More paired data changed win ratio estimates and confidence intervals

What the study found: The study found that increasing the number of matched data pairs affected the win ratio statistic and its 95% confidence intervals for two composite endpoints, but not in a regular way.
Why the authors say this matters: The authors conclude that the number of paired data affects the number of significant win ratios and the maximum win ratio, and that the win ratio confidence interval is more affected by data structure than by the size of the paired sample.
What the researchers tested: This was a simulation study using 35 scenarios with matched data pairs ranging from 10 to 1000. The researchers generated 34,101 data sets with the Python random library and examined the win ratio statistic and its 95% confidence intervals.
What worked and what didn't: As the number of paired data increased, the win ratio generally increased, although not regularly. The number of cases where the win ratio was greater than one did not change with the number of paired data, while the win ratio values considered important increased with larger paired-data numbers. The lower and upper confidence limits tended to move closer together, and the difference between them decreased, but again not regularly.
What to keep in mind: The abstract does not describe practical limitations beyond the simulation design. The findings are based on generated data across specified scenarios, so the summary is limited to those simulated conditions.

Key points

• The study examined how the number of matched data pairs affects the win ratio statistic and its 95% confidence intervals.
• Across simulated scenarios, the win ratio generally increased as paired-data numbers increased, but not regularly.
• The number of cases with a win ratio greater than one did not change with the number of paired data.
• The lower and upper confidence limits tended to come closer together as paired-data numbers increased, though not consistently.
• The authors say the confidence interval is more affected by data structure than by sample size.