In this project, we work on Gestational Diabetes Mellitus, a disease that results in increased blood glucose levels during the pregnancy of a mother. According to estimates of the International Diabetes Foundation, 18.4 million live births are affected by gestational diabetes in 2017 alone. Gestational diabetes is not only causing multiple negative outcomes for the mother and the newborn if not treated properly, such as high risk of type 2 diabetes (mother) or increased morbidity rates (newborn), but also represents a major financial burden for the health care system. With better ways of monitoring and diagnosing patients, the hope of this project is to treat gestational diabetes as early as possible so that such outcomes can be avoided.
In a healthy human body, blood glucose levels are regulated by insulin. If the blood glucose levels are high, for example due to food intake, the pancreas’ beta cells emit insulin which indirectly lower the blood glucose levels. However, if a patient suffers from gestational diabetes, the beta cells have a dysfunction, resulting in a disturbed blood glucose level regulation.
In this project, the goal was to monitor blood glucose levels with a metering device and use this data to predict model the risk of a patient on three tasks. First, we predict the risk of requiring drug treatment that counteracts the high blood glucose levels. Second, we predict the (bad) delivery outcomes itself to “early warn” the clinician about this risk. Third, we predict future blood glucose levels as an indirect way to make conclusions about therapy and diagnosis.
The data was gathered in multiple hospitals within the National Health Service in the UK.
More details will follow in a paper that we plan to submit to EMBC 2019.
In this project, we went from the stored, raw data tables in the medical system, cleaning and preprocessing the data, to clinically interesting (yet improvable) results in three tasks in less than 3 months. This progress would not have been possible without the expertise of my colleagues and collaborators, including Tingting Zhu, Carmelo Velardo, Prof. David Clifton, Prof. Lionel Tarassenko and Lucy McKillop at University of Oxford who were an incredible support in this project.