Estimating parameters of a microsimulation model for breast cancer screening using the score function method

In developing decision-making models for the evaluation of medical procedures, the model parameters can be estimated by fitting the model to data observed in (randomized) trials. For complex models that are implemented by discrete event simulation (microsimulation) of individual life histories, the Score Function (SF) method can potentially be an appropriate approach for such estimation exercises. We test this approach for a microsimulation model for breast cancer screening that is fitted to data from the HIP randomized trial for early detection of breast cancer. Comparison of the parameter values estimated using the SF method and the analytical solution shows that method performs well on this simple model. The precision of the estimated parameter values depends (as expected) on the size of the sample of simulated life histories, and on the number of parameters estimated. Using analytical representations for parts of the microsimulation model can increase the precision of the estimated parameter values. Compared to the Nedler and Mead Simplex method which is often used in stochastic simulation because of its ease of implementation, the SF method is clearly more efficient (ratio of computer time to precision of estimates). The additional analytical investment needed to implement the SF method in an (existing) simulation model may well be worth the effort.