Confidence intervals for right-censored outcome with no events

data analysis uncertainty
1

In the context of a variable with a binomial distribution, there are methods like the Wilson score interval that produce sensible confidence intervals even when no events are observed

For an analysis I’ve been working on, I’ve got two treatment groups, and one of them has a relatively small size. The outcome of interest is uncommon, so in the smaller treatment group, we observe 0 events. I can construct the cumulative risk curves for each group (it would be a flat line at 0 for the group with no events). What I can’t find guidance on is how to construct a 95% confidence interval for the cumulative risk curve when there are no observed events.

Can anyone offer guidance on how to tackle this?

2

Use this paper to find the effective sample size at each time (as if there was no censoring) and use the Wilson interval of the K-M point estimate and effective N as if it were binomial.

3

Thank you! I was thinking along those lines, but I hadn’t made the connection to use the effective sample size