Title: The Relationship Between CMS Quality Indicators and Long-term Outcomes Among Hospitalized Heart Failure Patients Mark Patterson, Ph.D., M.P.H. Post-doctoral Fellow Duke Clinical Research Institute (DCRI)
1The Relationship Between CMS Quality Indicators
and Long-term Outcomes Among Hospitalized Heart
Failure PatientsMark Patterson, Ph.D.,
M.P.H.Post-doctoral FellowDuke Clinical
Research Institute (DCRI)
2Acknowledgements
- Duke Clinical Research Institute (DCRI)
- Lesley Curtis, Ph.D.
- Adrian Hernandez, M.D.
- Bradley Hammill, M.S.
- Kevin Schulman, M.D.
- Eric Peterson, M.D.
- UCLA Medical Center
- Gregg Fonarow, M.D.
- Funding Sources
- Contract with GlaxoSmithKline
- Duke CERTs grant (AHRQ grant U18HS10548)
3Pay-for-Performance and Process Measures
- Goal of Pay-for-Performance Encourage providers
to follow recommended clinical care by providing
financial incentives - Theory Financial incentives ? improve
providers adherence ? improve clinical outcomes - Process Measures Estimate provider-level
adherence to this recommended clinical care
4CMS Heart Failure Process Measures
- Improving heart failure care remains a priority
for CMS - Prevalence 5 million Cost 30 billion
- 4 Core Process Measures
- Providing discharge instructions
- Conducting left ventricular ejection fraction
(LVEF) assessment - Prescribing ACE inhibitors or angiotensin
receptor blockers at discharge - Providing smoking cessation counseling
5Associations between process measures (PM) and
mortality
- Mixed evidence in regards to the associations
between process measures and mortality - Acute coronary syndrome1
- AMI2
- Heart failure3
- No evidence in regards to associations between PM
and long-term mortality
1 Peterson et al., JAMA, 2006 2. Bradley et al.,
JAMA, 2006 3. Fonarow et al., JAMA, 2007
6Objective
- Measure associations between the 4 current CMS
heart-failure process measures and 1-year
mortality - H1 Hospital-level process measures will be
associated with patient-level mortality
7Data Sources
- Retrospective cohort study
- Matched HF patients within the OPTIMIZE registry
with their Medicare Part A claims (2003 2004 - OPTIMIZE-HF
- Medicare Part A
- CMS denominator files
- Matched on age, gender, discharge date, and
hospital
8Participants
- Medicare fee-for-service HF patients matched to
the OPTIMIZE-HF registry (N22,483) - Excluding patients who died before discharge
- Excluding hospitals with
- missing process measures
- with less than 25 patients
- Final analytic dataset (N22,451)
9Hospital-level single process measures (PM)
- Discharge instructions N15,142 (67)
- LVEF assessment N20,061 (89)
- ACEI or ARBs at discharge N5,457 (24)
- Smoking cessation at discharge N902 (4)
- Frequency of PM documentation
- --------------------------------------------------
----------- - Number of patients eligible to receive PM
10Hospital-level combined process measures
- Composite N22,451
- Total number of processes documented
- --------------------------------------------------
---------- - Total number of opportunities to perform
- Defect-free N22,451
- Proportion of patients within the hospital having
documentation for ALL the PM that they were
eligible to receive
11Outcome and Control Variables
- Patient-level Mortality
- CMS denominator file
- Patient-level controls
- Demographics
- Comorbities
- Clinical measures
- Creatinine, weight, blood pressure
- Hospital-level volume
- Total HF discharges
- HF discharges of total
12Statistical Analysis
- Cox multivariate regressions
- Controlling for demographics, clinical measures,
selected co-morbidities, and hospital volume
indicators - Accounting for clustering of patients within
hospitals - 6 final models
- 4 Models for each single PM
- 2 Models for each combined PM
13Selected Baseline Characteristics (N22,451)
Variable
Mean age (s.d) 79 (7.8)
Male 44
White 84
Black 10
Other 6
Prior AMI 23
Prior PVD 15
Prior Hyperlipidemia 33
Mean Serum Creatinine (mg/dL) (s.d) 1.6 (1.2)
Mean Systolic BP (mmHg) (s.d) 142 (32)
Mean Weight (kg) (s.d) 77.4 (20.8)
14Hospital PM Adherence Rates (N178)
Process Measure (PM) Mean Score S.D. Range
Single
Discharge Instructions 0.52 0.30 (0 1.0)
LVEF Assessment 0.87 0.12 (0.29 1.0)
ACEI / ARBs at Discharge 0.75 0.16 (0.25 1.0)
Smoking Cessation 0.57 0.35 (0 1.0)
Combined
Composite 0.72 0.15 (0.32 1.0)
Defect-free 0.54 0.22 (0 1.0)
15Associations between hospital-level process
measures and patient mortality
HR (95 CI) HR (95 CI)
Process Measure (PM) N Unadjusted Adjusted
Single
Discharge Instructions 15,142 1.0 (0.99 1.02) 0.99 (0.98 1.01)
LVEF Assessment 20,061 1.0 (0.96 1.04) 1.0 (0.96 1.03)
ACEI / ARBs at Discharge 5,457 0.94 (0.89 0.99) 0.97 (0.93 1.02)
Smoking Cessation 902 0.99 (0.96 1.03) 0.98 (0.93 1.04)
Combined
Composite 22,451 1.0 (0.99 1.03) 1.0 (0.98 1.01)
Defect-free 22,451 1.0 (0.99 1.03) 1.0 (0.99 1.01)
16Discussion
- Current CMS heart failure process measures (PM)
are not associated with 1-year mortality in
Medicare beneficiaries diagnosed with HF - Explanation for null findings
- Care given at discharge may not affect 1-year
mortality - Documentation of care does not capture the
intensity or accuracy of care - High variation for PM may prevent ability to
detect small changes if they exist
17Limitations
- Cross-sectional design
- Unobserved factors confounding associations
- Patient-level
- Hospital-level
- Documentation of process measure at discharge may
not reflect the care given over 1 year
18Strengths
- First known study to link clinical registry data
with CMS data to examine associations between
process measures and long-term outcomes - Generalizeable to Medicare fee-for-service heart
failure patients1 - Models
- Include both patient and hospital-level
covariates - Account for clustering
1 Curtis et al., Abstract Proceedings at AHA,
2007
19Conclusions Recommendations
- Null findings do not undermine the need to
continue providing care that is good clinical
practice - Need to more firmly establish link between PM and
outcomes before broadly implementing P4P - Improve the accuracy of the measures
- Continue evaluating the effects of PM
- Within the context of longitudinal data
- Using PM with known clinical efficacy