A machine learning-based framework to identify type 2 diabetes through electronic health records.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Genetics

Background:

• Genome-wide association studies (GWAS) and phenome-wide association studies (PheWAS) require large cohorts for discovering genotype-phenotype associations.
• Identifying Type 2 Diabetes Mellitus (T2DM) cases and controls from Electronic Health Records (EHR) is crucial but challenging due to low recall rates of existing expert-based algorithms.

Purpose of the Study:

• To develop a semi-automated machine learning framework to improve the identification of T2DM subjects from EHR.
• To liberalize filtering criteria to enhance recall rates while maintaining a low false positive rate for T2DM case identification.

Main Methods:

• A data-informed framework utilizing feature engineering and machine learning models (k-Nearest Neighbors, Naïve Bayes, Decision Tree, Random Forest, Support Vector Machine, Logistic Regression) was developed.
• The framework was tested on 300 patient samples from a regional EHR repository, comparing its performance against a state-of-the-art expert algorithm.

Main Results:

• The machine learning framework achieved high identification performance, with an average AUC of approximately 0.98.
• This performance significantly surpasses the state-of-the-art expert algorithm, which had an AUC of 0.71.
• The framework demonstrated superior accuracy, precision, AUC, sensitivity, and specificity in classifying T2DM subjects.

Conclusions:

• The proposed machine learning framework offers a more accurate and efficient method for identifying T2DM subjects from EHR data.
• This approach effectively loosens conservative selection criteria, leading to a higher identification rate of both cases and controls compared to expert algorithms.