OBJECTIVES:

Model-based analyses are typically useful for assessing the cost-effectiveness of a few but not a vast number of alternative health care strategies. We aim to illustrate the potential advantages of using a meta-model to identify the best screening strategy for colorectal cancer accounting for colonoscopy capacity constraints.

METHODS:

We defined screening strategies by starting age, interval, number of screening rounds, and screening test positivity threshold (16,464 unique strategies). We evaluated a limited sample of predefined strategies with the validated ASCCA model and identified the best screening strategy therein, in terms of life-years gained (LYG), compared with no screening. Next, this limited sample was used to fit a Gaussian Process meta-model. Finally, discrete evolutionary programming was used to iteratively identify the best possible screening strategy according to the meta-model (GP-DEP approach). Colonoscopy demand was restricted to 500 per 1,000 simulated individuals. The sample size of predefined strategies was varied (n=50-200). GP-DEP performance was assessed with bootstrapping (n=200), brute force exhaustive search, and comparison with ASCCA outcomes.

RESULTS:

GP-DEP resulted in stable predicted best screening strategies when applied to a sample of >=100 strategies, identifying the exact same best strategy as exhaustive search in 94% of bootstrap samples. Compared with ASCCA, predicted colonoscopy demand, LYG and costs of the best strategies from GP-DEP were accurate, slightly too high and slightly too low, respectively. However, strategy ranking (in deciles) according to ASCCA and GP-DEP were similar. For sample size 100, average predicted benefit of the best strategy identified by GP-DEP compared to the best strategy identified by ASCCA equalled 0.028 LYG (95%CI 0.013-0.043) per individual.

CONCLUSIONS:

Extending the ASCCA model with GP-DEP enhances performance: the best screening strategy can be identified much faster, even when constraints apply, and will outperform the best screening strategy as typically identified from a limited sample of predefined strategies.