Socioeconomic predictors of 30-day hospital readmission of elderly patients with initial discharge destination of home health care

Introduction: The purpose of this study was to examine the association of social determinants of health with 30-day rehospitalization among elderly patients who were discharged from an acute care hospital with an initial discharge destination of home with home health care support and to explore the social determinants' ability to accurately predict the odds of rehospitalization. Methods: A secondary data analysis was performed on data obtained from the home healthcare agency. A total of 4,717 unique patients found in the data set. Some patients were represented in the data more than once during the observation period. To examine the relationship between socioeconomic factors and 30-day rehospitalization while controlling for select clinical factors, bivariate and multivariable analyses were performed using generalized linear mixed models (Proc GLIMMIX) and Cox proportional hazard models (Proc PHReg). A backward variable selection procedure was used to determine the best predictive model and a stepwise forward procedure was used to develop the survival model. Results: Bivariate analyses of the variables used for predictive modeling showed that race (p = 0.0019), overall prognosis (p < .0001), overall status (p < .0001), multiple hospitali-zations (p < .0001), multiple medications (p = .0001), multiple falls (p = 0.0004), mental disorder (p < .0001), no risk of hospitalization (p = 0.002), no high risk factors (p = 0.03), and clinical classification (p < .0001) were significantly associated with 30-day rehospi-talization. Bivariate analyses of the variables used for survival modeling showed that race (p = 0.0029), living arrangement (p = 0.037), overall prognosis (p < .0001), overall status (p < .0001), multiple hospitalizations (p < .0001), multiple medications (p < .0001), multiple falls (p < .0001), mental disorder (p < .0001), no risk of hospitalization (p = 0.0017), no high risk factors (p = 0.02), smoking (p = 0.04), and clinical classification (p < .0001) were significantly associated with 30-day rehospitalization. The predictive model was developed on 80% of the data and a random sample of 20% was used for independent validation of the predictive algorithm, Model sensitivity was 47% and specificity was 78% with a C statistic of 0.6 in the validation data. Conclusion: While the variables in the final predictive model were statistically significantly associated with rehospitalization, the observed effect sizes were small to moderate and the model lacked sensitivity and was not very useful in correctly predicting which patients were rehospitalized