Calibration of disease simulation model using an engineering approach.

Area of Science:

• Computational epidemiology
• Mathematical modeling
• Health policy analysis

Background:

• Calibrating disease simulation models to clinical data is crucial for predictive accuracy.
• Key challenges include defining a composite goodness-of-fit (GOF) score for multiple targets and optimizing parameter searches.
• The Lung Cancer Policy Model (LCPM) is a microsimulation model used for lung cancer research.

Purpose of the Study:

• To apply an engineering approach to calibrate the LCPM to multiple clinical data targets.
• To address challenges in defining a total GOF score and optimizing parameter searches.
• To enhance the reliability and predictive capability of the LCPM.

Main Methods:

• A weighted-sum approach combined 11 calibration targets into a total GOF score, incorporating user-defined target importance.
• Automated parameter search algorithms, simulated annealing (SA) and genetic algorithm (GA), were employed to minimize the total GOF score across 28 natural history parameters.
• Performance metrics for speed and model fit were used to evaluate the algorithms.

Main Results:

• Both SA and GA achieved total GOF scores below 95 within 1000 iterations.
• Simulated annealing (SA) demonstrated superior performance in finding a lower GOF score compared to GA.
• The calibrated LCPM generated predictions for lung cancer's natural history consistent with existing mathematical models.

Conclusions:

• The engineering-based calibration method successfully achieved simultaneous fitting of LCPM outputs to multiple targets.
• This approach offers benefits of computational speed and reduced human input, minimizing potential bias.
• The calibrated LCPM provides a reliable tool for understanding lung cancer dynamics and informing policy.