Use of Machine Learning and Routine Laboratory Tests for Diabetes Mellitus Screening.

Area of Science:

• Medical Informatics
• Computational Biology
• Data Science in Healthcare

Background:

• Diabetes mellitus often presents asymptomatically, delaying diagnosis and complicating treatment.
• Routine clinical laboratory examinations generate extensive lifetime health datasets.
• Computer processing offers potential for anomaly detection and disease prediction from clinical data.

Purpose of the Study:

• To evaluate machine learning models for diabetes screening using routine laboratory tests.
• To assess the performance of various classification and regression models in predicting glycated hemoglobin.
• To determine the efficacy of artificial neural networks in identifying diabetes and prediabetes.

Main Methods:

• Utilized a dataset of 62,496 patients' laboratory test results.
• Employed K-nearest neighbor, support vector machines, Bayes naïve, random forest, and artificial neural network models.
• Trained regression models to predict glycated hemoglobin, followed by classification.

Main Results:

• Artificial neural networks demonstrated the best performance in detecting prediabetes or diabetes.
• The artificial neural network classification model achieved 78.1% sensitivity, 78.7% precision, and 78.4% F1 score for the 'no healthy' group.
• Various models showed good performance depending on specific classification goals.

Conclusions:

• Machine learning models can effectively predict glycated hemoglobin values from routine laboratory tests.
• These models can serve as a valuable screening tool to identify patients requiring further diabetes evaluation.
• Integrating AI into routine diagnostics can improve early detection and management of diabetes mellitus.