The defining feature of a confidence interval is that it has a fixed minimu
m probability of covering the true value of the parameter being estimated,
whatever the value of the parameter. The authors demonstrate by simulation
that some recently proposed methods for interval estimation of the incremen
tal cost-effectiveness ratio (ICER) either do not satisfy this definition o
r have other problems that limit their usefulness in applications. The prob
lems are most prominent when the ICER is large and the true effectiveness d
ifference is small relative to its standard error. A modification of the pe
rcentile bootstrap confidence interval that involves a reordering of the sa
mple space provides a partial solution of the problem.