Predicting risk of hospitalisation or death: a retrospective population-based analysis.

Abstract

Develop predictive models using an administrative healthcare database that provide information for Patient-Centred Medical Homes to proactively identify patients at risk of hospitalisation for conditions that may be impacted through improved patient care. Retrospective healthcare utilisation analysis with multivariate logistic regression models. A population-based longitudinal database of residents served by the Emilia-Romagna, Italy, health service in the years 2004-2012 including demographic information and utilisation of health services by 3,726,380 people aged ≥18 years. Models designed to predict risk of hospitalisation or death in 2012 for problems that are potentially avoidable were developed and evaluated using the area under the receiver operating curve C-statistic, in terms of their sensitivity, specificity and positive predictive value, and for calibration to assess performance across levels of predicted risk. Among the 3,726,380 adult residents of Emilia-Romagna at the end of 2011, 449,163 (12.1%) were hospitalised in 2012; 4.2% were hospitalised for the selected conditions or died in 2012 (3.6% hospitalised, 1.3% died). The C-statistic for predicting 2012 outcomes was 0.856. The model was well calibrated across categories of predicted risk. For those patients in the highest predicted risk decile group, the average predicted risk was 23.9% and the actual prevalence of hospitalisation or death was 24.2%. We have developed a population-based model using a longitudinal administrative database that identifies the risk of hospitalisation for residents of the Emilia-Romagna region with a level of performance as high as, or higher than, similar models. The results of this model, along with profiles of patients identified as high risk are being provided to the physicians and other healthcare professionals associated with the Patient Centred Medical Homes to aid in planning for care management and interventions that may reduce their patients' likelihood of a preventable, high-cost hospitalisation.