Article type
Year
Abstract
Objectives: To assess the results of using seven different methods to calculate the Number Needed to Treat (NNT)from the pooled results of a convenience sample of 20 outcomes from Cochrane systematic reviews.
Methods: Point estimates and Confidence Interval of NNT for 20 outcomes of systematic reviews published on the Cochrane Library were calculated using Pooled Relative Risk Reduction (RRR), and Pooled Risk Difference (RD) with both fixed and random effects models. The lumped totals and Peto Odds Ratio were also used as a comparative methods.
Results: The raw overall lumped totals gave a broadly similar NNT to the other methods except in one review in which the control and treatment group sizes were not all equal. The CI for the lumped result was very similar to the pooled RD fixed effect CI, except where there are trials with zero outcomes in both arms. In most cases there was more heterogeneity in the Risk Differences than in the Relative Risk Reduction, supporting the view that Relative Risk Reduction is more stable across trials. The Control Event Rate also showed heterogeneity in some reviews which implies difficulty with using a Pooled Risk Difference to calculate NNT. Using a Pooled Relative Risk Reduction has the advantage that the reader of the review can apply this result (with its Confidence Interval) to the CER which best reflects the patients under consideration in their own Clinical Practice. Pooled OR used with the pooled Control Event Rate tends gives an overoptimistic estimate of NNT when there are large effects
Discussion: Whilst in many cases the point estimate and Confidence Interval of the NNT were similar using all the methods studied, application of the Pooled Relative Risk Reduction to an appropriate Control Event Rate may be most useful to readers wishing to apply the results of systematic reviews.
Methods: Point estimates and Confidence Interval of NNT for 20 outcomes of systematic reviews published on the Cochrane Library were calculated using Pooled Relative Risk Reduction (RRR), and Pooled Risk Difference (RD) with both fixed and random effects models. The lumped totals and Peto Odds Ratio were also used as a comparative methods.
Results: The raw overall lumped totals gave a broadly similar NNT to the other methods except in one review in which the control and treatment group sizes were not all equal. The CI for the lumped result was very similar to the pooled RD fixed effect CI, except where there are trials with zero outcomes in both arms. In most cases there was more heterogeneity in the Risk Differences than in the Relative Risk Reduction, supporting the view that Relative Risk Reduction is more stable across trials. The Control Event Rate also showed heterogeneity in some reviews which implies difficulty with using a Pooled Risk Difference to calculate NNT. Using a Pooled Relative Risk Reduction has the advantage that the reader of the review can apply this result (with its Confidence Interval) to the CER which best reflects the patients under consideration in their own Clinical Practice. Pooled OR used with the pooled Control Event Rate tends gives an overoptimistic estimate of NNT when there are large effects
Discussion: Whilst in many cases the point estimate and Confidence Interval of the NNT were similar using all the methods studied, application of the Pooled Relative Risk Reduction to an appropriate Control Event Rate may be most useful to readers wishing to apply the results of systematic reviews.