The Win Ratio: Ranking What Matters Most in a Composite Endpoint

The win ratio is a way to analyze a composite endpoint that ranks the outcomes by importance instead of treating them as equal. Patients in the two groups are paired and compared on the most serious event first, such as death, and only if that is tied are they compared on the next outcome, such as hospitalization. The win ratio is the number of pairs the treatment group won divided by the number it lost, so a value above one favors treatment.

The problem with ordinary composite endpoints

Trials often combine several outcomes into one composite endpoint, such as death, heart attack, and hospitalization, to gather enough events to detect a difference. The conventional analysis counts the time to whichever of these happens first and treats them as interchangeable. That has a well-known flaw: a hospitalization, which is common, counts the same as a death, which is rare and far more important, and often drives the result.

So a treatment can look successful mainly because it reduced hospitalizations while doing nothing for survival. The headline says the composite improved, but the components that moved were the least serious ones. Readers who do not open up the composite can be misled about what actually changed.

How the win ratio works

The win ratio was introduced to fix this by building a hierarchy. Patients in the treatment and control groups are formed into pairs, often matched on baseline risk. Each pair is compared on the most important outcome first: who died, and when. If one patient died and the other did not, or died later, the surviving patient's group wins that pair.

Only if the pair is tied on death, because neither died in the follow-up available, does the comparison move to the next outcome in the ranking, such as heart-failure hospitalization. This continues down the list. Every pair ends as a win, a loss, or a tie. The method gives priority to the outcome clinicians and patients care about most, instead of letting a frequent minor event dominate.

Reading a win ratio

The win ratio is the total number of winning pairs divided by the total number of losing pairs, reported with a confidence interval and a p-value. A win ratio of 1.0 means the groups traded wins and losses evenly. A win ratio of 1.4 means the treatment group won about 40 percent more comparisons than it lost, favoring treatment. Below 1.0 favors control.

One caution: the win ratio is a ratio of wins, not a risk ratio or a hazard ratio, so it does not translate directly into an absolute benefit or a number needed to treat. It tells you the direction and the weight of evidence across a ranked set of outcomes, not how many extra people survived. Ties, which can be numerous when follow-up is short, are dropped from the ratio itself, though related measures account for them.

Why the ranking is the point

The ranking is not a technical detail; it is the whole idea. By resolving the most serious outcome first, the win ratio ensures that a difference in deaths counts more than a difference in hospitalizations, which matches how anyone would weigh them. It also lets a trial combine outcomes of different types, including how many times an event recurred or a continuous measure like a quality-of-life score, within one hierarchy.

This flexibility is why the approach spread quickly in cardiovascular trials, where death, hospitalization, and symptom burden all matter but plainly not equally. The prespecified order of outcomes is therefore a design choice worth scrutinizing, because whoever sets the ranking sets what the trial rewards.

Strengths, and what it does not fix

The strength is priority. The win ratio stops a common, minor event from swamping a rare, serious one, and it uses more of the information in the data than time-to-first-event alone. That can also improve the power to detect a real difference.

The limits matter. The result depends on the chosen hierarchy, so a different but defensible ordering could shift the answer. The win ratio is harder to translate into absolute terms, which is what patients ultimately need to weigh benefit against harm. And matching pairs on baseline risk involves choices that can affect the estimate. Like any single summary, it is best read alongside the individual components, not instead of them.

What to check when a trial reports one

Look first at the hierarchy: what outcomes were ranked, in what order, and was that order prespecified? A ranking that buries mortality below softer outcomes defeats the purpose. Next, open the components: did the serious outcomes actually move, or is the win ratio carried by the lower tiers? A trustworthy report shows the breakdown.

Then read the precision: a win ratio with a confidence interval that comfortably excludes 1.0 is more convincing than one that barely does. Finally, remember what the number is not. It signals direction and priority-weighted evidence, so pair it with the absolute event rates before deciding how much the result should change anything.