Read on Twitter
Another paper out on whether central symptoms in networks are relevant to treatment. A (critical) thread to follow. https://bmcmedicine.biomedcentral.com/articles/10.1186/s12916-020-01740-5">https://bmcmedicine.biomedcentral.com/articles/...
The authors operationalize the problem as "does change in central symptoms relate to change in all symptoms more strongly than change in non-central symptoms". I think this is a good operationalization, though not only relevant one.
They find that nodes with high expected influence centrality do indeed predict change in other symptoms better than nodes with low expected influence.
The same finding did not extend to predictability or strength centrality.
The same finding did not extend to predictability or strength centrality.
But then, they exclude the amnesia node, a well-known outlier in terms of correlation to other PTSD symptoms, and the finding disappears. This is where the paper starts to go wrong.
Remember the operationalization was...
Remember the operationalization was...
"Does change in central symptoms relate to change in all symptoms more strongly than in non-central symptoms".
What the authors have done here is try to test this operationalization *after excluding from their analysis the only convincingly non-central symptom*!
What the authors have done here is try to test this operationalization *after excluding from their analysis the only convincingly non-central symptom*!
In addition, because the level of analysis is the nodes, *N = 17*. With amnesia excluded *N = 16*.
Because amnesia is least central, variance in the centrality metric is also highly restricted.
Because amnesia is least central, variance in the centrality metric is also highly restricted.
"Cross-sectional network centrality is useful for treatment" is definitely not the hill I want to die on.
There are many reasons I am very hesitant about this hypothesis, and as the authors point out in this paper, there are many other metrics that are more useful and available
There are many reasons I am very hesitant about this hypothesis, and as the authors point out in this paper, there are many other metrics that are more useful and available
However, I must say I am a bit peeved that this is the second paper to:
1. Find a large, significant effect of this hypothesis
2. Then (mostly) conclude the hypothesis is wrong anyways
(here& #39;s the other: https://pubmed.ncbi.nlm.nih.gov/30265042/ )">https://pubmed.ncbi.nlm.nih.gov/30265042/...
1. Find a large, significant effect of this hypothesis
2. Then (mostly) conclude the hypothesis is wrong anyways
(here& #39;s the other: https://pubmed.ncbi.nlm.nih.gov/30265042/ )">https://pubmed.ncbi.nlm.nih.gov/30265042/...
When you ponder this hypothesis, I think it becomes obvious that it really *must* be true -- it& #39;s nearly tautological. Of course things that are more connected are also going to be more related to change, on average.
I attempt to explain here:
I attempt to explain here:
The real debate is not this theoretical question, but a practical question:
Regardless of the theoretical question of centrality, is this actually going to help with our current symptom sets and our current patients?
Regardless of the theoretical question of centrality, is this actually going to help with our current symptom sets and our current patients?
