The effect of integration among generalist and specialist physicians on medical care

Despite longstanding interest among researchers and policymakers, little is known about the effects of organizational form in medicine (e.g. see Rebitzer and Votruba, 2011). Lack of evidence has not, however, precluded policies aimed at encouraging integration among providers, such as the Accountable Care Organization incentives in the Affordable Care Act. In this project I provide new evidence on the effect of integration among physicians -- specifically generalists (e.g. family practitioners) and specialists (e.g. cardiologists) -- on outcomes such as spending, physician effort, outpatient utilization, and new prescription drug usage. 

My analysis joins two important sources of data in health economics:

an extremely large and detailed data set of health insurance claims from the Health Care Cost Institute and a marketing database from SK&A that captures organizational form for most office-based physicians in the United States. Because I do not observe severity in the data, I score patients using a well-known and actively maintained risk adjustment application, the Johns Hopkins ACG System.

In my descriptive analysis, I compare spending and utilization for patients of integrated generalists against those for patients of non-integrated generalists. Which specialists define an integrated firm depends on the medical condition. Patients in my data are observed with at least one of three conditions: acute sinusitis, hypertension, and irritable bowel syndrome (IBS). For acute sinusitis patients, a generalist is integrated if an ear, nose, and throat specialist is in his firm. For IBS patients, I look for gastroenterologists. (For convenience, I call a patient initially treated by a generalist an ``integrated patient.'')

Differences in the descriptive analysis are striking. Integrated patients spend more on physician services than non-integrated patients, consume less physician services RVUs, and receive new prescriptions more frequently. Interestingly, in the case of outpatient services, integrated acute sinusitis and IBS patients consume less of such care while hypertension patients consume more of such care. Linear models estimated with ordinary least squares yield an even richer set of patterns. Among hypertensive patients, integration is again associated with greater physician services spending. However, the association holds only because integrated firms generate higher spending only for higher-severity patients. In contrast, while organizational form is associated with the probability of a specialist follow-up, patient severity matters little.

Because physicians observe severity in ways I do not and I observe severity-based sorting by patients in the data, OLS will not identify the effects of integration. Therefore I generate differential distance-type instruments a la McClellan, McNeil, and Newhouse (1994).

These instruments are clearly appropriate in my setting. If distance affects provider choice in a high-stakes situation, it will certainly matter when care is routine. Factors like search costs make it unlikely that these distances affect patients' initial housing choices.