Predicting Mild Cognitive Impairment in Type 2 Diabetes: A Machine Learning Approach.

Area of Science:

• Neuroscience
• Endocrinology
• Artificial Intelligence

Background:

• Diabetes Mellitus (Type 2 Diabetes Mellitus - T2DM) is a significant risk factor for cognitive impairment, including mild cognitive impairment (MCI).
• Early detection of MCI in T2DM patients is critical for effective intervention and management of cognitive decline.
• Developing predictive models can aid in identifying at-risk individuals within the T2DM population.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting mild cognitive impairment (MCI) in patients with Type 2 Diabetes Mellitus (T2DM).
• To identify key clinical and demographic predictors associated with MCI risk in T2DM.
• To assess the generalizability and reliability of the developed ML model across diverse populations.

Main Methods:

• Utilized a dataset of 2074 T2DM participants with cognitive assessments.
• Employed statistical methods and genetic programming for feature selection to mitigate collinearity.
• Trained and evaluated six classification models using cross-validation and hyperparameter tuning, including a Support Vector Classifier (SVC).
• Validated the model externally using the DECODE and NHANES III cohorts, and employed SHAP analysis for predictor interpretation.

Main Results:

• The Support Vector Classifier (SVC) model demonstrated superior performance, achieving an AUC of 0.74 in internal validation.
• External validation with the DECODE cohort showed improved performance with an AUC of 0.80.
• Key predictors identified include education, age, glycolipid metabolism (GCA index), systolic blood pressure, eGFR, BMI, and diabetes duration.

Conclusions:

• The developed ML-based SVC model is effective and accurate in predicting MCI in T2DM patients.
• Machine learning holds significant potential for the early diagnosis and risk stratification of MCI within the T2DM population.
• The model's performance across external validation cohorts underscores its clinical utility for identifying individuals at risk of cognitive impairment.