Within ‘evidence-based medicine and health care’ the ‘number needed to treat’ (NNT) has been promoted as the most clinically useful measure of the effectiveness of interventions as established by research. Is the NNT, in either its simple or adjusted form, ‘easily understood’, ‘intuitively meaningful’, ‘clinically useful’ and likely to bring about the substantial improvements in patient care and public health envisaged by those who recommend its use? The key evidence against the NNT is the consistent format effect revealed in studies that present respondents with mathematically-equivalent statements regarding trial results. Problems of understanding aside, trying to overcome the limitations of the simple (major adverse event) NNT by adding an equivalent measure for harm (‘number needed to harm’ NNH) means the NNT loses its key claim to be a single yardstick. Integration of the NNT and NNH, and attempts to take into account the wider consequences of treatment options, can be attempted by either a ‘clinical judgement’ or an analytical route. The former means abandoning the explicit and rigorous transparency urged in evidence-based medicine. The attempt to produce an ‘adjusted’ NNT by an analytical approach has succeeded, but the procedure involves carrying out a prior decision analysis. The calculation of an adjusted NNT from that analysis is a redundant extra step, the only action necessary being comparison of the results for each option and determination of the optimal one. The adjusted NNT has no role in clinical decision-making, defined as requiring patient utilities, because the latter are measurable only on an interval scale and cannot be transformed into a ratio measure (which the adjusted NNT is implied to be). In any case, the NNT always represents the intrusion of population-based reasoning into clinical decision-making.
Evidence-Based Medicine and Clinical Judgment: An Imaginary Divide