Identification and Evaluation of Informal Inter-Organizational Ties between Hospitals and Skilled Nursing Facilities
Discussant: Jia Yu
We used the Random Utility Maximization (RUM) model developed by McFadden to test shared physician personnel as a marker of hospital-SNF linkage. Specifically, we estimated whether patients were more likely to enter SNFs, among the potential set of SNFs patients may be discharged to, sharing physicians with the discharging hospital. The presence of shared physicians was derived from Medicare inpatient and carrier claims and was lagged by one year to reduce endogeneity. We controlled for numerous SNF attributes such as quality ratings, staffing levels, bed availability, existing vertical integration, and distances between SNFs to individuals’ residences. Given that our model incorporates individual fixed effects we could not control for observable patient- or hospital-level characteristics directly. Accordingly, we conducted numerous subgroup analyses (e.g. among patients eligible for Medicaid, admitted to an ACO-affiliated hospital, and undergoing joint replacement surgery) to ensure robustness.
We estimated differences in outcomes between patients admitted to SNFs with and without shared physicians via OLS and through an instrumental variables approach. Our instrument, developed from our McFadden model, was the probability of entering a SNF with shared physicians given bed availability of all nearby SNFs on the day of hospital discharge. Outcomes were the number of days spent in the following settings during the 180-day post-SNF admission period: acute hospital, SNF, community without home health (HH) care, community with HH, and death. Models were adjusted for individual-, nursing home- and zip code-level characteristics and included county-hospital fixed effects.
We found that the probability of admission to a SNF increased 2- to 3-fold when it shared physician personnel with the discharging hospital, and was comparable to the effect of a SNF being vertically integrated with the discharging hospital. We found no noticeably different finding in any subgroup analysis. The number of days spent in the hospital following SNF admission did not differ significantly between individuals attending SNFs with and without shared physicians. However, patients admitted to SNFs with shared physicians on average spent 1.2 fewer days in the SNF, 1.5 more days in the community with HH, 0.3 days more days in the community without HH, and 0.8 fewer days dead. In summary, we provide evidence that shared physician personnel signal an informal hospital-SNF linkage, and find that SNFs informally tied to hospitals produce better clinical and health service utilization outcomes.