Evaluating administrative compliance as a predictor of nursing home postdisaster outcomes in the USA

Introduction Determining whether compliance with the US Centers for Medicare we operationalised over 200 unique deficiencies as separate dichotomous indicator variables. We fit generalised linear models and applied elastic net regularisation to identify which regulatory deficiencies were most predictive of adverse postdisaster outcomes (30-day mortality (primary), 30-day hospitalisation and functional decline within 120 days). We selected the best fitting model for each outcome based on the lowest Bayesian Information Criterion and reported incidence rate ratios (IRRs) for retained variables. We performed 10-fold cross-validation and evaluated the predictive accuracy of the best-fitting models using root mean squared error (RMSE), compared with a null (intercept-only) model. Results Across 294 nursing homes with 21 945 residents with an average age of 81 years, there were 697 deaths, 1316 hospitalisations and 1274 instances of functional decline in the postdisaster period. No emergency preparedness deficiency predicted adverse postdisaster outcomes. In contrast, two building code deficiencies predicted postdisaster functional decline (IRRs 1.21 and 1.51). The best-fitting models demonstrated modest improvements in predictive accuracy compared with the null model for postdisaster mortality (RMSE 1.76 vs 1.79) and functional decline (RMSE 3.35 vs 3.44), although these differences were not statistically significant. Conclusions Measures of compliance with federal emergency-preparedness standards did not predict postdisaster mortality, hospitalisation or functional decline. These findings indicate a need to better align the measurement and oversight of nursing home emergency preparedness with the complexities of real-world disaster response.