Background: Covariate-adjusted ITC (CA-ITC) is a comparison approach using trial-level covariates within an ITC adjustment to mediate the effect of between-trial differences. However, guidance is lacking on when (and how) CA-ITC should be utilised, often resulting in over-relying on, or incorrect/absent use of this method.

Aim: This work aims to summarise the key considerations for utilising CA-ITC. Key considerations include: a) Study selection: determining which studies to include in a network is the first decision to make, and should be based on a transitivity assessment (Tremblay 2017). It should be understood that covariate adjustment cannot fully correct for between-trial differences, so including widely differing studies will pose a risk of biased results, regardless of covariate adjustment; b) Sufficient data: The variable(s) being adjusted for should be reported granularly and similarly in all included studies. Additionally, an adequate number of studies should be included to provide sufficient statistical power for the adjustment; c) Covariate interaction: the final step is to determine whether the adjustment should assume the same covariate interaction across all treatments (i.e., should the CA-ITC produce a single result that accounts for that covariate), or if it should vary between treatments (i.e., should the CA-ITC produce stratified results for that covariate). The former approach will generally require less observations to obtain sufficient statistical power, but in some cases assuming the same interaction effects may not be appropriate (Donegan 2017).

Conclusions: CA-ITC can be a powerful statistical tool in comparative research. However, if the key considerations cannot be satisfied, then CA-ITC should be avoided. While the theoretic CA-ITC requirements may seem clear, there is a significant deficiency in objective guidelines defining when its use is most appropriate, and when other comparison methods should be used (for example, unadjusted Bayesian ITC, matching-adjusted or simulated-trial comparisons when individual patient-level data are available).