Network Meta-Analysis, ML-NMA, MAICs and STCs

Network Meta-Analysis

Fully compliant with NICE Technical Support Document guidelines we offer a full range of Bayesian analytical services supplemented with Frequentist techniques:

  • Fixed & Random Effects Models for a host of different endpoints
  • Appropriate checks for consistency, heterogeneity & sensitivity of results to priors
  • Specialised models such as fractional polynomial survival modelling of patient level data (we digitalise published Kaplan-Meier curves to create such data).
  • Meta-regression techniques to account for treatment effect modifiers
  • All results we can program into an Excel HE models aimed at HTA submissions enabling probabilistic sensitivity analysis to be conducted.
  • Reports including completing relevant sections in HTA submissions and manuscripts
  • Full range of analysis outputs offered – absolute and relative differences, rankings, SUCRA, probability of non-inferiority/superiority (can be viewed for different thresholds) and much more.

Multilevel Network Meta-Analysis

  • Newest technique that offers advantages over all other methods in the joint modelling of both patient level (client’s own trial data) and aggregate (typically summary trial manuscript data).
  • Technique avoids aggregation/ecological bias by explaining within-trial variation within the aggregate data and typically increases the precision of estimates.
  • Superior in theory to MAICs and STCs in anchored (common trial comparator) comparisons.
  • By far the most complex method and may not always be appropriate dependent on intended audience.

Matched Adjusted Indirect Comparisons for Unconnected Networks

Meets all requirements of NICE TSD 18 and goes beyond it:

  • Ability to impose sensible cross-correlations (almost certainly missing from aggregate data supplying means/variances) in the algorithm that derives patient weights
  • Superior to simulated treatment comparisons for non-linear endpoints suffering from non-collapsibility (estimates at covariate average values different from averaging over all the individual estimates obtained from each patient’s covariate values).

Simulated Treatment Comparisons for Unconnected Networks

Often overlooked in favour of MAIC but superior in the sense that

  1. it is amenable to cross-validation techniques to gauge accuracy.
  2. From the cross-validation “systematic bias” can be judged and if thought present, remedial steps taken. Contact Steve for further details.