Presentation (CTI)

OBJECTIVES: Event-based number needed to treat (NNT) is increasingly used in health economic and outcomes research (HEOR) to summarize treatment effects when patients experience recurrent events. However, event-based NNT estimates can be <1 and are often reported without confidence intervals (CIs), making uncertainty difficult to interpret for cost-offset and budget impact analyses. We evaluated practical approaches for estimating and reporting CIs for event-based NNT and propose recommendations to support transparent economic decision-making.
METHODS: Event-based NNT was defined as the inverse of the difference in mean event rates between comparator scenarios over a fixed time horizon. Two approaches to uncertainty estimation were evaluated including a delta-method approximation based on the variance of the event-rate difference and non-parametric bootstrapping of model-generated event counts. Using simulated cohorts with recurrent events, we examined CI behavior across varying event frequencies, effect sizes, and follow-up durations. Performance was evaluated in terms of CI stability, interpretability, and frequency of sign changes around the null.
RESULTS: Event-based NNT point estimates were frequently <1 when multiple events were prevented per patient, yielding asymmetric and sometimes unbounded CIs. Delta-method CIs performed adequately when event-rate differences were large and stable but became unreliable near the null. Bootstrap-based CIs were more robust across scenarios and better captured skewness and sign uncertainty. Reporting the CI for the underlying event-rate difference alongside the event-based NNT improved interpretability, particularly when NNT estimates were extreme or unstable.
CONCLUSIONS: CIs for event-based NNTs are feasible and informative but require careful methodological handling. For health economic models with recurrent events, we recommend reporting CIs using bootstrap methods where possible, prioritizing rate-difference CIs when NNT estimates are unstable, and explicitly stating the time horizon and recurrent-event nature of outcomes. Adoption of standardized CI reporting for event-based NNT would improve transparency and decision-making in health economic models.