Article type
Year
Abstract
Background: Randomised control trials may include relevant treatment arms for a systematic review but only present data comparing these treatments to other treatment arms not relevant for the review. To be able to use the existing data, indirect comparisons have to be made.
Objectives: To develop a method to calculate the treatment effect and its variance between two treatment arms indirectly from other reported treatment arm comparisons.
Methods: In a Cochrane review comparing inhaled corticosteroids (ICS) alone to long-acting beta2-agonists (LABA) alone for treatment of COPD, none of the included trials reported the results of direct comparisons between these treatments. However, these treatment arms were compared to the treatment effect of ICS/LABA combination treatment and/or placebo. Indirect estimates of treatment effects could be calculated from trials comparing e.g. ICS versus placebo and LABA versus placebo. Dichotomous data were then obtained as the log risk ratio (LRR) for ICS versus LABA. LRR (ICS versus LABA) = LRR (ICS versus placebo) - LRR (LABA versus placebo). The standard error (SE) for ICS versus LABA could be calculated, using the additive characteristic of independent variances, when SE, exact P values or confidence intervals (CI) were available for the indirect comparisons.
Results: The assumptions of the indirect calculations were found to be validated using the method in trials which reported results of direct comparisons.
Conclusions: Treatment effects with standard errors can be calculated from multiple-arm trials which include relevant treatment arms for inclusion in a systematic review, but do not report direct comparison data.
Objectives: To develop a method to calculate the treatment effect and its variance between two treatment arms indirectly from other reported treatment arm comparisons.
Methods: In a Cochrane review comparing inhaled corticosteroids (ICS) alone to long-acting beta2-agonists (LABA) alone for treatment of COPD, none of the included trials reported the results of direct comparisons between these treatments. However, these treatment arms were compared to the treatment effect of ICS/LABA combination treatment and/or placebo. Indirect estimates of treatment effects could be calculated from trials comparing e.g. ICS versus placebo and LABA versus placebo. Dichotomous data were then obtained as the log risk ratio (LRR) for ICS versus LABA. LRR (ICS versus LABA) = LRR (ICS versus placebo) - LRR (LABA versus placebo). The standard error (SE) for ICS versus LABA could be calculated, using the additive characteristic of independent variances, when SE, exact P values or confidence intervals (CI) were available for the indirect comparisons.
Results: The assumptions of the indirect calculations were found to be validated using the method in trials which reported results of direct comparisons.
Conclusions: Treatment effects with standard errors can be calculated from multiple-arm trials which include relevant treatment arms for inclusion in a systematic review, but do not report direct comparison data.