Mohsen Mesgarpour1, Thierry Chaussalet2, Salma Chahed3. 1. HSCMG, Faculty of Science and Technology, University of Westminster, 115 New Cavendish Street, W1W 6UW London, UK. Electronic address: mohsen.mesgarpour@gmail.com. 2. HSCMG, Faculty of Science and Technology, University of Westminster, 115 New Cavendish Street, W1W 6UW London, UK. Electronic address: T.Chaussalet@westminster.ac.uk. 3. HSCMG, Faculty of Science and Technology, University of Westminster, 115 New Cavendish Street, W1W 6UW London, UK. Electronic address: S.Chahed@westminster.ac.uk.
Abstract
INTRODUCTION: About half of hospital readmissions can be avoided with preventive interventions. Developing decision support tools for identification of patients' emergency readmission risk is an important area of research. Because, it remains unclear how to design features and develop predictive models that can adjust continuously to a fast-changing healthcare system and population characteristics. The objective of this study was to develop a generic ensemble Bayesian risk model of emergency readmission. METHODS: We produced a decision support tool that predicts risk of emergency readmission using England's Hospital Episode Statistics inpatient database. Firstly, we used a framework to develop an optimal set of features. Then, a combination of Bayes Point Machine (BPM) models for different cohorts was considered to create an optimised ensemble model, which is stronger than the individual generative and non-linear classifications. The developed Ensemble Risk Model of Emergency Admissions (ERMER) was trained and tested using three time-frames: 1999-2004, 2000-05 and 2004-09, each of which includes about 20% of patients in England during the trigger year. RESULTS: Comparisons are made for different time-frames, sub-populations, risk cut-offs, risk bands and top risk segments. The precision was 71.6-73.9%, the specificity was 88.3-91.7% and the sensitivity was 42.1-49.2% across different time-frames. Moreover, the Area Under the Curve was 75.9-77.1%. CONCLUSIONS: The decision support tool performed considerably better than the previous modelling approaches, and it was robust and stable with high precision. Moreover, the framework and the Bayesian model allow the model to continuously adjust it to new significant features, different population characteristics and changes in the system.
INTRODUCTION: About half of hospital readmissions can be avoided with preventive interventions. Developing decision support tools for identification of patients' emergency readmission risk is an important area of research. Because, it remains unclear how to design features and develop predictive models that can adjust continuously to a fast-changing healthcare system and population characteristics. The objective of this study was to develop a generic ensemble Bayesian risk model of emergency readmission. METHODS: We produced a decision support tool that predicts risk of emergency readmission using England's Hospital Episode Statistics inpatient database. Firstly, we used a framework to develop an optimal set of features. Then, a combination of Bayes Point Machine (BPM) models for different cohorts was considered to create an optimised ensemble model, which is stronger than the individual generative and non-linear classifications. The developed Ensemble Risk Model of Emergency Admissions (ERMER) was trained and tested using three time-frames: 1999-2004, 2000-05 and 2004-09, each of which includes about 20% of patients in England during the trigger year. RESULTS: Comparisons are made for different time-frames, sub-populations, risk cut-offs, risk bands and top risk segments. The precision was 71.6-73.9%, the specificity was 88.3-91.7% and the sensitivity was 42.1-49.2% across different time-frames. Moreover, the Area Under the Curve was 75.9-77.1%. CONCLUSIONS: The decision support tool performed considerably better than the previous modelling approaches, and it was robust and stable with high precision. Moreover, the framework and the Bayesian model allow the model to continuously adjust it to new significant features, different population characteristics and changes in the system.