The effect size is, as a rule, the main driver of the cost-effectiveness of an intervention, and therefore deserves special attention in uncertainty analysis.
In modeled health-economic evaluations the uncertainty is mostly quantified using Monte Carlo simulation (aka probabilistic sensitivity analysis, or PSA). In this procedure key variables in the model are represented by statistical distributions instead of point estimates. By repeatedly drawing random values from these distributions and each time recalculating the model, a distribution of outcomes is obtained that embodies the combined uncertainty of the key variables.
The validity of PSA crucially depends on choosing appropriate distributions and parameters for the key variables. As argued in a recent study published in Value in Health, an appropriate distribution for the effect size can be obtained by taking our cue from epidemiology, where its uncertainty is modeled by a Lognormal distribution with the natural log of the relative risk and its standard error as parameters.
However, the study, "The effect size in uncertainty analysis" by Jan Barendregt of the University of Queensland, shows that this choice poses a dilemma.
Says Dr. Barendregt, "This choice fails to achieve a few simple and quite reasonable demands, such as that the mean of the randomly drawn values equals the point estimate of the effect size. My article discusses the dilemma, looks at two correction methods, and proposes a preferred one".
The preferred correction method is implemented in Ersatz, a powerful but low cost Monte Carlo add-in program for MS Excel™, with specific features for uncertainty analysis in health economic evaluation.
Value in Health (ISSN 1098-3015) publishes papers, concepts, and ideas that advance the field of pharmacoeconomics and outcomes research and help health care leaders to make decisions that are solidly evidence-based. The journal is published bi-monthly and has a regular readership of over 4,000 clinicians, decision-makers, and researchers worldwide.
Source:
ISPOR