Skip to main content
Search

Bayesian methods in health economics: probabilistic sensitivity analysis 4

Part 4 of 5
Published on February 16, 2009 Reviewed on March 30, 2022   56 min

A selection of talks on Methods

0:00
Hi, I'm Tony O'Hagan. Welcome to the fourth talk in my series, Bayesian Methods in Health Economics. This talk is entitled Probabilistic Sensitivity Analysis.
0:14
Let's quickly review one more time the way the whole course looks. The first two talks dealt with Bayesian approach to statistical inference in a general way. Then in the third talk, we moved on to explaining how Bayesian methods are applied in the area of health economic evaluation, and in particular, we looked at the uncertainties in health economic evaluation. This talk is about probabilistic sensitivity analysis, which is the most important way that health economists use Bayesian methods. The final talk will deal with another aspect of probabilistic sensitivity analysis, that of formulating input uncertainties.
0:56
Here's the road map for this fourth talk, we begin by discussing economic modelling, which is the context in which probabilistic sensitivity analysis takes place. We then move on to PSA itself, and identify three key steps in doing that. Now, the fifth talk is entirely devoted to one of those steps, that of quantifying input uncertainties. The remainder of this talk deals with the other two important steps. One is computing output uncertainties, and finally, how we express those uncertainties and how they influence decision making.
1:31
The context, as I explained in the previous talk, number 3, one way that Bayesian methods are used in the health economic evaluation is to analyse the outcomes of clinical trials in which we measure both costs and efficacy. But as I also explained then, this is rarely the case that health economists find themselves in. A single clinical trial will rarely provide all the evidence that you need for cost-effectiveness study. There are a number of reasons for this. First of all, let's emphasise the single clinical trial. Very often there is more than one trial, and we need to be able to combine the information from several things. But in that case, almost surely some of them will not deal with costs, and were only looking at efficacy. Indeed, the common situation is where clinical trials have only been evaluating efficacy. Even where we have a trial in which costs might be measured as well the conditions of those trials are rarely what we need for deciding on cost-effectiveness because in a clinical trial situation, we invariably are looking at limited follow-up, and that may be very important in clinical diseases which take some time to evolve, we're invariably looking at restricted enrollment, which ensures that you get a patient group for which you might be able to observe an effect, whereas in the real world, we might be wanting to use this treatment in a wider population group. We have in clinical trials an unrealistic level of compliance. The patients typically will comply with the taking of the medicines far more so than they would out in their own homes. We have the wrong comparators very often. We are looking at comparisons against placebos or against treatments that are not the ones that we're really comparing this treatment when it comes to cost-effectiveness. Finally, we very often have the wrong outcomes in a clinical trial. People run clinical trials in order to demonstrate some effect, but it may not be the effect that in economic terms would be most valuable to see. There are many ways in which a clinical trial fails to live up to what a health economist needs in order to evaluate cost-effectiveness for the treatment under consideration against the relevant comparators. Instead, health economists rely on a process of modelling. We look at a very simple example in order to
Show
Hide

Bayesian methods in health economics: probabilistic sensitivity analysis 4

Embed in course/own notes