Machine learning models can provide promising results for the task of predicting current A1C levels, according to a study published in JMIR Medical Informatics. Researchers investigated the performance of predictive models to forecast A1C elevation levels by employing several machine learning models and examined the use of patient EHR longitudinal data in the performance of the predictive models. The study team used multiple logistic regression, random forest, support vector machine, and logistic regression models, and a deep learning model (multiplayer perception) to classify patients with normal (<5.7%) and elevated (≥5.7%) levels of A1C, also integrating current visit data with historical data from previous visits. To interrogate the models and provide an understanding of the reasons behind the decisions the models made, they used explainable machine learning methods. A data set with more than 18,000 unique patient records was used to train and test all models. When coupled with historical data, the machine learning models outperformed a multiple logistic regression model used in a comparative study in predicting current A1C elevation risk. A multilayer perceptron model achieved an accuracy of 83.22% for the area under receiver operating characteristic curve when used with historical data. With and without longitudinal data, all models showed a close level of agreement on the contribution of random blood sugar and age variables.
