Purpose: Our previously developed deterministic model has been used to assess the cost-effectiveness of the 2002 American Cancer Society (ACS) guidelines for cervical cancer screening. To evaluate more complex screening strategies that tailor to the risk-profile and history of an individual woman, we developed a first-order Monte Carlo simulation model, employing a two-step calibration approach. We explored policy implications using different model fit criteria, and compared results from the new and original models.

Methods: In the first step of calibration, we capitalized on the availability of primary data from a longitudinal study, and identified a plausible range for each parameter that produced model output that fell within the 95% confidence intervals of the data. In the second step, we included data from the literature to establish 32 calibration targets and ranges. Through multiple simulations, we conducted a comprehensive search over 14 input parameters simultaneously, to identify parameter sets that produced output that fit with both the primary and secondary data. We evaluated clinical benefits and cost-effectiveness of the original 2002 ACS screening strategies using different definitions of good fit.

Results: We scored 140,000 uniquely sampled parameter sets according to the number of times the model output fell outside target ranges. Among the best scoring sets, reductions in lifetime cancer risk ranged from 58-82%, depending on screening frequency and strategy. As we incrementally relaxed our fit criterion to include 1%, 10% and 50% of the top runs, the range of projected benefits widened considerably. For example, for the top 50% scoring sets, while the average benefits were similar to those obtained with the best scores, the variability more than doubled. Similar trends were apparent with the cost-effectiveness ratios associated with different strategies. Finally, when we used our initial set of parameters prior to calibration, which achieved a score between the best and the mean scores, the benefits and cost-effectiveness results were similar to those based on the best-fitting calibrated sets.

Conclusions: Depending on the fit criterion, uncertainty around policy results may differ, but the model parameters prior to calibration may have been “good enough” in evaluating the cost-effectiveness of the 2002 ACS guidelines. We anticipate an analysis of the 2006 guidelines, which will involve more individualized and complex screening protocols, may not produce the same results.