Background: The number needed to treat (NNT) is the inverse of the absolute risk reduction (ARR) and represents the estimated average number of patients needed to be treated with a new treatment to prevent one adverse event compared with standard therapy. NNTs are conventionally applied in randomized controlled trials (RCTs) with binary data where the observation time is the same for all patients. In this case, the calculation of NNTs is based on estimated proportions (rates from a 2x2 table). In the situation of RCTs where the outcome is time to an event (survival time data), the naive use of proportions in general is invalid because different follow-up times due to staggered entry and censoring are not taken into account. Two methods have been proposed to calculate the NNT from survival time data. First, the NNT based on risks estimated by the Kaplan-Meier survival curve concerning a specific time point1, and second, the NNT calculated as the reciprocal of the difference of hazards2,3. The latter approach, however, is only adequate if important limitations are taken into account (exponential distribution, rare events)3.
Objective: To give an overview of NNT applications in publications of RCTs where the outcome is time to an event.
Methods: We chose four frequently cited journals for our investigations (BMJ, JAMA, NEJM, and Lancet) and searched for RCTs using survival time techniques in three consecutive publication years (2003-2005). We only included RCTs with parallel group design and individual randomization. We investigated whether NNTs were used to present results, whether the method to calculate NNTs was adequate, and whether confidence intervals for the NNT were reported.
Results: In 2005, we identified 130 published RCTs where the outcome was time to an event. Of these RCTs, 13 (10%) reported NNTs. An adequate method for NNT calculation was used only in 4 of 13 RCTs, mostly because naive proportions were used (8 of 13 RCTs). Confidence intervals for NNTs were only provided in 2 of 13 RCTs. The publications of 2003 and 2004 are currently under investigation.
Conclusions: The NNT is used as an effective measure to present the results of RCTs not only for binary but also for time to event (survival time) data. In the case of survival time data, inadequate methods for NNT calculation are frequently applied (9 of 13 studies). Confidence intervals are rarely provided, even though the CONSORT explanation and elaboration paper4 requests the provision of confidence intervals to indicate the precision of estimates. In conclusion, the current application of NNTs to express results of RCTs where the outcome is time to an event should clearly be improved.

References:
1. Altman D, Andersen PK (1999). Calculating the number needed to treat for trials where the outcome is time to an event. BMJ; 319: 1492-5.
2. Lubsen J, Hoes A, Grobbee D (2000). Implications of trial results: the potentially misleading notions of number needed to treat and average duration of live gained. Lancet; 356:1757-9.
3. Mayne TJ, Whalen E, Vu A (2006). Annualized was found better than absolute risk reduction in the calculation of number needed to treat in chronic conditions. J Clin Epidemiol; 59: 217-23.
4. Altman DG, Schulz KF, Moher D et al. (2001). The revised CONSORT statement for reporting randomized trials: explanation and elaboration. Ann Intern Med; 134: 663-94.