MACHINE LEARNING-BASED PREDICTION OF CATASTROPHIC HEALTH EXPENDITURE AMONG HOSPITAL‑BASED PATIENTS WITH DIABETES IN VIETNAM
Author(s)
Thi Bao Ngoc Nguyen, MPH1, Thi Tao Tran, PhD2, Binh Thang Tran, Sr., MPH, DrPH3.
1Science - Technology & International Relations Office, University of Medicine and Pharmacy, Hue University, Hue, Viet Nam, 2Faculty of Public Health, University of Medicine and Pharmacy, Hue University, Hue, Viet Nam, 3Lecturer, University of Medicine and Pharmacy, Hue University, Hue, Viet Nam.
1Science - Technology & International Relations Office, University of Medicine and Pharmacy, Hue University, Hue, Viet Nam, 2Faculty of Public Health, University of Medicine and Pharmacy, Hue University, Hue, Viet Nam, 3Lecturer, University of Medicine and Pharmacy, Hue University, Hue, Viet Nam.
OBJECTIVES: We aimed to apply machine learning algorithms to classify CHE at the 40% capacity‑to‑pay threshold and identify key associated factors among households with at least one member with diabetes at Hue University of Medicine and Pharmacy Hospital in 2024.
METHODS: A cross-sectional study was conducted on 240 households with diabetic patients attending outpatient and inpatient services. CHE was defined as out-of-pocket spending on diabetes care exceeding 40% of household capacity to pay over one year. Sociodemographic, economic and clinical variables were collected through structured interviews and medical records. Five algorithms: logistic regression, random forest, XGBoost, decision tree and AdaBoost were trained and evaluated using accuracy, precision, recall, F1-score and the area under the ROC curve (AUC).
RESULTS: The annual mean direct medical cost was dominated by medicines, while non-medical costs were driven by functional foods and traditional medicines. The incidence of CHE at the 40% capacity‑to‑pay threshold was 33.3% among households with at least one member with diabetes. Logistic regression achieved the highest accuracy (0.90) with F1-score 0.93, while all models demonstrated high discriminative performance with AUC ≥ 0.90. Prior inpatient admission in the previous year was the most influential predictor of CHE, followed by number of diabetic patients in the household, number of older adults, household size and economic status.
CONCLUSIONS: Machine learning-based classification models can accurately predict CHE among households with diabetes using routinely collected data. Early identification of high-risk households could inform targeted financial protection policies and integrated support programs to prevent CHE in Viet Nam.
METHODS: A cross-sectional study was conducted on 240 households with diabetic patients attending outpatient and inpatient services. CHE was defined as out-of-pocket spending on diabetes care exceeding 40% of household capacity to pay over one year. Sociodemographic, economic and clinical variables were collected through structured interviews and medical records. Five algorithms: logistic regression, random forest, XGBoost, decision tree and AdaBoost were trained and evaluated using accuracy, precision, recall, F1-score and the area under the ROC curve (AUC).
RESULTS: The annual mean direct medical cost was dominated by medicines, while non-medical costs were driven by functional foods and traditional medicines. The incidence of CHE at the 40% capacity‑to‑pay threshold was 33.3% among households with at least one member with diabetes. Logistic regression achieved the highest accuracy (0.90) with F1-score 0.93, while all models demonstrated high discriminative performance with AUC ≥ 0.90. Prior inpatient admission in the previous year was the most influential predictor of CHE, followed by number of diabetic patients in the household, number of older adults, household size and economic status.
CONCLUSIONS: Machine learning-based classification models can accurately predict CHE among households with diabetes using routinely collected data. Early identification of high-risk households could inform targeted financial protection policies and integrated support programs to prevent CHE in Viet Nam.
Conference/Value in Health Info
2026-09, ISPOR Asia Pacific 2026, Bangkok, Thailand
Value in Health, Volume 55, Issue S1
Code
EE85
Topic
Economic Evaluation
Topic Subcategory
Cost/Cost of Illness/Resource Use Studies
Disease
SDC: Diabetes/Endocrine/Metabolic Disorders (including obesity)