Risk Modelling Framework for Emergency Hospital Readmission, Using Hospital Episode Statistics Inpatient Data

Abstract

The comparisons were made using positive predictive value, sensitivity and specificity for different risk cut-offs, risk bands and top risk segments, together with the receiver operating characteristic curve. The constructed Bayes Point Machine using this feature selection framework produces a risk probability for each admitted patient, and it was validated for different timeframes, sub-populations and cut-off points. At risk cut-off of 50%, the positive predictive value was 69.3% to 73.7%, the specificity was 88.0% to 88.9% and sensitivity was 44.5% to 46.3% across different timeframes. Also, the area under the receiver operating characteristic curve was 73.0% to 74.3%. The developed framework and model performed considerably better than existing modelling approaches with high precision and moderate sensitivity.

Ensemble Risk Model of Emergency Admissions (ERMER)

“…In addition, it improves the previous model (Mesgarpour et al, 2016a), which does not use the ensemble of subpopulations. The ROC and precision percentage of the any-acute model increase by 2.83 and 7.16, respectively, and sensitivity decreases in consequence.…”

“…There are three main layers of difficulties in the preparation of features: correlations, recategorisations and selections (Mihaylova et al, 2011;Walpole et al, 2014;Yang et al, 2005). In this study, the variables were generated and selected based on the previously developed preprocessing framework (Mesgarpour et al, 2016a). Based on this framework, a large pool of variables was generated and reduced based on a set of defined criteria.…”

“…Based on the framework developed in the previous stage of our research (Mesgarpour et al, 2016a), four steps of feature filtering were carried out, in order to reduce the number of features and to better capture the underlying structure.…”

“…In addition, similarly to the PARR model, each sample covers about 20% of unique patients within the trigger year of the selected time-frame (Table 1). Before the modelling stage, four stages of data preprocessing were carried out (Mesgarpour et al, 2016a). Firstly, the extracted data was sorted by patients and the order of episodes.…”

“…The HES inpatient data was extracted from English hospitals and maintained by the Health and Social Care Information Centre (HSCIC) (HSCIC, 2016c). Based on a preprocessing framework (Mesgarpour et al, 2016a), features were cleaned, generated, filtered and ranked.…”