Case study: methods for NMA using IPD – depression


Network meta-analysis (NMA) is used to summarise relative treatment effects from randomised trials (RCTs) that compare multiple competing interventions for the same condition (for more about NMA, see Overview of Evidence Synthesis and NMA). Most NMAs are based on published aggregate data (AD), but this limits the ability to investigate the extent of network consistency and between-study heterogeneity. As individual participant data (IPD) are considered the gold standard in evidence synthesis, it may be possible to use this when conducting NMA.

What was examined in this case study?

The case study aimed to:

  • explore statistical methods for NMA using IPD in the existing literature
  • investigate the potential advantages and limitations of IPD NMA compared with NMA based on summary data from the literature
  • provide technical details on how to conduct an NMA using IPD
  • advise on circumstances in which use of IPD may be advantageous.

IPD was obtained from 18 RCTs investigating the comparative efficacy of several antidepressants. All participants were diagnosed with major depressive disorder, and monitored for depression severity using the Hamilton depression rating scale. All statistical models used a continuous endpoint (the Hamilton depression score).

The case study did not aim to develop methods for combining IPD and published AD, because the RCTs included IPD only.

What ‘effectiveness challenge(s)’ was addressed in this case study?

The case study is not directly linked to specific ‘effectiveness challenges’ because it only considers RCT data. However, findings from the case study also apply to NMA that are based on non-randomised studies. These studies typically suffer from missing data and confounding, and methods to address these challenges were discussed.

What were the findings and conclusions?

IPD NMA offers several potential advantages over NMA that are solely based on AD. Its use should be considered when studies are affected by substantial drop-out, and when treatment effects are potentially influenced by participant-level covariates. Also, when considering the inclusion of non-randomised studies, IPD may help to properly adjust for confounders, thereby increasing the credibility of NMA results.

Debray et al. (2016) presents a generic NMA framework to:

  • combine IPD
  • include covariates (prognostic factors and/or effect modifiers)
  • address missing response data
  • account for longitudinal responses.

What do stakeholders say?

  • IPD is not typically available for all studies. Usually researchers will have access to a mixture of IPD and AD. Methods for combining both types of data exist, but have not been explored in this case study.
  • For some stakeholders the statistical methods might be difficult to understand or carry out.

Key contributor

Thomas Debray, University Medical Center Utrecht