Prediction of Metformin-Associated Lactic Acidosis Using Machine Learning Techniques : A Case Study of Surin Hospital

Abstract

Objective: To develop a predictive model for metformin-associated lactic acidosis (MALA) and compare the prediction of MALA among various models using different machine learning techniques. Methods: The study retrospectively collected data from electronic medical records of diabetic patients aged 18 years and over who received metformin at Surin Hospital between January 2017 and December 2021. The factors used to build the model were body mass index, metformin daily dose, obesity, NSAIDs use within the past 6 months, thiazides use, use of beta blockers, use of statins, renal function, dyslipidemia and duration of being diagnosed with diabetes. The study used a supervised machine learning technique with classification algorithm. Machine learning techniques used were Logistic Regression (LR), Multi-layer Perceptron (MLP), Support Vector Machine (SVM), and Random Forest (RF). Results: There were 8,505 diabetic patients receiving metformin including 8,387 of those without MALA and 118 with MALA. RF and MLP models were significantly more accurate than LR. The RF model showed sensitivity 93.30%, specificity 93.57%, accuracy 95.26%, and AUROC of 0.992. Conclusion: The MALA prediction model using RF machine learning technique was the most efficient model.