Purpose: The purpose of this study is to illustrate situations in which bootstrapping, used to correct for the bias of the incremental cost-effectiveness ratio (ICER) estimate, can lead to a “corrected” estimate that is even more biased.

Methods: To illustrate how correcting for the bias using bootstrapping can actually make things worse, we generate a hypothetical data set of individual patient level data (e.g., from a clinical trial). This allows us to know the true cost and effect means, and thus the true ICER. We compute the bias-corrected ICER using the method recommended in the literature. We compare the true ICER estimate with the bias-corrected and uncorrected ICER estimates.

Results: In the illustrative example, the bias-corrected estimate was farther from the true estimate than the uncorrected estimate. Correcting for the bias in the sampling process made things worse. Graphing the results hints at circumstances where bias-correction is contraindicated.

Conclusion: Decision makers in healthcare are assumed to be facing a constrained optimization problem with the desire to maximize population health curtailed by limited budgets to pay for new treatments and novel drugs. A proposed solution involves comparing an ICER to a threshold. Calculation of the ICER has become quite popular in published cost-effectiveness analysis (CEA) studies. CEA studies are frequently used by decision makers to choose which healthcare treatments will be funded and which will not. Better decisions might be reached with better estimates of the ICER. In this paper, we illustrate how the bootstrapping process to correct bias can lead to a “corrected” estimate that is even more biased. We explain how this is because of paradoxical scenarios that can occur with ICERs. We urge researchers to look at the uncertainty in their ICER estimates so that they may create better ICER estimates.