TPGraph: a hospital readmission prediction method based on temporal phenotype graphs

Accurate hospital readmission prediction in a vast amount of healthcare data is important to the reducing healthcare costs and improving treatment patterns. Due to the temporality and sequentiality of the medical records, we propose a method for predicting hospital readmission based on temporal phenotype graphs in this paper, namely the TPGraph. Firstly, we constructed a temporal graph for each patient based on their medical event sequence. Then, we developed an approach to identify the most significant frequent subgraphs as temporal phenotype graphs. After that, an improved greedy algorithm was designed to obtain the optimal expression coefficient of temporal phenotype graphs. Finally, the optimal expression coefficient as a feature, we use random forest algorithm to predict whether the patient will perform hospital readmission. Our experiments demonstrate the effectiveness of our proposed method, and show that our approach to gain better predictive performance compared with the baselines.