The authors developed a mixed-ensemble model to predict hospital readmission rates, which allows the tradeoff between reasoning transparency and predictive accuracy to be controlled.

Hospital readmission classifications are highly imbalanced because most patients are not readmitted to the same hospital or admitted to a different hospital within 30 days of being discharged. The model created by the authors provides a way to control the classification error for positive readmission instances. This ensembling method can be used effectively for predicting all-cause hospital readmissions for congestive heart failure (CHF) patients, without the limitations of individual consideration of classifiers and of traditional ensembling methods. This model is also more accurate in classifying positive readmission instances, especially when strong predictors of such readmissions are not available.

This paper highlights the first readmission model that addresses the conflict between predictive accuracy and reasoning transparency. The authors acknowledge the need for exploring a single model for readmission for all patients or different models for individuals to predict how readmission chances change over time for each patient. There are also suggestions to incorporate knowledge-based clinical information in the model’s structure, which supports evidence-based research in this arena and presents an exciting avenue of research in this field.