RTI uses cookies to offer you the best experience online. By clicking “accept” on this website, you opt in and you agree to the use of cookies. If you would like to know more about how RTI uses cookies and how to manage them please view our Privacy Policy here. You can “opt out” or change your mind by visiting: http://optout.aboutads.info/. Click “accept” to agree.
Evaluation of matching-adjusted indirect comparison implemented by a resampling method
Wang, J., Odom, D., Chirila, C., & Zheng, Q. (2015). Evaluation of matching-adjusted indirect comparison implemented by a resampling method. Value in Health, 18(3), A31. https://doi.org/10.1016/j.jval.2015.03.186
Objectives Matching-adjusted indirect comparison (MAIC) has been proposed as a new tool when conducting indirect treatment comparisons in the situation where individual-level data are available from one study, but only summary data are available from another study. This study evaluated the performance of the MAIC method proposed by Malangone and Sherman (2011) which is implemented by a resampling (bootstrapping) technique.
Methods Two patient-level data sets, similar to two clinical trials, were generated: the first with treatments A and placebo, and the second with treatments B and placebo. Other variables included in both data sets were survival time, censoring indicator, and two baseline categorical variables. In both data sets, interactions between baseline characteristics and treatments were incorporated such that differential treatment effects across baseline strata were present. The SAS programs illustrated in Malangone and Sherman were adopted for the MAIC analysis. First, MAIC was applied to a situation in which only summary data were available from the first data set and individual-level data were available from the second data set. Subsequently, the roles of two data sets were switched and the MAIC analysis was applied once again.
Results Using MAIC, when the first data set provided summary statistics, the hazard ratio (HR) (95% confidence interval [CI]) for A versus placebo was 0.283 (0.246-0.325); the HR (95% CI) for B versus placebo was 0.586 (0.466, 0.740). When the second data set provided summary data, the HR (95% CI) for A versus placebo was 0.489 (0.390-0.612) and for B versus placebo was 0.237 (0.205-0.273). The two comparisons produced opposite significant inferences.
Conclusions The method proposed by Malangone and Sherman is an interesting addition to the MAIC field, but results could be misleading under some circumstances. Therefore, the conditions under which this method is suitable should be explored further.
