Predictive Model for Dementia in Patients with Type 2 Diabetes Using a Common Data Model

Speaker(s)

Lee S1, Choi K2, Lee J3, Suh HS4
1Department of Regulatory Science, Graduate School, Kyung Hee University, Institute of Regulatory Innovation through Science (IRIS), Kyung Hee University, Seoul, Seoul, Korea, Republic of (South), 2Department of Regulatory Science, Graduate School, Kyung Hee University, Institute of Regulatory Innovation through Science (IRIS), Kyung Hee University, Seoul, Korea, Republic of (South), 3Department of Regulatory Science, Graduate School, Kyung Hee University, Institute of Regulatory Innovation through Science (IRIS), Kyung Hee University, Seoul, South Korea, 4College of Pharmacy, Kyung Hee University, Department of Regulatory Science, Graduate School, Kyung Hee University, Institute of Regulatory Innovation through Science, Kyung Hee University, Seoul, Korea, Republic of (South)

OBJECTIVES: This study aimed to establish a predictive model of dementia in patients with type 2 diabetes (T2DM) using the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) and machine learning techniques.

METHODS: This study was conducted using CDM data from Ajou University Medical Center, which is a tertiary general university hospital between January 01, 1993, and November 30, 2023. The index date was defined as the earliest date of diagnosis of T2DM between January 01, 2001, and April 30, 2013. We used a 1-year wash-out period to ensure that the sample had no previous diagnoses of diabetes. The outcome was defined as the earliest dementia event during the 10 years following the index date. We used sex, age, procedure, measurement, CHA2DS2VASc, diabetes complication severity index, and Charlson comorbidity index as covariates. The data were randomly split into distributions of 75% for training and 25% for testing. LASSO logistic regression, random forest, gradient boosting machine, and naive Bayes were used to develop the models. Model performances were assessed by the area under the ROC curve (AUC), accuracy, recall, precision, and specificity.

RESULTS: Of the 2,765,800 patients who visited Ajou University Medical Center, 9,279 were newly diagnosed with T2DM. Over the 10-year follow-up period, 169 patients were diagnosed with dementia (mean age= 69.3 years; females=58.2%). The LASSO logistic regression model had the best AUC (AUC=78.8%; accuracy=50.5%; recall=92.3%; precision=3.1%, and specificity=49.8%), demonstrating acceptable predictive ability. Age, disorder of the nervous system, individual psychotherapy, osteoporosis, and candesartan were selected as features for predicting dementia.

CONCLUSIONS: We developed and validated a predictive model for dementia in T2DM patients. By identifying dementia in patients with T2DM, clinicians can take proactive measures to prevent dementia, improving patient's quality of life. Therefore, this model could potentially be utilized to improve clinical decision-making and optimize dementia care.

Code

RWD29

Topic

Study Approaches

Topic Subcategory

Electronic Medical & Health Records

Disease

Diabetes/Endocrine/Metabolic Disorders (including obesity), No Additional Disease & Conditions/Specialized Treatment Areas