Patient-specific treatment effects in RCTs: why should one present them?

[arthur_albuquerque]

What is the value of presenting patient-specific treatment effects in RCT/IPD-MAs articles?

Background

In this individual patient data network meta-analysis (IPD-NMA), the authors presented patient-specific treatment effects, along with a nice Shiny app.

Similarly, in this randomized clinical trial, @granholm et al. calculated estimates for three different representative reference patients, e.g.:

Dr. @f2harrell has suggested that picking reference patients to present absolute risk differences (in logistic regressions) is a suitable approach:

Methods to calculate the posterior distribution of the absolute risk reduction

You’d need to pick reference patients as @arthur_albuquerque described, as absolute risk difference is a dramatic function of baseline risk. We tend to not put priors on risk differences because we don’t know that much about them apriori. Instead it is more common to put a wide prior on a reference log odds and a narrower prior on a log odds ratio for comparing A and B.

Prognostic

There was no clear goal of developing a prognostic score in both examples above.

On the other hand, in this other study, @Ewout_Steyerberg and colleagues developed a prognostic score from one RCT, and validated it externally using IPD from three other RCTs.

They then present this figure to depict the application of such score in three different representative reference patients:

Question

When there is no predictive goal, what are the advantages of presenting patient-specific treatment effects along with conditional/marginal relative effects in RCTs/IPD-MAs?

[s_doi]

I don’t know the direct answer to this question but if we keep in mind that covariate adjustment helps generate the closest empiric estimate of individual treatment benefit then there should be no advantage of presenting such estimates. One can simply use the risk based approach to heterogeneity of treatment effect, termed risk magnification (see Frank’s blog post on this) and modify this empiric estimate of individual treatment benefit by the estimated effect of the main covariates of interest to an individual and arrive at the individual benefit. It is therefore more important to provide the latter rather than patient-specific treatment effects.