[Number needed to treat: Interpretation and estimation in multivariable analyses and censored data]

Abstract

Number needed to treat has been recommended as an easy way to transmit results from a trial, especially controlled clinical trials. Most articles estimate it from a 2×2 table, as the inverse of the absolute risk reduction. However, some limitations have been pointed out: The interpretation is not as easy as claimed, confidence intervals are frequently not estimated, and the estimation from 2×2 tables is inadequate when the main effect measure has been estimated adjusting for confounding factors. In this paper, we revise how to obtain point estimations and confidence intervals of number needed to treat in 4 situations: 2×2tables, logistic regression, Kaplan-Meier method, and Cox regression.

Keywords: Absolut risk reduction; Controlled clinical trials; Cox regression; Ensayos clínicos; Epidemiologic method; Estimador de Kaplan-Meier; Estudios epidemiológicos; Kaplan-Meier estimation; Logistic regression; Number needed to treat; Número necesario de tratamientos; Reducción absoluta de riesgos; Regresión de Cox; Regresión logística.