Interpretation of absolute measures of disease risk in comparative research

Abstract

When comparing two groups, one receiving an experimental intervention and the other a placebo or nothing, researchers often wish to assess the disparity in risk of experiencing an event of interest, such as onset of disease. Relative risk, relative risk reduction, and odds ratio are often used to measure the association between potential benefit or harm and the intervention. However, these summary measures reflect relative disparities and are perhaps less useful in clinical practice than measures of absolute benefit or harm. We demonstrate that relative risk reduction is unaffected by the risk of an event in the control group and hence may either overestimate or underestimate the treatment effect. Absolute risk reduction accounts for the baseline control group event rate and is a more realistic quantification of treatment effect than relative measures. Number needed to treat (NNT) estimates the therapeutic effort needed to prevent one additional adverse event. NNT incorporates both relative risk reduction and the event rate without treatment. For a given relative risk reduction, we demonstrate the NNT will increase as the control event rate decreases. Thus, NNT has more-obvious implications for clinical decision making than risk estimates expressed in relative terms.