Methods

Using publicly available MA contract and enrolment files from the Centers for Medicare & Medicaid Services (CMS), we constructed a data set identifying all provider-offered contracts from 2011 to 2015 (n=2030 contract years). For the purposes of this study, provider-offered means that the parent organisation of the MA contract and provider organisation are or are owned by the same firm. To make this determination, we followed the approach of Frakt et al and Johnson et al.3 4 We used contract identifiers to link these data to yearly MA healthcare and drug coverage quality data. These quality data aggregate across plans within MA contracts, hence our analysis is at the contract-year level. Twenty per cent of contract-year observations were excluded, including those that were special needs plans or regional organisations.5 An additional 178 contract years were dropped because they were entirely missing quality data, our final analytic data set included 1446 unique contract years. However, not all contract-year observations had data on all quality measures we analyse (see online supplementary appendix table 1). Our data set begins with 70 measures. Of these, we excluded 24 from our analysis because they are missing data in more than half of our observations at the contract-year level.

To examine quality of care, for each measure of quality as the outcome variable, we estimated an ordinary least squares regression model with an indicator of provider-offered status as the key explanatory …