Filling the need for trusted information on national health issues

Do People Who Sign Up for Medicare Advantage Plans Have Lower Medicare Spending?

This analysis focuses on beneficiaries in traditional Medicare who were enrolled in both Medicare Part A and Medicare Part B in 2015, examining average adjusted 2015 Medicare Part A and B spending for these beneficiaries, based on their 2016 enrollment in Medicare Advantage plans or traditional Medicare. Beneficiaries who enrolled in Medicare Advantage plans at any point during the 2016 calendar year were categorized as Medicare Advantage enrollees.

To conduct this analysis, we excluded beneficiaries who: (1) became Medicare beneficiaries after 2013 or were not in traditional Medicare with both Part A and Part B in 2013, 2014, and 2015 (5.8 million people) because three years of claims data were required for each person to collect sufficient information about chronic conditions; (2) died prior to January 2016 (1.5 million people) because they would not have had the same opportunity to enroll in Medicare Advantage as other beneficiaries; (3) had end-stage renal disease in 2015 or 2016 (290 thousand people) because the vast majority were not eligible to enroll in a Medicare Advantage plan in 2016; (4) were unlikely to have actively selected (and instead may have been passively enrolled in) a Medicare Advantage plan, including beneficiaries who enrolled in Medicare-Medicaid Plans (MMPs) and employer group health plans (183 thousand people); (5) lived in Puerto Rico and other territories because some elements in the Medicare claims data are not as reliable or accurate for these beneficiaries; (6) enrolled in cost, Medical Savings Account (MSA), or PACE plans in 2016 (21 thousand people) because these plans are paid differently than Medicare Advantage plans; and (7) enrolled in a Special Needs Plan for people with specified chronic conditions (C-SNP; 13 thousand people) because the design of these plans may disproportionately attract healthier people with chronic conditions. When we relaxed the first inclusion requirement, for beneficiaries to be in traditional Medicare with both Part A and Part B in 2013, 2014, and 2015, and instead only required included beneficiaries to be in traditional Medicare with Part A and B in 2014 and 2015, the findings did not materially change, with the adjusted percent difference in spending remaining 13%. Similarly, when we included people in C-SNPs, the adjusted percent difference in spending did not change. In total, the primary analysis included 24 million beneficiaries who were in traditional Medicare in 2015.

The brief uses claims data from a five percent sample of Medicare beneficiaries from the Master Beneficiary Summary Files of CMS’s Chronic Conditions Data Warehouse for 2013 through 2016. The analysis first examined the bivariate differences in spending and use of services by demographics, chronic conditions, and other factors. To control for differences in health status and other factors that could account for the difference in Medicare spending, a multivariate generalized linear log link model with a gamma distribution was developed that mimics as closely as possible the CMS-HCC Risk Adjustment Model, which is used to risk-adjust payments to Medicare Advantage plans. The model for this study includes the same structure of the demographic variables and interaction terms as the HCC Risk Adjustment Model. This study’s model also includes the only available (although imperfect) variable to indicate whether someone who used a Part D covered drug was residing in a long-term care facility at any point during the year; this approach misses information about institutional residency status for the people who do not take drugs covered under Part D.

This study examined bivariate differences in traditional Medicare spending across counties, for those county residents who enrolled in Medicare Advantage compared to those who did not. The data used in the study did not include a sufficient number of people to adjust these county-level values for health risk factors. Future studies could examine whether the observed bivariate differences across counties hold, after adjusting for health risk factors.

The model used in this analysis does not include the HCCs in the Risk Adjustment Model that are not recorded as chronic conditions in the Chronic Conditions Data Warehouse, the majority of which are HCCs for acute or relatively rare conditions. The margins command, with values as observed, was used to generate the adjusted spending values. Alternative models for this analysis also included as covariates the per capita traditional Medicare spending for each county, beneficiaries’ race/ethnicity as defined by the RTI race variable, and additional chronic conditions, with no meaningful change in the results. We also looked at the sensitivity of the findings to the inclusion criteria; when we included beneficiaries who were in traditional Medicare with Part A and B in 2014 and 2015 but either were not in traditional Medicare or did not have both Part A and Part B in 2013, the findings did not materially change, with the risk adjusted difference in spending rising from $1,253 to $1,298.

Issue Brief Tables