Readmission Risk Prediction

Predicting the Probability of Readmission After Cancer Surgery with LSTM

Hospital readmissions cost the US healthcare system billions of dollars annually, are associated with high mortality rates, and are a source of stress and suffering for patients and family members. Traditional approaches to readmission risk stratification rely on static administrative and medical record data and generally classify all surgical oncology patients at high-risk. However, different factors related to daily behavior and activities may contribute to or signal increased or decreased risk of readmission. Our research utilizes mobile sensing and deep learning to measure daily readmission risk in cancer patients after discharge. Using data from mobile and Fitbit devices of 49 patients collected over 90 days after discharge from the hospital, we build a probabilistic model in an LSTM structure to infer the risk progression trajectory in each patient. Our results show that using only sensor data, the model can predict the risk progression trajectory aligned with the ground truth data.

Related Publications