Estimating and Testing an Index of Bias Attributable to Composite Outcomes in Comparative Studies

Abstract

Background: Little guidance is available on how composite outcomes should be interpreted, especially in situations of varied direction in the association across the event subtypes. I proposed an index to evaluate the bias attributable to composite outcomes (BACO) and applied it in recently published clinical trials.Methods: I defined the BACO index as the ratio between logarithms of the association measures of both a composite outcome and its most relevant component (e.g., any-cause mortality). By using the non-linear combination of parameters, based on the delta method, I calculated the confidence intervals and performed Wald-type tests for the null hypotheses (BACO index = 1). I applied this method in systematically selected clinical trials, and in two other preselected trials which I considered "positive controls". These last trials have been recognized as examples of primary composite outcomes that were disregarded because of inconsistency with the treatment effect on mortality.Results: BACO index values different from one were classified according to whether the use of composite outcomes overestimated (BACO index >1), underestimated (BACO index between zero and <1), or inverted (BACO index <0) the association between exposure and prognosis. In three of 23 clinical trials and the two positive controls, the BACO indices were significantly lower than one (using p <0.005 as a preset cutoff).Conclusion: Based on the BACO index testing, researchers could predefined rules to make impartial decisions about maintaining a composite outcome as the primary endpoint or to state cautions regarding its interpretation.