The welfare effect of integration in professional services markets: Evidence from physician practices

In response to the growth of medical knowledge physicians have increasingly specialized (Becker and Murphy, 1992). Both the number of medical specialties and the number of MEDLINE-indexed publications have grown rapidly and roughly in parallel since the 1970s. Widespread emergence of richer organizational forms has not accompanied this growth, as many physicians still operate in small, single-specialty practices.

The welfare effect of provider integration -- specifically, combining primary care physicians (PCPs) and specialists into multi-specialty practices -- on welfare is an open question. Health services researchers have long argued that fragmentation among providers is a root cause of inefficiency in the US health care system. However, little evidence is available to support this view (Rebitzer and Votruba, 2011). Economists have found that integration solves a referral incentive problem at the cost of inducing moral hazard (Pauly, 1979; Garicano and Santos, 2004). In single-specialty primary care practices, specialist referrals lead to foregone revenue. In multi-specialty practice, revenue sharing reduces the PCP's opportunity cost of referral. However, revenue sharing also introduces a free-rider problem. PCPs may choose to refer excessively and PCPs and specialists may shirk in providing services.

My project aims to add new evidence to the scientific understanding of physician organizations. I link an extremely large data set of health insurance claims from the Health Care Cost Institute (HCCI) with a census of physicians and practices collected by SK&A. Jointly these data allow me to directly observe how physician choices vary with organizational form.

The HCCI claims data are drawn from four large insurers serving the US employer-sponsored market. The coverage is national and exceptionally broad, the data consisting of 6 billion claims for 40 million insured individuals. Patient diagnosis, services rendered, and the negotiated price for each of these services is recorded in each claim. The SK&A data are constructed from a census of office-based physicians and their practices. For each physician, SK&A records the practice to which the physician belongs as well as the physician's specialty. Because I observe each physician's practice, I can construct the size and specialty composition of all practices in the data. Using the National Provider Identifiers (NPIs) recorded in each data set, I am able to link each physician with his or her claims.

To estimate the effect of integration on referrals and treatment patterns, I first estimate a discrete choice model of physician behavior. The model captures essential features of the physician's clinical problem, i.e. whether to refer or what treatment to choose given the patient's condition. It also captures the essential economic tensions such as financial tradeoffs in treatment due to price differences and practice size. Once estimated, the model allows me to counterfactually simulate physician choices in different practice environments and, therefore, the effect of integration on health care costs and the quality of care (e.g. hospitalizations, Heathcare Effectiveness Data and Information Set measures, etc.).