A self-adaptive 30-day diabetic readmission prediction model based on incremental learning

Hospital readmissions within 30 days after discharge are costly and it has been a prior for researchers to identify patients at risk of early readmission. Most of the reported hospital readmission prediction models have been built with historical data and thus can outdate over time. In this work, a self-adaptive 30-day diabetic hospital readmission prediction model has been developed. A diabetic inpatient encounter data stream was used to train the self-adaptive models based on incremental learning algorithms. The result indicated that the model can automatically adapt to the newly arrived data. The best model achieved an average AUC of 0.655 ± 0.078, which is consistent with static models built with the same dataset.