Most of us like our news (medical or otherwise) pre-digested into understandable chunks for consumption. This need for this is inevitable because we expect to have an understanding of a broader range of things than it is possible to be an expert in. This places huge responsibility on the shoulders of the chunkifiers, particularly the researchers themselves.
The take home from major cardiovascular trials has traditionally been the hazard ratio. If it’s 0.5 that’s a halving of the rate of a composite of mortality and cardiovascular events, say. We are all familiar with this even if we don’t fully get it so it serves as a reasonable way of communicating results.
Are we ready for something new? The win ratio is able to include more information from trial events (not simply the first event that occurs which is what the hazard ratio considers) and allows prioritization of endpoints by clinical importance (although this ordering has to be done carefully as the win ratio is sensitive to this). I say new but actually the win ratio has been around for years. In essence each patient in the active arm is compared to every patient in the control arm to see who has the better outcome – who “wins”. There are data to suggest this approach is valid. And trials are using it, like the 2018 COAPT trial in which the win ratio for MitraClip for all-cause death or heart failure hospitalization at 2 years was 1.61 with a 95% confidence interval of 1.29 to 2.04 and p value <0.001. 1.61 is the number of “winners” divided by the number of “losers”. So greater than 1 indicates more wins. Seems straightforward enough, doesn’t it?
So what’s the issue? Redfors et al have thrown out any remaining excuses by providing R code for estimating sample size by simulation and for analysis. They also explain in their excellent appendix:
“the win ratio is effectively a weighted sum of the effect on individual components of the primary outcome. A consequence of this is that the underlying win ratio is sensitive to the choice of follow-up time. For example, suppose we use the win ratio to analyse a trial where time to death is the highest priority outcome and time to heart failure hospitalization is the second priority outcome. Suppose that early on there are far more heart failure hospitalisations than deaths; the win ratio will mainly describe the effect of treatment on heart failure hospitalization. However, in the long term the win ratio gradually becomes more strongly weighted towards the effect of treatment on death. The win ratio in the short term is therefore often not equal to the win ratio in the long-term because effects of treatment on each component may differ”
But, since hazard ratios are also influenced by follow-up duration, this limitation shouldn’t hold us back from using win ratios. Are we just reluctant to adopt new methods? I suppose another issue is lag; trials being published in 2020 might have been designed in 2010, for example.
Win ratios seem to make it even easier for people to understand the results of trials, not to mention being a better reflection of what we need a therapy to achieve in clinical practice.