Factors predicting the hospital episode costs of emergency medical admissions

Abstract

BackgroundImportant outcome predictor variables for emergency medical admissions are the Manchester Triage Category, Acute Illness Severity, Chronic Disabling Disease and Sepsis Status. We have examined whether these are also predictors of hospital episode costs. MethodsAll patients admitted as medical emergencies between January 2008 and December 2012 were studied. Costs per case were adjusted by reference to the relative cost weight of each diagnosis related group (DRG) but included all pay costs, non-pay costs and infra-structural costs. We used a multi-variate logistic regression with generalized estimating equations (GEE), adjusted for correlated observations, to model the prediction of outcome (30-day in-hospital mortality) and hospital costs above or below the median. We used quantile regression to model total episode cost prediction over the predictor distribution (quantiles 0.25, 0.5 and 0.75). ResultsThe multivariate model, using the above predictor variables, was highly predictive of an in-hospital death-AUROC of 0.91 (95% CI: 0.90, 0.92). Variables predicting outcome similarly predicted hospital episode cost; however predicting costs above or below the median yielded a lower AUROC of 0.73 (95% CI: 0.73, 0.74). Quantile regression analysis showed that hospital episode costs increased disproportionately over the predictor distribution; ordinary regression estimates of hospital episode costs over estimated the costs for low risk and under estimated those for high-risk patients. ConclusionPredictors of outcome also predict costs for emergency medical admissions; however, due to costing data heteroskedasticity and the non-linear relationship between dependant and predictor variables, the hospital episode costs are not as easy to predict based on presentation status.