| Popis: |
BackgroundUndiagnosed type 2 diabetes continues to represent a significant challenge for all national healthcare systems. Although diagnostic criteria and laboratory screening procedures are well-established, clinical tests have limitations, and in many cases, diagnosis confirmation and more accurate interpretation of the test results are required. Machine learning methods, when applied to clinical outcome risk prediction, demonstrate great effectiveness, as they recognize specific patterns in data dynamics and thus can be used for the identification of at-risk cases where diabetes and complications can be delayed or even prevented. The aim of this study was to develop a type 2 diabetes machine learning model capable of efficient early identification of diabetes presence based on the results of common laboratory tests.MethodsReal-world medical data from electronic medical records were subjected to a multistage processing, including feature selection, missing values imputation. The machine learning algorithms adopted in this study were XGBoost, multilayer perceptron, ridge classifier, ridge classifier with polynomial features, bootstrap aggregating, dynamic ensemble selection, stacked generalization. An external dataset was analyzed via the same workflow to validate the initial results. The study was designed in accordance with the TRIPOD statement.ResultsWe have developed a machine learning classification model for type 2 diabetes that possesses several important advantages over conventional clinical methods (specifically, FINDRISC, ADA risk score). Performance metrics for the diabetes diagnostic model were 0.96 AUC, 92% specificity, and 89% sensitivity (mean values).ConclusionsThe study results potentially have major clinical implication and provide a contribution to the field of conventional diabetes risk assessment tools. Being specifically trained on real-world laboratory data and based on satisfactory external validation results, the present diagnostic type 2 diabetes model demonstrates high generalizability and can serve as a medical decision support and health monitoring tool. |
