Presentation (CTI)

OBJECTIVES: Number needed to treat (NNT) is widely used to summarize absolute treatment benefit and is often used in economic evaluations to estimate cost per event avoided and budget impact. When endpoints are time-to-event or recurrent, forcing a single NNT can create unit mismatches (including NNT<1) and inconsistent cost-offset estimates. We assessed how endpoint structure and estimand choice drive these differences and propose a unit-consistent framework and minimum reporting standard for NNT-type metrics.
METHODS: We considered three endpoint structures commonly used in clinical trials and economic models: binary patient outcomes (≥1 event within a fixed horizon), time-to-first event outcomes, and recurrent events summarized as event rates. Absolute treatment effects were expressed as patient-level NNT derived from absolute risk reduction, time-specific NNT(t) derived from survival risk differences, and recurrent-event benefit expressed as events avoided per patient-time. Using an illustrative one-year example, we estimated events avoided under each approach and translated these into cost offsets using per-event costs. Based on these comparisons, we defined minimum reporting elements required for correct interpretation.
RESULTS: For the same underlying treatment effect, estimated events avoided, cost per event avoided, and budget impact differed materially depending on endpoint structure and estimand choice. Applying the traditional NNT formula to recurrent-event rate differences produced values <1, which are not interpretable as “patients needed to treat”. Expressing recurrent outcomes as events avoided per patient-year yielded transparent cost-offset estimates and avoided fractional-patient artifacts. Differences among patient NNT, NNT(t), and recurrent-event metrics reflected distinct analytic questions and are not interchangeable for value assessments.
CONCLUSIONS: NNT should be reserved for patient-level binary outcomes or explicitly time-specific survival contrasts. For recurrent-event endpoints, treatment benefit should be reported as events avoided per patient-time to support valid cost per event avoided and budget impact analyses. A unit-consistent framework and minimum reporting standard can improve transparency, comparability, and interpretability for decision-makers.