Machine learning and translational approaches to personalised care for women with gestational diabetes

点击翻译按钮获取中文摘要

Gestational diabetes mellitus (GDM), glucose intolerance with first onset or recognition during pregnancy, has a global prevalence of 14.0%, making it one of the most common disorders of pregnancy. With increasing obesity and maternal age worldwide, the incidence of GDM is likely to increase, causing a strain on health systems.Gestational diabetes can have adverse short-term consequences on both the mother and the fetus. Women with GDM are at a higher risk of experiencing pregnancy complications and adverse neonatal outcomes such as caesarian section (C-section), preterm delivery, macrosomia, large for gestational age babies, neonatal respiratory distress syndrome, neonatal jaundice, and admission to a neonatal ICU. Although GDM usually subsides after birth, it may lead to long-term consequences for both the mother and the child. The odds of developing type 2 diabetes (T2DM) are substantially higher for women with GDM than without. Additionally, women with GDM are at a significantly higher risk of hypertension, obesity, and cardiovascular morbidity. Children exposed to GDM also have a higher risk of obesity, cardiovascular morbidity and glucose intolerance compared to unexposed children.While appropriate detection, treatment and follow-up reduce the complications associated with GDM, they also place a significant burden on overwhelmed health systems as it is unclear which women are at risk. Risk stratification is important to effectively allocate resources and direct secondary prevention towards at risk populations, especially with the increasing prevalence of GDM. While models have been developed to identify which women are at risk of developing GDM, few have focused on identifying which women with GDM are at risk of experiencing short or long-term adverse outcomes. Data-driven machine learning models have the potential to personalise GDM management and treatment, allowing clinicians to move away from a one-size-fits-all approach. This project intends to leverage the power of large datasets and machine learning methods to develop prediction models capable of stratifying women with GDM based on their risk of developing short and long-term adverse outcomes of GDM. The overarching aim of the project is to develop and validate data-driven models to stratify women with GDM based on risk of adverse outcomes during pregnancy, at birth, and after pregnancy using the GDm-health tagged glucose dataset and electronic health records. Prior to developing risk stratification models, our first step is to conduct a systematic literature review. The search strategy will be guided by the main objective to understand the current models available to predict poor individual outcomes in women with GDM. The next step would be to consolidate the clinical datasets and explore their limitations. Following the extraction and selection of features and outcomes, we will explore different models ranging from classical statistical models to more elaborate machine learning methods. In parallel, I will work with the regulatory, legal, and business development teams at EMIS Health (industry partner) to learn about how to translate a product developed in an academic setting to commercial markets.