Article type
Year
Abstract
Background: Often, it is not sufficient to prove statistical significance of effects. Rather, the clinical relevance of effects has to be shown. Usually, clinical trials are powered for statistical significance only. Within the framework of systematic reviews the power (precision) is increased by pooling results of multiple trials. Therefore, more stringent conclusions towards clinical relevance can be drawn without loss of power.
Objectives: To develop an approach to prove clinical relevance in meta-analyses of binary data.
Methods: The starting point is a specified true effect that is considered clinically relevant. Also in order to prove clinical relevance, the probability of statistical errors should be controlled. Therefore, the precision of the estimates has to be considered. This is fulfilled by determining a threshold for the confidence interval (CI) obtained by meta-analysis which corresponds to the testing of shifted hypotheses. This threshold has to be precisely selected so that a specified power is preserved. The main idea is that the power of the meta-analysis should be identical to the power for proving statistical significance in the individual studies included in the meta-analysis. Balanced sample sizes are assumed. This approach will be investigated for the effect measures relative risk (RR) and odds ratio (OR).
Results: Simulations show that the resulting thresholds depend on the baseline risk to some extent. The OR is substantially more severely affected than the RR. Therefore, the RR is favoured. In addition, the baseline risk dependency nearly vanishes if using the restricted maximum likelihood approach (REML) instead of the Wald approach for the calculation of the CIs.
Conclusions: Clinical relevance can be shown without loss of power by taking advantage of the higher precision of meta-analyses. The thresholds for clinical relevance are nearly independent of the baseline risk if the REML-approach is applied. Thus, the determination of these thresholds only depends on the higher precision gained by meta-analyses.
Objectives: To develop an approach to prove clinical relevance in meta-analyses of binary data.
Methods: The starting point is a specified true effect that is considered clinically relevant. Also in order to prove clinical relevance, the probability of statistical errors should be controlled. Therefore, the precision of the estimates has to be considered. This is fulfilled by determining a threshold for the confidence interval (CI) obtained by meta-analysis which corresponds to the testing of shifted hypotheses. This threshold has to be precisely selected so that a specified power is preserved. The main idea is that the power of the meta-analysis should be identical to the power for proving statistical significance in the individual studies included in the meta-analysis. Balanced sample sizes are assumed. This approach will be investigated for the effect measures relative risk (RR) and odds ratio (OR).
Results: Simulations show that the resulting thresholds depend on the baseline risk to some extent. The OR is substantially more severely affected than the RR. Therefore, the RR is favoured. In addition, the baseline risk dependency nearly vanishes if using the restricted maximum likelihood approach (REML) instead of the Wald approach for the calculation of the CIs.
Conclusions: Clinical relevance can be shown without loss of power by taking advantage of the higher precision of meta-analyses. The thresholds for clinical relevance are nearly independent of the baseline risk if the REML-approach is applied. Thus, the determination of these thresholds only depends on the higher precision gained by meta-analyses.