Predicting individual patient outcomes using prognostic models in economic evaluations

Objective When estimating incremental quality adjusted life years (QALYs) and costs in economic evaluations, prognostic models can be applied to predict survival times. However, these models do not themselves estimate whether the event, e.g. death or survival, would actually occur or not. When this projection is needed it is important to fully incorporate the uncertainty around it.

Study Design and Setting This paper compares two methods for estimating patient specific outcomes. The average probability method uses the mean estimated proportion of survivors at a particular time point and assumes the patients with the longest survival times are the survivors. The second method uses probabilistic sensitivity analysis (PSA) to simulate individual patient outcomes. The two methods are illustrated using a prognostic model for estimating survival in the absence of liver transplantation.

Results The mean survival, QALYs, costs and incremental cost-effectiveness ratio (ICER) were similar for the two methods. 95% confidence intervals were slightly wider for survival and QALY estimates and substantially wider for cost and ICER estimates when using PSA to estimate patient outcomes, thus capturing outcome uncertainty at the individual level.

Conclusion PSA gives more realistic confidence intervals representing uncertainty than an average probability method and is the recommended method when estimating individual patient outcomes from prognostic models.