The Use of the Number Needed to Treat (NNT) to Assess the Effects of Medicinal Interventions: A Quality Assessment

Abstract

The number needed to treat (NNT) is interpreted as the number of patients needed to treat with one therapy versus another for one patient to encounter an additional outcome of interest over a given period of time1-2. There are few methods to calculate NNTs, which use depends on the study design. This study aims to describe how NNT has been reported in medical literature, and assess the appropriateness of calculating methods according to different study designs and type of variables. Top-25 high impact-factor medical journals were screened to identify studies (2006-2015), assessing medicines and reporting NNTs. Studies were categorized according to their design, and type of variables (binary or time-to-event). NNT estimates were assessed for completeness (baseline risk, time-horizon, and confidence intervals [CI]). The appropriateness of methods used for calculating NNTs was assessed based on published evidence. Data analyses comprehended descriptive statistics. Further the chi-square test was used to test differences between study designs and also type of variables on the likelihood of applying inadequate methods. The search returned 138 citations; 57 were selected. Nearly half were meta-analyses (49.1%), followed by clinical trials (29.9%), cohort (17.5%) and case-control studies (3.5%). Binary variables were more common (82.5%) than time-to-event (17.5%) outcomes. Baseline risk (64.9%), time-horizon (68.4%) and CI (57.9%) for NNT were not always reported. Overall, 29% of studies applied inadequate methods for calculating NNTs; this proportion was higher in meta-analyses (54%) as compared to other research designs (P=0.003). No differences were found between type of variables and appropriateness of methods (p=0.972). A considerable proportion of studies, particularly meta-analyses, applied inadequate methods for calculating NNTs. Despite their usefulness in assisting clinical decisions, NNTs are uninterpretable if incompletely reported, and may be misleading if calculating methods are inadequate to study designs and variables under evaluation.