Answers are grounded in this paper and cite their sources.
Prognostic reports claim that the risk of an event by a fixed time lies below a threshold, citing event counts or rates, although follow-up often ends earlier. Building on known shortcomings of event rates and worst-case bounds, we treat the claim as a one-sided alternative and ask which summaries admit a uniformly valid and consistent rule. Under administrative censoring, a claim of low risk can rest on the patients whose follow-up could reach the fixed time, known in advance from enrollment dates, and their events. Event counts can demonstrate a high risk but not a low one. If follow-up differs, cohorts with nondecreasing hazards can share expected numbers of events and person-years while their risks lie on either side of the threshold. When hazards rise and follow-up is short, rules that convert event rates into risks demonstrate false claims more often in larger cohorts, even with the same follow-up. In simulations, the log-log confidence limit of the Kaplan–Meier estimate failed to keep the nominal level with few patients at risk or heavy loss. The beta product confidence procedure kept the nominal level within Monte Carlo error under independent censoring in the scenarios studied, but lost power under heavy loss.
No takes yet. Share an insight, caveat, or question.
Satoshi Nakamura (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: