How can artificial intelligence revolutionize the way we model health-state utilities in health technology assessments (HTA)? We explored this question by applying Symbolic Regression (SR) through Genetic Programming (GP), an AI technique capable of discovering mathematical relationships automatically, to model utilities as functions of health states.
The goal of this work was to assess whether Genetic Programming-based Symbolic Regression could more effectively model utilities than conventional linear regression. Accurate utility modelling is critical in HTA, influencing key measures such as Quality-Adjusted Life Years (QALYs) and Incremental Cost-Effectiveness Ratios (ICERs).
Using data from the TOPICAL trial (Erlotinib vs Best Supportive Care in advanced non-small cell lung cancer [NSCLC]*), we defined three key health states:
- Progression-free
- Progressive disease
- Death
The Genetic Programming approach, implemented via Mathematica®’s Data Modeler®, automatically generated and optimized models based on criteria such as R² and Akaike Information Criterion (AIC).
We found that:
- Over 1,200 models were generated in under three minutes.
- The top-performing models achieved R² ≈ 9%, although at the cost of high complexity.
- Simpler models (complexity 42) achieved R² ≈ 6.5% while offering far better interpretability.
- Compared with standard linear regression, SR models consistently demonstrated higher R², lower AIC, and more accurate mean utility estimates.
This study highlights the potential of AI-driven Symbolic Regression to enhance utility modeling within HTA. By automatically identifying nuanced, non-linear relationships between health states and utilities, SR can deliver more accurate, interpretable, and time-efficient models.
Ultimately, these AI-based approaches could significantly impact QALY and ICER outcomes, paving the way for smarter, data-driven health economic evaluations.
📄 Explore the full poster:
Modelling Utilities as Functions of Health States Using AI and ML through Genetic Programming (PDF)