Type 2 Diabetes Mellitus (T2DM) is a chronic, progressive metabolic disorder, associated with an increased risk of developing micro- and macrovascular complications. Because of its interactive and heterogeneous nature, the management of T2DM is very complex. For the successful management of T2DM, the use of individualized and evidence-based clinical guidelines is necessary. Randomized controlled trials (RCTs) are considered the gold standard for clinical research. However, the results from RCTs can be inconclusive, leaving many aspects of T2DM management unaddressed. Therefore, there exists a huge gap between the optimal individualized and the current patient care. To fill some of the gap, there are opportunities of artificial intelligence (AI) in medicine, because big data and advanced machine learning (ML) techniques offer a new way to generate evidence that enhances clinical practice guidelines with more personalized recommendations. My overarching goal is to build clinically useful and transferable machine learning models on big data that can influence individual T2DM patient care towards the implementation of precision medicine. Under this goal, I had three specific aims, which I successfully achieved. • Specific aim 1: To develop a semi-supervised divisive hierarchical clustering algorithm for a subpopulation-based T2DM risk score. • Specific aim 2: To develop a Multi-Task Learning (MTL)-based methodology to reveal outcome-specific effects by separating the overall deterioration of metabolic health from progression to individual complications. • Specific aim 3: To demonstrate that even a complex ML model built on nationally representative data can be transferred to two local health systems without significant loss of predictive performance. In the management of T2DM, which is complex, the availability of reliable clinical evidence is critical for clinicians to make the right decision and produce high-quality care in healthcare delivery. Against the backdrop of RCTs, AI in medicine can reduce the gap between optimal individualized and current T2DM patient care. And building clinically useful and transferable ML models will especially facilitate the implementation of precision medicine in T2DM.
University of Minnesota Ph.D. dissertation. February 2019. Major: Health Informatics. Advisor: Gyorgy Simon. 1 computer file (PDF); xii, 124 pages.
The Use of Artificial Intelligence for Precision Medicine in the Metabolic Syndrome.
Retrieved from the University of Minnesota Digital Conservancy,
Content distributed via the University of Minnesota's Digital Conservancy may be subject to additional license and use restrictions applied by the depositor.