Why you need to test for violations of the proportional hazards assumption

An interesting article in Value in Health reviews how HTA bodies have (or have not) required testing for violation of the proportional hazards assumption. The Academic Health Economists Blog describes why simply assuming proportional hazards is a bad idea and reviews the Monnickendam et al. (ViH 2019) article:

If you are an HTA body…[testing for violation of the proportional hazards assumption] is important for at least three reasons. First, if hazards are non-proportional, this can mean that the average hazard ratio (treatment effect) from the trial is a poor representation of what is likely to happen beyond the trial period – a big issue if you are extrapolating data in an economic model. Second, if hazards are non-proportional, this can mean that the median survival benefit from the trial is a poor representation of the mean benefit (e.g. in the case of a curve with a “big tail”). If you don’t account for this, and rely on medians (as some HTA bodies do), this can result in your evaluation under-estimating, or over-estimating, the true benefits and costs of the medicine. Third, most approaches to including indirect  comparison in economic models rely on proportionality so, if this doesn’t hold, your model might be a poor representation of reality. Given these issues, it makes sense that HTA bodies should be looking for violations in proportional hazards when evaluating oncology data.

…the authors review the way different HTA bodies approach the issue of non-proportionality in their methods guides, and in a sample of their appraisals. Of the HTA bodies considered, they find that only NICE (UK), CADTH (Canada), and PBAC (Australia) recommend testing for proportional hazards. Notably, the authors report that the Transparency Committee (France), IQWiG (Germany), and TLV (Sweden) don’t recommend testing for proportionality. Interestingly, despite these recommendations, the authors find that solely the majority of NICE appraisals they reviewed included these tests, and that only 20% of the PBAC appraisals and 8% of the CADTH appraisals did. This suggests that the vast majority of oncology drug evaluations do not include consideration of non-proportionality – a big concern given the issues outlined above.

One important example where non-proportional hazards is important is for measuring survival gains from immuno-oncology treatments for cancer. These treatments have a fat tail meaning that benefits are modest for many individuals but for a subset of people there are very long survival gains compared to previous standard of care (e.g., chemotherapy).

Why you need to test for violations of the proportional hazards assumption posted first on http://dentistfortworth.blogspot.com