Readmission prediction for heart failure patients using features extracted from SS-MIX

Antonio Tejero-de-Pablos, Hiroaki Yamane, Yusuke Kurose, Tatsuya Harada

Abstract:

Increased demand for healthcare has resulted in necessity of utilizing electronic medical records. Standardized Structured Medical Information eXchange (SS-MIX) is a standard data storage format used in Japan to share clinical data. SS-MIX contains standard medical receipt data which is automatically stored. In this paper, we utilize the SS-MIX data to predict readmission of patients, who are suffering from heart failure. (i.e.,We define the readmission prediction task as a binary prediction task, where whether a patient readmits within N days after hospital discharge or not.) We train classifiers based on features (e.g., disease history, allergy, prescription, etc.) that are extracted from medical receipt data which is contained in SS-MIX. The experimental result showed that above features are effective for the task of readmission prediction; in particular, LightGBM outperformed other classifiers. Moreover, features from ICD10 codes are shown to be effective to predict readmission.