A simulation study to compare methods for constructing confidence intervals for the incremental cost-effectiveness ratio

The incremental cost-effectiveness ratio (ICER) is a well accepted measure of cost-effectiveness in health sciences. Due to the complexity of a ratio of random variables, constructing an appropriate confidence interval for the ICER is challenging. Many methods have been proposed, yet systematic comparisons of these methods are rare. Also, there have been conflicting recommendations for the optimum methodology. Polsky et al. (Health Econ. 6: 243-252, 1997) and Briggs et al. (Stat. Med. 18: 3245-3262, 1999) have performed simulation studies comparing four and eight methods, respectively. Cook and Heyse (Stat. Med. 19: 2989-3003, 2000) have proposed an angular transformation for addressing the problem of improper ordering of ICER estimates using bootstrap resampling, and compared several methods with and without the transformation. In this paper we have conducted an extensive simulation to compare the most commonly used methods as well as new methods that have not yet been evaluated in the aforementioned simulation studies. We simulated samples from a wide range of distributions, and considered the less desired scenarios when the true standardized incremental effect was near zero. The results suggest that confidence intervals based on Fieller's method, bootstrap-percentile method, and bootstrap-standard method consistently yield reasonable coverage percentages across different sample distributions. Other methods depend heavily on sample sizes, as well as how far the standardized incremental effect is away from zero. Confidence intervals based on the transformed methods are exceptionally problematic when the standardized incremental effect is small. We have also illustrated our results with a real example.