In our last blog we highlighted the importance of exploring covariate effects in NMA models. But what about unknown risk factors that aren’t captured in reported patient characteristics? Exploring the baseline risk in each trial population using meta-regression methods may address the impact of potential unknown risk factors in networks where trials include a common treatment arm (such as placebo). The baseline risk is not a measurable quantity and clinicians only know about underlying risk through measurable patient characteristics (1). Baseline risk in the context of meta‑regression is usually defined as the outcome event rate in the common comparator treatment arm (e.g. outcome event rate in the placebo arm). This represents a potentially important source of heterogeneity, particularly among studies where the baseline risk varies.

Meta-regression on baseline risk requires special consideration compared with using known risk factor covariates. This is because a standard regression model will ignore regression dilution bias and causes the covariate association to be overestimated (2, 3). Therefore, when including baseline risk as a covariate, the regression model is modified to include trial‑specific baseline risk estimated by the NMA model (rather than using observed data and ignoring uncertainty in its estimation). The trial-specific baseline risk is used as the covariate, taking into account the uncertainty in each baseline (2).

We believe that our recently published paper (in collaboration with Leicester University), is the first analysis to explore the impact of baseline risk on a clinically important efficacy outcome for stroke prevention in atrial fibrillation (4). More generally, and regardless of indication, in our experience the exploration of baseline risk is uncommon in NMA publications. We recommend the consideration of baseline risk as a proxy or surrogate for multiple (and potentially unmeasured) treatment confounding effects when exploring heterogeneity within NMA evidence networks.

For further information on any aspects of SR and NMA please contact our in-house experts.


  1. Thompson SG, Smith TC, Sharp SJ. Investigating underlying risk as a source of heterogeneity in meta-analysis. Stat Med. 1997;16(23):2741-58.
  2. Dias S, Welton, N.J., Sutton, A.J., Valdwell,D.M., Guobing, L., & Ades, A.E. NICE DSU Technical Support Document 3: Heterogeneity:subgroups, meta-regression, bias and bias-adjustment. wwwnicedsuorguk. 2011.
  3. Welton NJ, Sutton AJ, Cooper NJ, Abrams KR, Ades AE. Evidence synthesis for decision making in healthcare. 1 ed: A John Wiley and Sons, Ltd; 2012. 282 p.
  4. Batson S, Sutton A, K A. Exploratory Network Meta Regression Analysis of Stroke Prevention in Atrial Fibrillation Fails to Identify Any Interactions with Treatment Effect. PLoS One. 2016;11(8):e0161864.


DRG becomes Clarivate

View Now