Tutorial: The Practical Application of Longitudinal Structural Equation Mediation Models in Clinical Trials (2017) Chalder et al

H

hixxy

Senior Member (Voting Rights)
Goldsmith KA, MacKinnon DP, Chalder T, White PD, Sharpe M, Pickles A.

Abstract

The study of mediation of treatment effects, or how treatments work, is important to understanding and improving psychological and behavioral treatments, but applications often focus on mediators and outcomes measured at a single time point. Such cross-sectional analyses do not respect the implied temporal ordering that mediation suggests. Clinical trials of treatments often provide repeated measures of outcomes and, increasingly, of mediators as well. Repeated measurements allow the application of various types of longitudinal structural equation mediation models. These provide flexibility in modeling, including the ability to incorporate some types of measurement error and unmeasured confounding that can strengthen the robustness of findings. The usual approach is to identify the most theoretically plausible model and apply that model. In the absence of clear theory, we put forward the option of fitting a few theoretically plausible models, providing a type of sensitivity analysis for the mediation hypothesis. In this tutorial, we outline how to fit several longitudinal mediation models, including simplex, latent growth and latent change models. This will allow readers to learn about one type of model that is of interest, or about several alternative models, so that they can take this sensitivity approach. We use the Pacing, Graded Activity, and Cognitive Behavioral Therapy: A Randomized Evaluation (PACE) trial of rehabilitative treatments for chronic fatigue syndrome (ISRCTN 54285094) as a motivating example and describe how to fit and interpret various longitudinal mediation models using simulated data similar to those in the PACE trial. The simulated data set and Mplus code and output are provided. (PsycINFO Database Record.
Click to expand...

https://www.ncbi.nlm.nih.gov/pubmed/29283590
http://psycnet.apa.org/fulltext/2017-57776-001.pdf
 
Oh dear I think they should stay away from stats. From what I remember with the PACE mediation model they failed to model important temporal aspects. Also I seem to remember that the scales they use broke the basic assumptions of the methods.

But mediation models as whole seem to be a bad idea trying to replace complex methods and corrections with trying to use the stats to gain an understanding of the basic system. David Freedman wrote a good article "statistics and shoe leather" with the point being:
https://web.math.rochester.edu/people/faculty/cmlr/Advice-Files/Freedman-Shoe-Leather.pdf
"Statistical techniques can seldom be an adequate substitute for good design, relevant data, and testing predictions against reality in a variety of situations"

It seems like they use mediation analysis to cover for having an untestable hypothesis and bad data (subjective measures in a trial aimed at changing subjective feelings of symptoms).

Still I may have a read as I was thinking of looking at mediation analysis for a data prediction problem.
 
For crying out loud. Is the PACE team going to teach us about Structural Equation Models? Embarrassing. Their analysis uses subjective outcomes and excludes all biomedical factors. Talk about begging the biopsychosocial question.

Any expert in the field could explain for Goldsmith et al. that you cannot prove causality from SEMs. They can be used to test causal models and they can be used for exploratory analysis, but statistical methods alone cannot determine the direction of causality. Moreover, any SEM depends on the assumption that you have included all relevant factors. The causal inference from SEMs is only as strong as the underlying theoretical support for the model and for the choice of causal factors. Temporal analysis can in some cases resolve direction of causality, but can never solve the problem of relevant factors.

I think that the PACE team should contemplate the meaning of scientific models, in particular the difference between causal models and black box correlational models.
 
Suggested alternative title:

Tutorial: The Practical Guide How to Create a (Bio)psychosocial Pseudomodel for an Illness of Arbitrary Etiology
 
I suspect the statistician on the PACE team needed another publication for his portfolio, or a project for a post grad student to take on, and they invented this stuff using a simulation of PACE as a pseudo data set to try it out on.

And because they used PACE it will get more readers and therefore greater kudos than any old made up data set, and can attach the names of the PACE team to the author list to add weight and make their number games look 'useful'.
 
You must log in or register to reply here.
Back
Top Bottom