143
Hospital Connectivity as a Measure of Unobserved Disease Severity Among COPD Patients

Tuesday, June 14, 2016
Lobby (Annenberg Center)

Author(s): Jacob E Simmering; Linnea A. Polgreen; Philip M Polgreen

Discussant: Cole G Chapman

Introduction
Adjustment for mortality risk in observational data is an active area of research. Both academic studies and programs focusing on defining "excessive" rates for Medicare depend heavily on risk adjustment. However, adjustment on the observable patient variables, such as sex, age, payer, and comorbidities may leave significant unobserved heterogeneity between the patients and their risk of inpatient mortality. We propose that including a measure of the number of transfers into a hospital may capture some of this unobserved heterogeneity in patient severity.

Methods
Using the Healthcare Costs and Utilization Project State Inpatient Database for California for years 2005 to 2011, we extracted all cases of primary hospitalization due to chronic obstructive pulmonary disease (COPD). In addition, we counted all cases of transfer between one hospital and another in the data and the number of hospital discharges. We regressed the patients' inpatient mortality on the patients' age, sex, primary payer, race, year of admission, month of admission, the Elixhauser comorbidities, the number of all-cause admissions to the hospital in the same year (by decile) and the number of transfers into the hospital in that year (by decile).

Results
Relative to patients in hospitals with the fewest transfers (0-25), patients in hospitals with the second decile (26-56) had odds of death 1.17 times greater (p = 0.0020). Patients in hospitals in the fifth decile (132-179) had odds 1.66 (p < 0.0001) greater. Patients in the highest decile (712+) had odds 1.98 (p < 0.0001) greater.

Conclusions
Patients in hospitals with a greater number of incoming transfers have a higher risk of inpatient mortality after adjustment for age, sex, payer, timing of admission and a common adjustment for patient severity. This risk persists after adjustment for hospital size. We theorize that the greater risk associated with hospitals with higher rates of inpatient transfers is due to selection by high severity patients into these facilities. Network connectivity contains information about the severity of patients in a facility even after adjustment for commonly-used variables. This information may improve the quality of models adjusting for medically severe patients.