In a world where health systems are under constant pressure to allocate resources effectively, understanding and predicting healthcare demand is no longer optional—it’s essential. A new study published in BMC Health Services Research by Orhan and Kurutkan (2025) harnesses the power of machine learning (ML) to dissect and forecast healthcare utilization across Türkiye, using nationally representative data and a well-established theoretical framework.
The Theoretical Lens: Andersen’s Behavioral Model
The research applies Andersen’s Behavioral Model of Health Services Use, which categorizes the drivers of healthcare usage into three domains:
- Predisposing Factors (e.g., age, gender, education),
- Enabling Factors (e.g., income, insurance, access barriers),
- Need Factors (e.g., chronic illness, perceived health).
This tripartite structure allows for a nuanced view of what prompts people to seek medical care—and how.
Data & Methodology
The study analyzed data from the 2022 Turkey Health Survey conducted by the Turkish Statistical Institute (TÜİK), involving over 22,000 participants. Seven machine learning models were tested:
- Decision Tree
- Random Forest
- Support Vector Machine (SVM)
- K-Nearest Neighbors (KNN)
- Logistic Regression
- XGBoost
- Gradient Boosting
Key metrics for evaluating model performance included recall, precision, F1-score, and ROC AUC.
Key Findings
- Model Performance:
The standout performers were Gradient Boosting, XGBoost, and Logistic Regression, each showing superior F1 scores (~0.88) and recall rates (~0.90), with balanced accuracy and generalization capability. - Most Influential Features by Category:
- Predisposing: Gender, education level (especially no formal education), and age groups (especially children and seniors).
- Enabling: Self-paid treatment costs, community interest, appointment delays, and payment difficulties for dental and mental health services.
- Need: Smoking status, chronic illnesses (especially depression and hypertension), and perceived health status.
- Integrated Model Insights:
When all factors were combined, the most critical predictors included smoking status, gender, BMI, community support, chronic diseases, and mental health challenges. - Policy Implications:
The study underscores the usefulness of machine learning not just for prediction, but for guiding policy interventions. For example, identifying populations with payment difficulties or mental health needs could help design targeted subsidy programs or outreach initiatives.
Why This Matters
Unlike traditional statistical models, ML algorithms can capture complex, nonlinear relationships and interactions among variables. This makes them ideal for modeling the multi-layered nature of healthcare demand. In particular, this study exemplifies how integrating sociological theory (Andersen’s model) with data science can yield actionable insights for health planners.
Conclusion
Orhan and Kurutkan’s work offers a compelling blueprint for predictive healthcare policy using machine learning. By clearly identifying which factors most influence service use—whether demographic, economic, or clinical—health systems can move towards data-driven, equity-focused planning.
This paper is a must-read for health policymakers, data scientists in healthcare, and anyone interested in the future of personalized and anticipatory health systems.
Reference
Orhan, F., & Kurutkan, M. N. (2025). Predicting total healthcare demand using machine learning: separate and combined analysis of predisposing, enabling, and need factors. BMC Health Services Research, 25(366). https://doi.org/10.1186/s12913-025-12502-5
Podcast link: https://notebooklm.google.com/notebook/6ab1a940-bda8-4171-a340-24c5efc83b17/audio