(42) Identifying Factors with and Effects of Metformin on The Risk of Alzheimer's Disease and Other Dementias: A Machine Learning Approach

Saturday, August 27, 2022

8:00 AM – 6:00 PM CEST

Location: Hall C

Publication Number: 716

Presenting Author(s)

Xiaomo Xiong
– University of South Carolina College of Pharmacy, West Columbia, South Carolina, United States

Co-Author(s)

Minghui Li, PhD
– University of Tennessee Health Science Center, United States
Kevin Lu, PhD, BPharm, FISPE
– Associate Professor, USC College of Pharmacy, Columbia, South Carolina, United States

Background: Alzheimer's and related dementia (ADRD) affects more than 5 million people in the U.S., and the number is estimated to double by 2060. The literature documents that metformin may decrease the risk of ADRD although results are controversial.

Objectives: To identify factors with and assess the effects of metformin on the risk of Alzheimer’s disease and other dementias using real-world big data and machine learning methods.

Methods: Data were from Medicare Current Beneficiary Survey (MCBS), 2015–2017. The selected features included metformin, year, age, gender, race, marital status, education attainment, rural-urban residence, census regions, household income, Charlson comorbidities index (CCI), body mass index (BMI), and health status. Metformin was identified based on the generic name of metformin and the glucophage, glucophage XR, fortamet, glumetza, and riomet. Patients with ADRD were identified based on ICD-9 and ICD-10 codes. stochastic gradient boosting (XGboost) was used to examine the effects of metformin on the risk of ADRD. In addition, Shapley Additive exPlanations (SHAP) was used to identify the contribution and association of features to the risk of ADRD. All analyses were conducted using python 3.8.

Results: 33,590 individuals were included in this study. One-third of the included people (11,197) were assigned to the training group, and two-thirds (22,393) were assigned to the test group. The top 5 factors associated with the risk of ADRD were age, comorbidities, marital status, BMI, and race. Based on the area under the receiver operating characteristic (ROC) curve, the prediction performance was 0.51. According to the confusion matrix, the accuracy and the precision of the model were 94.4% and 45.7%, respectively. According to SHAP, although metformin ranked low among all features, it predicted a low risk of ADRD with the SHAP ranging from -0.022 to 0.003.

Conclusions: Our results suggest that age, comorbidities, marital status, BMI, and race are the most important predictors of the risk of ADRD. In addition, metformin may have a protective effect against the risk of ADRD.