gms | German Medical Science

Introduction: In medical practice, physicians often face situations where various therapy options exist. Ideally, all these therapies were previously compared in one (or several) trials. However, often only two- (or even one-) arm trials were conducted comparing only a subset of all possible therapies. In this situation, the question arises whether and how reliable and valid conclusions on the choice of the best treatment option can be drawn. NaÏvely combining the results from different trials can lead to severe bias due to cross-trial differences in baseline characteristics. Therefore, valid methods for so-called indirect comparisons are needed.

Methods: We investigate the method of Bucher [1] and the matching-adjusted indirect comparison (MAIC) method [2]. The method of Bucher preserves the within-study randomization and needs a common comparator. A problem may be the insufficient comparability of studies according to patient characteristics, comparator therapies, and the adjustment of effect estimates. The MAIC approach is based on a matching procedure that selects a weight for each patient and thus adjusts individual patient data with respect to baseline characteristics. After that, the method of Bucher is applied to the weighted data. However, individual patient data needs to be available for at least one trial to conduct an indirect comparison by MAIC. We performed a simulation study for a wide range of practically relevant scenarios aiming to investigate the above described methods for indirect comparisons for time-to-event and binary endpoints. We assess and compare the statistical properties of the methods, including bias in the estimated therapy effects, type I error rates, and power.

Results: Bucher as well as MAIC show a negligible bias when no interactions are present and the effect estimates are adjusted for baseline characteristics. We identified situations were differences between the methods became apparent. Our investigations showed that the availability of adjusted or unadjusted hazard ratios, as well as the presence of interactions between baseline characteristics and treatment selection influence the extent of the bias. MAIC shows a better performance in situations when differing patient characteristics are observed, and additionally when only unadjusted effect estimated are available. In case interactions between therapy and patient characteristics are present, the MAIC approach leads to less biased effect estimates.

Discussion: Indirect comparisons allow for estimation of therapy effects when no studies comparing these therapies directly are available. An important step prior to conducting an indirect comparison is the identification of the possible underlying differences between the trials. Based on this knowledge, the method for the indirect comparison should be chosen carefully to avoid bias. In the case of equal patient characteristics and adjusted effect estimates Bucher, has the advantage of preserving the within-study randomization.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.