BACKGROUND AND OBJECTIVE: Whether the number needed to treat (NNT) is sufficiently precise to use in clinical practice remains unclear. We compared unadjusted NNTs to quality-adjusted life years (QALYs) gained, a more comprehensive measures of health benefit. STUDY DESIGN AND SETTING: From a subset (n = 65) of a dataset of 228 cost-effectiveness analyses, we compared how well NNTs predicted clinically important QALY gains using correlation analysis, multivariable models and receiver-operator curve (ROC) analysis. RESULTS: NNT was inversely correlated with QALY gains (P < .001); this relationship was affected by quality of life and life-expectancy gains of treatment (P <or= .04). The NNT is a moderately accurate predictor of treatments that provide large health benefits (area under ROC 0.74-0.81). For ruling out therapies with low QALY gains (threshold <or=0.125 to <or=0.5 QALYs), an NNT >15 had a sensitivity of 82% to 100%. For ruling in therapies with high QALY gains (threshold >or=0.125 to >or=0.5 QALYs), an NNT <or=5 had a specificity of 77%. CONCLUSION: Using NNT thresholds of <or=5 and >15 to rule in and out therapies with large QALY gains may provide general guidance regarding the magnitude of health benefit.
BACKGROUND AND OBJECTIVE: Whether the number needed to treat (NNT) is sufficiently precise to use in clinical practice remains unclear. We compared unadjusted NNTs to quality-adjusted life years (QALYs) gained, a more comprehensive measures of health benefit. STUDY DESIGN AND SETTING: From a subset (n = 65) of a dataset of 228 cost-effectiveness analyses, we compared how well NNTs predicted clinically important QALY gains using correlation analysis, multivariable models and receiver-operator curve (ROC) analysis. RESULTS: NNT was inversely correlated with QALY gains (P < .001); this relationship was affected by quality of life and life-expectancy gains of treatment (P <or= .04). The NNT is a moderately accurate predictor of treatments that provide large health benefits (area under ROC 0.74-0.81). For ruling out therapies with low QALY gains (threshold <or=0.125 to <or=0.5 QALYs), an NNT >15 had a sensitivity of 82% to 100%. For ruling in therapies with high QALY gains (threshold >or=0.125 to >or=0.5 QALYs), an NNT <or=5 had a specificity of 77%. CONCLUSION: Using NNT thresholds of <or=5 and >15 to rule in and out therapies with large QALY gains may provide general guidance regarding the magnitude of health benefit.