Content area
Full Text
With an increasing demand for comparative effectiveness research (CER), there is often interest in comparing alternative treatments that have not been directly compared in a head-to-head randomized trial. For example, one may wish to compare drug A with drug B when randomized trials have only compared each with drug C (e.g. placebo). To make a reliable indirect comparison of drug A versus drug B across the separate trials, it is desirable that the trials be as similar as possible.
The most widely used methods for indirect comparison rely exclusively on aggregate data, such as published trial results, and use common comparator arms (e.g. drug C) to adjust for any differences between trials. For example, if one trial provides the relative response rate of drug A versus drug C (RR A:C) and another trial provides the relative response rate of drug B versus drug C (RRB:C), the relative response rate of drug A versus drug B is estimated as RRA:C/RRB:C. This approach, introduced by Bucher et al.,[1] can incorporate multiple trials per treatment, and has been extended to simultaneously compare multiple treatments linked through a network of indirect comparisons,[2] and to adjust for aggregate baseline characteristics in a meta-regression framework.[3-5] Several comprehensive and detailed reviews of these and other methods for indirect treatment comparisons have been published.[5,6]
Although aggregate data are widely available, indirect comparisons that rely exclusively on aggregate data are subject to well appreciated limitations, including the potential for bias due to differences between trial populations.[1,5-9] For example, if baseline disease severity was greater for patients in the trial of drug A versus C compared with the trial of drug B versus C, an indirect comparison of drug A versus B across trials could be biased, even with treatment effects compared relative to drug C. Table I provides an extreme hypothetical example of such bias. An adjusted indirect comparison based on aggregate data, in the upper portion of table I, suggests that drug A is inferior to drug B, with a lower response rate by 10 percentage points. However, when individual patient data (IPD) are used to stratify the comparison by baseline disease severity, as in the lower portion of table I, drug A is revealed...