Prediction of Hospital Readmission Combining Rule-based and Machine Learning Model
Kareen Teo, Khin Wee Lai, Ching Wai Yong, Belinda Pingguan‐Murphy, Joon Huang Chuah, Clarence Augustine TH Tee · 2020
Hospital readmission shortly after discharge is threatening to plague the quality of inpatient care. In the United States, the Hospital Readmissions Reduction Program has been established to improve patient outcomes by reducing payments to hospitals with excess readmissions. This concern calls for an incentive for healthcare facilities to reduce their readmission rates by predicting patients at a high risk of readmission. Recently, scientists have developed prediction models by using machine learning algorithms. Extensive feature engineering, testing multiple algorithms, and algorithm tuning were conventionally performed to better identify potential high-risk readmission. To alleviate these challenges, we develop a readmission risk ensemble prediction model that combines the data- and knowledge-driven feature scores in risk computation. The proposed model showed better prediction performance, i.e improvements of 2-17% in sensitivity and 1% in the area-under-the-receiver operating characteristic curve, proving that the combination of the machine learning and rule-based score can detect patients that are not evident on the prediction of the standard model alone.