L-HRP: A Remote Health Risk Prediction model based on LSTM for elderly

With the continuous progress of the aging process, the prevalence of chronic diseases and disability among the elderly increases, resulting the corresponding medical demand and medical costs create an intensified pressure on the healthcare infrastructures. To this end, this topic proposes a remote health risk prediction model for the community- and home-based scenarios based on wearable sensor technology and deep learning technology to alleviate the pressure on healthcare infrastructure. The model is designed in three functional considerations. The wearable sensor component is responsible for remotely and continuously collecting vital signs of the elderly, including five variables: systolic blood pressure (SBP), heart rate (HR), respiratory rate (RR), temperature (TEMP), and oxygen saturation (SPO2). The neural network component consists of five different 5- input-1-output Long Short-Term Memory (LSTM) networks, which are responsible for predicting values of the vital signs. The risk prediction component consists of a simplified version of the National Early Warning Score (NEWS), which is responsible for predicting the health risk level based on the predicted values of vital signs. The model is developed and tested using existing electronic health record (EHR) that mimic vital signs data collected via wireless sensor network. We found that the model performed the best using a dataset in a size of 1000 admissions, within a time window of 12 hours, and with a configuration of 5 neurons and 50 epochs. The accuracy was over 74% for the risk level calculated from the predicted values of vital signs. Our results suggest that remote monitoring and prediction of health risks in the elderly using deep learning models is a feasible new strategy for community- and home-based monitoring systems.