Exploring the effects of network analysis on depression trial outcomes: Protocol for secondary analysis of individual participant data

BackgroundNetwork analysis (NA) is a modern statistical method for exploring relationships and patterns in complex data. NA techniques can also be used to eliminate unstable or nonperforming items, potentially reducing measurement error. However, the use of NA to improve measures used in randomised controlled trials has been limited, and it is unknown whether applying such techniques could impact trial effect size outcomes.AimWe aim to determine whether network analysis can impact clinical trial effects by reducing measurement error in depression models and subsequently modifying trial effect size outcomes. MethodWe will analyse individual participant data (IPD) from multiple depression trials that used the Montgomery Åsberg Depression Rating Scale (MADRS) as a depression measure. Data will be accessed from Vivli.org. A sequence of network modelling, followed by bootstrapping and stability analysis, will be performed to revise models of the MADRS at baseline and outcome. Net scores will be derived from abbreviated models and the differences in original trial outcomes versus the abbreviated outcomes utilising net scores, will be the effect of interest. Effect size outcomes will be modelled using multilevel linear regression. DiscussionThis study will determine whether network modelling can improve precision and inform better estimates of effect sizes in antidepressant treatment trials. The outcome of the proposed study could inform a shift in the way in which clinical trial data, and indeed data from other types of studies, may be analysed.