search

DARTH-git / dampack

Decision Analytic Modeling (DAM) package is a suite of functions for analyzing and visualizing the health economic outputs of mathematical models
34 stars 9 forks source link

Request - add confidence interval to function calculate_icers #160

Closed krijkamp closed 3 years ago

krijkamp commented 3 years ago

Can we add an argument CI where we can also have the output of 95%-CI of the costs and effects? Maybe by default FALSE, but having the option would be very nice.

gknowlt commented 3 years ago

@krijkamp Were you envisioning that these 95% CI's would be constructed on the basis of the distribution of outcomes realized across a PSA? I think this could have some useful applications, but I'm a bit wary of the scenarios in which the ordering of strategies change across different PSA samples. We could end up seeing negative ICERs, which are very unintuitive and ambiguous in terms of whether the numerator or denominator is negative in the ICER, or extremely large SEs for the ICERs if the distribution of incremental QALYs sits anywhere near 0. Do you have any thoughts on this, or sources that establish the best ways for handling these issues? I think if we set the default to FALSE like you recommended as well as provide some sort of warning message about the unreliability of ICER CIs that could work out well.

krijkamp commented 3 years ago

Great points @gknowlt! I was envisioning having 95%-CI around the costs and effects based on the estimates costs and effects during all PSA runs indeed. And this in turn can also apply to the incremental costs and effects and maybe even ICER, but than your points are very valid, as the order of strategies might changes based on a PSA iteration. I belief for the mean costs and mean effects of each strategy this should not be an issue. For those we can provide a 95-%CI.

For the incrementals, I don't have an immediate solution. But let me think about this and I will also discuss with Petros as this idea came up while we were teaching a course. Maybe he has a good idea.

krijkamp commented 3 years ago

@gknowlt I think I know how we should approach this. Lets discuss in a call.