Diabetes in pregnancy is one such case in point. A potentially serious condition affecting a growing number of women, it requires careful monitoring in order to reduce the risk of maternal and neonatal complications (e.g. pre-term birth, caesarean section). Standard clinical management for the condition is lifestyle advice and pharmacological intervention when needed.
GDm-Health is a remote patient monitoring solution for pregnant women with diabetes, now used by over 50 NHS Trusts in the UK. Instead of maintaining a paper-based diary of blood glucose readings and associated meal information, reviewed periodically at routine clinic visits, GDm-Health allows ‘between clinic’ recognition of patterns in the data. Clinicians can intervene in a more timely manner, and women can achieve tighter control of their blood glucose readings, leading to improved clinical outcomes for the woman and her baby.
The underlying dataset offers the potential to improve clinical care still further. Comprising multiple variables from over 16,000 deliveries, including more than 2.8 million real-time blood glucose recordings, the data might be fundamental in helping clinicians identify earlier those women that need pharmacological intervention to effectively control blood glucose levels.
Machine learning offers the potential to quickly analyse large quantities of data that are now available as a result of using remote patient monitoring systems like GDm-Health, enabling the development of algorithms that can provide risk predictions to support clinical decision making. Predicting the need for pharmacological treatment could likely benefit pregnant women with diabetes by improving glycaemic control, potentially leading to improved perinatal outcomes and avoiding complications.
Through the analysis of 411,785 blood glucose readings from 3,029 de-identified and anonymised patient records, Sensyne developed a machine learning model that identifies when a woman needs to switch to medications (Insulin or Metformin) by analysing the data related to blood glucose and other risk factors.
Repeat testing showed the Sensyne algorithm (called SYNE-GDM) reliably predicted requirement for pharmacological therapy, out-performing against a heuristic currently used in clinical settings which simply relies on counting the number of alerts generated in the last three days. The next steps are to further evaluate this algorithm in real clinical settings.
This work is underway, with the aim of deploying the algorithm alongside the GDm-Health remote patient monitoring software to monitor real-time performance (i.e., predictions done on a daily basis on updated daily blood glucose readings) against decisions performed by healthcare professionals.
The research has led to the submission of SYNE-GDM for regulatory approval, as a machine learning algorithm for the early identification of the need for pharmacological intervention to supporting the delivery of clinical decisions quickly and efficiently to patients.
Our work also supports the hypothesis that machine learning algorithms, like SYNE-GDM, may in the future lead to improved clinical decision-making and support for patients across a wide range of healthcare conditions.
Senior Research Fellow, Sensyne Health
Read the full paper here