Controlling for the effects of hospital attributes: The efficacy of factor analysis

PURPOSE: The assessment of risk often requires controlling for the potentially confounding effects of hospital attributes. This paper (1) describes the use of factor analysis as a means of quantifying hospitalsʹ obstetrical care, and (2) compares within respective unconditional logistic regression models the performance of factor scores with that of the “obstetrical unit service level” (OBUSL) classification as defined by the American Hospital Association (AHA). METHODS: A principal components factor analysis was performed on fourteen variables from 116 hospitals in Upstate New York. These variables, descriptive of hospitalsʹ obstetrical care, were obtained from the 1992 AHAʹs Annual Survey of Hospitals. Factor scores were correlated with the OBUSL. Factor scores were matched to 89,341 women with vaginal deliveries in the 1992 Live Birth Registry for Upstate New York. The performance of factor scores and the OBUSL variable was compared in separate unconditional logistic regression analysis designed to identify determinants of obstetrical anesthesia care. RESULTS: Principal components factor analysis with varimax rotation identified three factors which were strongly correlated with hospital OBUSL. In a model which included the OBUSL variable, the adjusted odds ratios (AORs) or receiving an epidural for vaginal delivery were lower among mothers with Medicaid, HMO, or no insurance coverage (i.e., 0.45 [95% CI, 0.43–0.48], 0.68 [0.64–0.71], and 0.44 [0.38–0.52], respectively) than among those with private coverage. In a model in which factor scores were substituted for the OBUSL variable, respective AORs were 0.48 (0.45–0.52), 0.63 (0.60–0.66), and 0.45 (0.39–0.53). CONCLUSIONS: Factor analysis provided a parsimonious description of 14 hospital variables, was useful as a control within the regression model, and may prove similarly useful in other areas of clinical care.