changing Survival after MI
Trend in Characteristic of MI Patients
•Age, gender, and race
•Diagnosis of MI*
(Use and level of troponin for
diagnosis)
•Characteristics for MI*
(location, infarct size, affected
vessels)
•Complication of MI
•Comorbidity
Calendar
Year
Trend in In-hospital Management
•Thrombolytic therapy
•Antiplatelet agents and other drugs*
•Coronary angioplasty
•Surgery
Trend in Post-discharge Management
•Initiation and maintenance of drug
therapy
(aspirin*, BB, ACEI/ARB and statins)
•Life-style modification*
Effect on
Survival
Improving trend of long-term prognosis for MI
Hazar Ratio for Calendar Year
1.2
1.0
0.8
0.6
Not Adjusted for CV drug use/coronary intervention
0.4
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
Calendar Year
Improving trend of long-term prognosis for MI disappeared
after adjusting for the recommended cardiovascular drug
use
Hazar Ratio for Calendar Year
1.2
1.0
0.8
0.6
Not Adjusted for CV drug use/coronary intervention
Adjusted for CV drug use
0.4
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
Calendar Year
Use of CV procedures did not eliminate the calendar year
effect completely
Hazar Ratio for Calendar Year
1.2
1.0
0.8
0.6
Not Adjusted for CV drug use/coronary intervention
Adjusted for CV drug use
Adjusted for CV procedure use
0.4
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
Calendar Year
Lessons learned

The criteria for diagnosing MI have changed over the
decade studied
 likely resulting in an increasing fraction of patients
having non-ST elevation MI (NSTEMI).
 Unlikely to explain the findings completely.

No information on aspirin use and life style modification.
 Studies suggest that use of aspirin is relatively
stable after 1995
 unclear whether lifestyle has changed over time in
the elderly population

Further investigation is necessary to elucidate the
relative and individual contributions of these factors.
Download