624143 - Dynamic Forecasting and Control of Diabetes Using Mobile-Based Health-Lifestyle Data

Monday, May 20, 2019

2:40 PM – 3:00 PM

Location: Gatlin E1

Primary Contact(s)

Type 2 diabetes mellitus (T2DM) is a major health problem and is identified as one of the most common chronic diseases with approximately 16 million affected people in the United States alone. It not only creates a huge public health burden but also contributes to the tremendous patient burden of carrying out intensive and complex self-management regimens. Adherence to these complex regimens on day to day basis is particularly challenging. While an experience patient with a stable metabolism can attain necessary control of their glucose level, this is not the case for an unexperienced or elderly patient. Thus, we propose to use a machine learning approach to analyze streams of daily captured data from individual patients’ mobile health data in order to help patients and clinicians manage T2DM more efficiently. We develop a deep learning model based on dynamic recurrent neural networks to estimate and forecast the trajectory of glucose levels in individual patients. Given past lifestyle regimens data, the deep learning network monitors the patient daily regiments (diet, activity, weight control) and alarms any unusual trend/patterns that can lead to undesirable glucose level. The trend or patterns are monitored using a control chart tailored specifically for the patients.