Prediction of Out-of-Pocket Health Expenditures: A Methodological Comparison of Machine Learning Approaches – A Methodological Review
Redefining Leadership: Adapting to AI-Driven Dynamics in Organizations, İzmir, Türkiye, 31 Ağustos - 04 Eylül 2026, ss.216-225, (Özet Bildiri)
- Yayın Türü: Bildiri / Özet Bildiri
- Basıldığı Şehir: İzmir
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.216-225
- Kütahya Sağlık Bilimleri Üniversitesi Adresli: Evet
Özet
We provide a methodological review of
machine learning (ML) approaches that predict out-of-pocket (OOP) and catastrophic
health expenditures (CHE). We review 14 studies published after 2010 and
evaluate data sources, variable selection, class imbalance, and explainable AI
usage. The evidence shows that ensemble algorithms, particularly Random Forest
and XGBoost, consistently demonstrate superior predictive performance because
they can handle complex and skewed data. The literature also identifies
advanced age, chronic diseases, low income, and lack of insurance as the
primary risk factors driving financial hardship. Despite ML’s strong
forecasting potential, the current literature faces significant methodological
limitations. These limitations include substantial heterogeneity in
dependent-variable and CHE-threshold definitions, rare use of data balancing
techniques, and heavy reliance on household survey data. Survey-only designs
introduce critical measurement validity issues, such as recall bias and
underreporting of informal payments. To develop reliable decision-support tools
for health policy, we recommend stronger theoretical grounding, adherence to
standardized reporting guidelines such as TRIPOD, and integration of household
surveys with administrative records through data linkage to overcome
measurement errors.
Keywords: catastrophic health
expenditure, explainable artificial intelligence, health financing, machine
learning, out-of-pocket health expenditures