Greg Peterson, MPA, PhD candidate Melissa McCarthy, PhD Presentation for 2013 AcademyHealth Annual Research Meeting
Medicare Coordinated Care Demonstration (MCCD) Established in Balanced Budget Act of 1997 to test whether care management for chronically-ill Medicare beneficiaries could Improve quality of health services and health outcomes Lower Medicare expenditures 15 programs Unique populations/interventions Ran for 4 to 11 years (one ongoing) Randomized trials 2
Assess long-term effects of the MCCD programs on patient survival Although programs primary goal was to reduce hospitalizations, they could lengthen survival Filled gaps in self and clinical care for conditions that are lead causes of death No effects on two-year survival for any program a May be due to short follow-up and low statistical power Study s contribution: Study objective Longer follow-up for 13 programs (adding 2 to 4 years) Larger sample (from 14,069 to 20,601, treatment + control) a Peikes et al. 2009 3
The MCCD programs shared core features, but also varied Core features Care managers (typically nurses) assigned a caseload of patients Comprehensive assessments Self-management support Longitudinal monitoring Information exchange (patientsproviders, providers-providers) Variation Target population Monitoring frequency and mode Use of behavioral change models in self-management education Efforts to improve provider adherence to care guidelines Extent of transitional care 4
Separate analysis for each of 13 programs Compare survival in treatment vs. control groups (1:1) Sample: Enrollees through one year before program s end N ranged from 176 to 4,314 Intent to treat Outcome: Days from randomization to death or program end Follow-up period varied by program Mean: 2.5 to 5.1 years Max: 4 to 6 years Methods Data: Medicare Enrollment Database and claims (2000 2008) linked to randomization file 5
Statistical analysis Cox proportional hazard model hazard ratio Control variables, measured at baseline: Demographics (age, gender) Medical conditions (using Chronic Condition Warehouse) Medicaid enrollment Hierarchical Condition Category (HCC) score Tested proportionality assumption with Schoenfeld residuals 6
Enrollee characteristics Diagnosis, % Baseline characteristic (Percentage unless noted) All 13 programs Medicare average (2003) Coronary Artery Disease (CAD) 68 41 a Congestive Heart Failure (CHF) 54 41 a Diabetes 39 21 Chronic Obstructive 27 15 Pulmonary Disease Hierarchical Condition Category score, mean 2.2 1.0 Hospitalizations in prior year, mean 1.31 0.30 a Baseline data for program enrollees are not directly comparable to Medicare average because the Medicare data are reported as heart disease, which includes CAD and CHF 7
Wide variation in mortality rates reflects differences in targeting Target population (number of programs) Percent of control group enrollees who died within two years of enrollment Recently hospitalized for heart failure (4) 30 Varied conditions + recent hospitalization or other high-risk filter (6) 22 Varied conditions (2) 6 Cancer under active treatment (1) 44 8
Evidence of survival effects for only 2 of 13 programs, but effects large No detected effect a Number of programs Low statistical power b 4 Sufficient power c 6 Lengthened survival (hazard ratio < 1) 2 Inconclusive due to violation of proportionality assumption 1 Two programs with effects reduced hazard by 18 and 37 percent. a p-value for hazard ratio > 0.05 b Less than 80% power to detect a 25% or larger decrease in the hazard of death c Greater than 80% power n.a. = not applicable 9
Programs that lengthened survival closely monitored CHF patients Program (N)* Telemonitor Mean # contacts per month Hazard ratio (95% CI) U. of MD (176) Yes 3.9 0.63 (0.39, 1.00)* Avera (1,133) Yes 8.2 0.82 (0.69, 0.99)* Georgetown (232) Yes 5.9 0.76 (0.48, 1.21) CorSolutions (4,314) No 2.6 1.03 (0.94, 1.13) Suggests that programs with intensive monitoring for patients recently hospitalized for CHF can improve survival Consistent with a meta analysis a * P <= 0.05 a Inglis et al. 2011 10
Program with inconclusive result focused on end-of-life counseling Targeted patients with complex, advanced illnesses, but not in hospice (43% died within two years) Follow-up years Hazard ratio (95% CI) 1-3 0.90 (0.78, 1.03) 4-6 1.48 (1.08, 2.03)* Results may be due to Limitation in method: violation of proportional hazard assumption may bias results in later years of follow-up a End-of-life counseling: some patients may have chosen less intensive treatment, shortening survival time * P < 0.05 a Therneau and Grambsch, 2000 11
Possible reasons for lack of measured effects for most programs Low statistical power (4 programs) Implementation barriers (1 program) Difficulty hiring care managers; fewer than 60% of enrollees ever received intervention Low mortality risk (2 programs) Lack of anticipated effects on self management (all programs) No measured changes in patient diet, exercise, or medication adherence (self reported) a May be due to high adherence rates in control group a Brown et al. 2007 12
Summary The 13 MCCD programs in this study tested a range of care management models Most (10 of 13 programs) did not have measurable effects on survival Two programs targeting CHF patients with extensive monitoring improved survival Results inconclusive for one program focusing on endof-life counseling may have increased risk in later years of follow-up 13
Policy implications Heart failure programs focused on monitoring and responding to warning signs may improve survival among FFS Medicare beneficiaries Survival gains may increase costs Neither of the CHF programs that improved survival decreased all-cause hospitalizations One increased costs to Medicare by 17% a Importance of shared decision-making in end-of-life care a Peikes et al. 2009 14
References Brown, R., Peikes, D., Chen, A., Ng, J., Schore, J., & Soh, C. (2007). The Evaluation of the Medicare Coordinated Care Demonstration: Findings for the First Two Years. Princeton, NJ: Mathematica Policy Research. Inglis, S. C., Clark, R. A., McAlister, F. A., Ball, J., Lewinter, C., Cullington, D.,... Cleland, J. G. (2011). Structured telephone support or telemonitoring programmes for patients with chronic heart failure. Cochrane Database of Systematic Reviews (Online). Peikes, D., Chen, A., Schore, J., & Brown, R. (2009). Effects of care coordination on hospitalization, quality of care, and health care expenditures among Medicare beneficiaries: 15 randomized trials. JAMA, 301(6), 603-618. Therneau, T., and P. Grambsch. Modeling Survival Data: Extending the Cox Model. New York, NY: Springer, 2000. 15
Acknowledgements CMS: Bill Clark, Negussie Tilahun Mathematica: Debbie Peikes, Randy Brown, Carol Razafindrakoto Dissertation committee members: Paula Lantz, Jean Johnson, Donna Infeld Note: While I work for Mathematica Policy Research, the evaluator for the MCCD, I did this study as a graduate student through a data reuse agreement with CMS. 16