Article type
Year
Abstract
Background: Meta-analyses utilize several indices of treatment effectiveness to summarize their results, such as odds ratio (OR), risk ratio (RR), risk difference (RD) and/or number needed to treat (NNT). In applying the results of a meta-analysis to individual patients, some textbooks of evidence-based medicine assume the constancy of RR but no study to date has examined this empirically. We will therefore examine the generalizability of several indices of treatment effectiveness for meta-analyses empirically.
Methods: We selected a random subset of meta-analyses contained in a recent issue of the Cochrane Library. When a meta-analysis pooled more than three RCTs to produce a summary measure for an outcome, we compared the OR, RR and RD of each of the included RCTs with the OR, RR and RD, respectively, meta-analytically pooled from all the other RCTs. The main outcome measure was the percentage of pairs of OR, RR or RD where they were not statistically significantly different at the conventional p value of 0.05.
Results: We made 1, 843 comparisons, extracted from 55 meta-analyses. The random effects model OR had the highest concordance rate, closely followed by the random effects model RR, and these were applicable to individual RCTs dealing with comparable patient-intervention-outcome combinations approximately 90% of the time, even when the baseline risk differed substantially.
Conclusions: Coupled with the ease of interpretation of RR, the present findings suggest that clinicians may wish to rely on the random effects model RR and its control-event-rate-adjusted NNT when they apply the results of a meta-analysis in their practice.
Methods: We selected a random subset of meta-analyses contained in a recent issue of the Cochrane Library. When a meta-analysis pooled more than three RCTs to produce a summary measure for an outcome, we compared the OR, RR and RD of each of the included RCTs with the OR, RR and RD, respectively, meta-analytically pooled from all the other RCTs. The main outcome measure was the percentage of pairs of OR, RR or RD where they were not statistically significantly different at the conventional p value of 0.05.
Results: We made 1, 843 comparisons, extracted from 55 meta-analyses. The random effects model OR had the highest concordance rate, closely followed by the random effects model RR, and these were applicable to individual RCTs dealing with comparable patient-intervention-outcome combinations approximately 90% of the time, even when the baseline risk differed substantially.
Conclusions: Coupled with the ease of interpretation of RR, the present findings suggest that clinicians may wish to rely on the random effects model RR and its control-event-rate-adjusted NNT when they apply the results of a meta-analysis in their practice.