Tracy J Mayne1, Edward Whalen, An Vu. 1. Outcomes Research, Pfizer Pharmaceuticals, 235 East 42nd Street, 205/9/10, New York, NY 10017, USA. Tracy.Mayne@Pfizer.com
Abstract
BACKGROUND AND OBJECTIVE: Recent studies have calculated number needed to treat (NNT) estimates based on annualized rates; however, the ramifications of altering the NNT statistic have not yet been explored in the literature. Here we introduce the concept of annualized NNT (ANNT), and apply it to data from randomized controlled trials (RCTs). METHODS: Incidence rates from RCTs for serious adverse events for three medicines were compared to an older class of drugs. NNT and ANNT were calculated from the event rates for these events. RESULTS: Based on the data, the NNT to prevent one adverse event a year vs. older medications was drug A, ANNT = 88; drug B, ANNT = 77; drug C, ANNT = 68. Equivalent calculations based on Bayesian statistics are drug C, ANNT = 54; drug B, ANNT = 49. Drug A produced a bimodal distribution, with one mode within the NNT range and the other in the number needed to harm range. CONCLUSIONS: NNT can erroneously inflate differences between treatments when based on absolute and not differential safety. We propose that NNT be limited to acute conditions with short-term, well-defined treatment courses, and that ANNT be used for chronic conditions.
BACKGROUND AND OBJECTIVE: Recent studies have calculated number needed to treat (NNT) estimates based on annualized rates; however, the ramifications of altering the NNT statistic have not yet been explored in the literature. Here we introduce the concept of annualized NNT (ANNT), and apply it to data from randomized controlled trials (RCTs). METHODS: Incidence rates from RCTs for serious adverse events for three medicines were compared to an older class of drugs. NNT and ANNT were calculated from the event rates for these events. RESULTS: Based on the data, the NNT to prevent one adverse event a year vs. older medications was drug A, ANNT = 88; drug B, ANNT = 77; drug C, ANNT = 68. Equivalent calculations based on Bayesian statistics are drug C, ANNT = 54; drug B, ANNT = 49. Drug A produced a bimodal distribution, with one mode within the NNT range and the other in the number needed to harm range. CONCLUSIONS: NNT can erroneously inflate differences between treatments when based on absolute and not differential safety. We propose that NNT be limited to acute conditions with short-term, well-defined treatment courses, and that ANNT be used for chronic conditions.
Authors: Stacy Loeb; Edward F Vonesh; E Jeffrey Metter; H Ballentine Carter; Peter H Gann; William J Catalona Journal: J Clin Oncol Date: 2010-12-28 Impact factor: 44.544
Authors: Zhao Lan Liu; Jian Ping Liu; Anthony Lin Zhang; Qiong Wu; Yao Ruan; George Lewith; Denise Visconte Journal: Cochrane Database Syst Rev Date: 2011-07-06
Authors: Pim A de Jong; Martijn J A Gondrie; Constantinus F M Buckens; Peter C Jacobs; Willem P T h M Mali; Yolanda van der Graaf Journal: PLoS One Date: 2011-10-12 Impact factor: 3.240
Authors: Zoe N Memel; Ashwini Arvind; Oluwatoba Moninuola; Lisa Philpotts; Raymond T Chung; Kathleen E Corey; Tracey G Simon Journal: Hepatol Commun Date: 2020-11-13