JultikaUniversity of Oulu repository

Abstract

Type 1 diabetes patients are prone to life-threatening conditions. Severe hypoglycemia and diabetic ketoacidosis are such conditions that often require urgent hospital care. Recently, artificial intelligence (AI) techniques have been used to improve the quality of diabetes care and management. These techniques provide a more comprehensive and better experience for patients and their loved ones. The objective of this study is to implement an AI-based explainable solution to predict possible severe hypoglycemia and diabetic ketoacidosis events in T1D patients within the next 12 months. The initial models in this study were built with baseline factors identified in prior research. However, baseline factors alone did not provide enough information, and the models were improved by introducing more features and separating the population by gender. The final predictive models highlighted some of the baseline factors in the original study when predicting the outcomes. Decision support systems based on machine learning models have become a viable way to enhance patient safety by locating and prioritizing high-risk patients. The final models were used to build a decision support system that facilitates precision medicine by prioritizing the high-risk patient group. Moreover, it helps to potentially reduce medical expenses through more efficient resource management.