Case Study: Propensity Score Matching Design and Sensitivity Analysis for Trauma Care Evaluation

More young Americans die from traumatic injuries—such as car crashes, falls, or homicides—than from any other cause, which makes injury the leading cause of death among persons aged 1 to 44 years old. Trauma centers provide specialized medical services and resources to patients suffering from traumatic injuries. Hospitals in the United States can be categorized as Trauma Center (TC) and Non-Trauma Center (NTC), according to resources and expertise. Patients are admitted to TCs or NTCs depending on various factors, including severity of the injury, geographical restrictions, or other patient characteristics. This is by no means a random process, so the evaluation becomes an RWE study.

Using the 2006-2010 National Emergency Data Sample (NEDS) (Agency for Healthcare Research and Quality 2016), we evaluate the performance of the two levels of trauma care with respect to a key outcome, emergency department (ED) mortality. We consider trauma patients, aged 18-64, characterized by a severe trauma (injury severity score [ISS] >25). Detailed description of the data can be found elsewhere (Shi et al. 2016).

The exposure under investigation was the admission to NTCs (vs. TCs) and the binary outcome was ED mortality. The research question was "would the outcome of patients treated at a NTC be different if these patients had been treated at a TC?" The answer to this question has important implications to regional trauma care planning. TCs are supposed to offer the best care to trauma patients, with active research agenda and assuming a leader role in education, but they are also resource craving, which makes them less available in rural areas. A matching design is appropriate to evaluate this causal effect, which corresponds to the exposure effect on the exposed. Our original sample consisted of 21,855 patients, of whom 5314 (24.3%) and 16,541 (75.7%) patients were admitted to NTCs and TCs, respectively. For illustration purposes, we used NTC patients admitted in 2008 (1085) to ensure a large enough pool of available controls for matching design.