Using Medicare data to measure vertical integration of hospitals and physicians

Researchers, healthcare providers, and policy makers have become increasingly interested in the cost and quality effects of vertical integration (VI) between hospitals and physicians. However, tracking VI is often financially costly. Because the Medicare Data on Provider Practice and Specialty (MD-PPAS) annual dataset may be more cost-effective for researchers to access than private data sources, we examine the accuracy of MD-PPAS in identifying VI by comparing it to physician and hospital affiliations reported in Blue Cross Blue Shield Texas (BCBSTX) PPO claims data for 2014-2016. The BCBSTX data serve as a gold standard, because physician-hospital affiliations are based on the insurer's provider contract information. We merged the two datasets using the physician National Provider Identifier (NPI), then determined what percentage of physicians had the same Tax Identification Number (TIN) in both sources, and whether the TIN implied the physician belonged to a physician- or hospital-owned practice. We found that 71.3% of successfully matched NPIs reported the same TIN, and 95.1% of patient-level observations were attributed to organizations with the same ownership type in both datasets, regardless of TIN. We compared regression estimates of patient-level annual spending on an indicator variable for physician versus hospital ownership for the primary attributed physician and found that estimates were within one percentage point whether one determined VI based on the BCBSTX or the MD-PPAS data. The results suggest that MD-PPAS, which costs less to obtain than from a for-profit data source, can be used to reliably track VI between hospitals and physicians.