Article type
Year
Abstract
Introduction/Objective: Meta-analyses of randomized controlled trials traditionally summarise treatment effects using the relative risk or odds ratio. Currently available software also calculates pooled estimates of numbers needed to treat and increasingly, these are provided as summary measures of treatment effects. We believe that these estimates of treatment effects may be seriously misleading.
Methods: Primary and secondary prevention meta-analyses for interventions commonly used in coronary heart disease (CHD) were examined using relative risk reductions and NNTs. Comparisons were made between NNTs for trials carried out at different times, in different settings and in different places. Pooled relative risk reduction estimates for a range of interventions were applied to a range of baseline cardiovascular disease risks to calculate NNTs.
Results: NNTs were not constant between age groups, men and women, between places and over time and were consequently much less stable than relative risk reduction estimates. Trials carried out in the early 1980s experienced 50% higher background rates of CHD than rates that currently prevail, so in these circumstances, NNTs were very markedly affected by the outcomes included, with small NNTs for "all bad things that could happen". Some interventions were given constantly (eg. antihypertensive drugs), but others were given only once (eg. smoking cessation advice, dietary advice), producing NNTs that were not directly comparable. Calculating pooled NNTs for trials with different follow up periods required an assumption of constant treatment effects over time to permit presentation of NNT for a fixed time period. Confidence intervals of cooled NNTs were wide.
Discussion: A pooled NNT is less useful for clinical management than individual trial NNTs where characteristics and baseline risk of patients can be examined. NNTs derived from trials are unlikely to be generalisable to clinical practice as placebo group event rates are typically lower than expected. It would be preferable to confine meta-analysis summary estimates to relative risk reductions, and then use illustrative examples of the NNTs obtained when applying data to high, intermediate and low risk populations.
Methods: Primary and secondary prevention meta-analyses for interventions commonly used in coronary heart disease (CHD) were examined using relative risk reductions and NNTs. Comparisons were made between NNTs for trials carried out at different times, in different settings and in different places. Pooled relative risk reduction estimates for a range of interventions were applied to a range of baseline cardiovascular disease risks to calculate NNTs.
Results: NNTs were not constant between age groups, men and women, between places and over time and were consequently much less stable than relative risk reduction estimates. Trials carried out in the early 1980s experienced 50% higher background rates of CHD than rates that currently prevail, so in these circumstances, NNTs were very markedly affected by the outcomes included, with small NNTs for "all bad things that could happen". Some interventions were given constantly (eg. antihypertensive drugs), but others were given only once (eg. smoking cessation advice, dietary advice), producing NNTs that were not directly comparable. Calculating pooled NNTs for trials with different follow up periods required an assumption of constant treatment effects over time to permit presentation of NNT for a fixed time period. Confidence intervals of cooled NNTs were wide.
Discussion: A pooled NNT is less useful for clinical management than individual trial NNTs where characteristics and baseline risk of patients can be examined. NNTs derived from trials are unlikely to be generalisable to clinical practice as placebo group event rates are typically lower than expected. It would be preferable to confine meta-analysis summary estimates to relative risk reductions, and then use illustrative examples of the NNTs obtained when applying data to high, intermediate and low risk populations.