Medicare managed care plan performance: a comparison across hospitalization types.

OBJECTIVE The study evaluates the performance of Medicare managed care (Medicare Advantage [MA]) Plans in comparison to Medicare fee-for-service (FFS) Plans in three states with historically high Medicare managed care penetration (New York, California, Florida), in terms of lowering the risks of preventable or ambulatory care sensitive conditions (ACSC) hospital admissions and providing increased referrals for admissions for specialty procedures. STUDY DESIGN/METHODS Using 2004 hospital discharge files from the Healthcare Cost and Utilization Project (HCUP-SID) of the Agency for Healthcare Research and Quality, ACSC admissions are compared with 'marker' admissions and 'referral-sensitive' admissions, using a multinomial logistic regression approach. The year 2004 represents a strategic time to test the impact of MA on preventable hospitalizations, because the HMOs dominated the market composition in that time period. FINDINGS MA enrollees in California experienced 22% lower relative risk (RRR= 0.78, p<0.01), those in Florida experienced 16% lower relative risk (RRR= 0.84, p<0.01), while those in New York experienced 9% lower relative risk (RRR=0.91, p<0.01) of preventable (versus marker) admissions compared to their FFS counterparts. MA enrollees in New York experienced 37% higher relative risk (RRR=1.37, p<0.01) and those in Florida had 41% higher relative risk (RRR=1.41, p<0.01)-while MA enrollees in California had 13% lower relative risk (RRR=0.87, p<0.01)-of referral-sensitive (versus marker) admissions compared to their FFS counterparts. CONCLUSION While MA plans were associated with reductions in preventable hospitalizations in all three states, the effects on referral-sensitive admissions varied, with California experiencing lower relative risk of referral-sensitive admissions for MA plan enrollees. The lower relative risk of preventable admissions for MA plan enrollees in New York and Florida became more pronounced after accounting for selection bias.