Calibrating Parameters for Microsimulation Disease Models: A Review and Comparison of Different Goodness-of-Fit...

Area of Science:

• Biostatistics
• Health Economics
• Computational Biology

Background:

• Model calibration involves comparing model outcomes with observed data to estimate parameters.
• Goodness-of-fit (GOF) criteria quantify discrepancies between model and observed outcomes.
• A lack of consensus exists on the optimal GOF criterion for model calibration due to limited comparative studies.

Purpose of the Study:

• To systematically compare the performance of commonly used GOF criteria in microsimulation model calibration.
• To evaluate GOF criteria based on parameter estimation accuracy, computational efficiency, and impact on cost-effectiveness ratios.
• To identify the most suitable GOF criterion for calibrating the MISCAN-Colon microsimulation model for colorectal cancer.

Main Methods:

• Systematic comparison of sum of squared errors (SSE), Pearson chi-square, and likelihood-based deviance criteria.
• Assessment of GOF criteria performance using root mean squared prediction error (RMSPE) for parameters.
• Evaluation of computation time and impact on estimated cost-effectiveness ratios across various calibration scenarios.

Main Results:

• Likelihood-based deviance demonstrated the lowest RMSPE in most scenarios, indicating superior parameter accuracy.
• SSE and Pearson chi-square exhibited significantly higher RMSPE in specific data scenarios.
• Likelihood-based deviance provided the most accurate cost-effectiveness ratio estimation compared to SSE and Pearson chi-square.

Conclusions:

• Likelihood-based deviance criteria ensure accurate parameter estimation across diverse circumstances in microsimulation models.
• These criteria are recommended for model calibration, offering advantages over traditional methods like SSE and Pearson chi-square.
• The findings support the use of likelihood-based deviance for robust calibration in disease modeling.