Testing the Hypothesis that Matters for Multiple Primary Endpoints

Abstract

In clinical trials with multiple primary endpoints, the effectiveness of a treatment may be established using either a composite endpoint or by employing a decision rule based on the individual endpoints. In this papec the latter method is examined. Emphasis is placed on the careful selection of primary endpoints and the need to specib the clinical objectives and statistical decision rules in advance. It is argued that Bonferroni-type adjustments are appropriate only in the case of a nonspecific hypothesis. Since requiring an intervention to demonstrate positive effects in several primary endpoints is more stringent than requiring an effect in any, some upward adjustment of individual a-levels should be allowed. In addition, physiologidsymptomatic classification of endpoints with separate decision rules for each may also be warranted in certain circumstances. An example is presented.

A decision rule for sequential monitoring of clinical trials with a primary and supportive outcome

“…For low to moderate correlation, critical values based on the OBF error spending function are lower than those based on univariate boundaries that do not consider the correlation. Our finding that critical values vary substantially according to the correlationincreasing with increasing correlation for 'and' rules -is consistent with fixed sample results reported by Capizzi and Zhang [18]. An advantage of the approach considered here compared to ad hoc methods or more general guidelines is that trial designers can discuss and then determine the degree of 'support' required for the secondary endpoint and plan monitoring accordingly.…”

Vascular Biomarkers and Surrogates in Cardiovascular Disease

“…Nevertheless, it is critically important to study biomarkers (for example, IVUS) and clinical end points in the same study to evaluate their correlation in pharmacological intervention trials. 42 This approach represents indeed the best strategy to validate a biomarker. There may be instances where clinical events influence the analysis of the surrogate.…”

Applied Statistics in Biomedicine and Clinical Trials Design

“…The more variables that are chosen, i.e., the greater the value of K, the more serious this problem will become. Hence, K should be no larger than necessary and the outcome variables should not be highly correlated with each other as well as presumably reflecting a treatment effect (Capizzi and Zhang 1996).…”