Severity of Depression and treatment effects

[masterofmonkeys]

I may be a crazy person, or I may just not know how to do statistical analysis. Both are strong possibilities. I just found something I didn't really expect when I ran on a meta-regression and I was wondering if anyone could bring me some clarity on this finding. Namely, that the size of the treatment effect was positively correlated with the severity of depression as measured by BDI.

Far as I know, most treatments (SSRIs I know for sure) show smaller treatment effects the higher the BDI score is. (i.e. the more depressed you are the worse they work)

If it turns out I'm not crazy, I'm kind of excited. Although that means a hell of a lot more work before this paper is ready.

---

Members don't see this ad.

 

[dianamd]

Far as I know, most treatments (SSRIs I know for sure) show smaller treatment effects the higher the BDI score is. (i.e. the more depressed you are the worse they work)

I would expect more severe depression to show a larger treatment effect.

For example, a person with a Beck depression index (BDI) score of 12 can only drop, say, 5-7 points, with treatment. -- 12 = mild depression.

Someone with a BDI of 34 (severe depression) has a lot more room to go -- if they are treated till they have mild depression, that could be 20+ points on the BDI. So, larger treatment effect.

SSRIs may not "work as well" with severely depressed people -- I never heard that, btw -- but they can certainly have as great of an absolute effect on more severely depressed patients, because with those patients there is so much more room to make an impact (as measured by the BDI).

 

[Faebinder]

It's a well known fact that SSRIs truly work on the severely depressed patients, moderate or mild. As said above, this is probably because there is a lot more room to drop (i.e. improve) in severe depression.

That's why SSRIs given to those with depression post cocaine or depression from alcohol do not have a big effect.

 

[masterofmonkeys]

sorry I got my whatchamacallits backward there. I know SSRIs work on severely depressed people. Wasn't saying they didn't. I was saying that in severely depressed people they are less efficacious. In some ways, we overstate the effect of SSRIs in the severely depressed compared to less severely depressed patients because of the variable placebo effect. In recent years, research has shown that we may be mis-measuring the efficacy of SSRIs because the placebo (and therefore difference from placebo) is a significant confounder.

http://medicine.plosjournals.org/pe...ure&doi=10.1371/journal.pmed.0050045&id=96823

Not sure that link'll work or not, but that's what I was talking about.

CBT also shows lower treatment effect at more severe levels of depression.

Treatment effects are calculated using the SD. Generally the higher the scores are numerically for a cohort, the higher the SD will be, which serves to counteract the magnitude of calculated effect size. That's the rationale for using metrics like hedges and cohen's d instead of a pearson or similar R.

 

[Faebinder]

sorry I got my whatchamacallits backward there. I know SSRIs work on severely depressed people. Wasn't saying they didn't. I was saying that in severely depressed people they are less efficacious. In some ways, we overstate the effect of SSRIs in the severely depressed compared to less severely depressed patients because of the variable placebo effect. In recent years, research has shown that we may be mis-measuring the efficacy of SSRIs because the placebo (and therefore difference from placebo) is a significant confounder.

http://medicine.plosjournals.org/pe...ure&doi=10.1371/journal.pmed.0050045&id=96823

Not sure that link'll work or not, but that's what I was talking about.

CBT also shows lower treatment effect at more severe levels of depression.

Treatment effects are calculated using the SD. Generally the higher the scores are numerically for a cohort, the higher the SD will be, which serves to counteract the magnitude of calculated effect size. That's the rationale for using metrics like hedges and cohen's d instead of a pearson or similar R.

Check the time frame as well... the time frame by which they calculated the change. Personally, I favor Spearman's correlation over pearson, but remember you are doing a meta-regression, so you will be assuming so much regarding the populations... perhaps you included populations of different trials that shouldn't be included together in a meta-regression... Get a second set of eyes to check your raw data action, though I realize trying to get a professor interested in this is like trying to talk a model to do algebra.

 

[masterofmonkeys]

I don't know anyone who does this kind of research in psych. Got two people (both family though, neither in psych, but both with strong stats backgrounds in research positions) to look over the data. I haven't been able to find anything that would indicate I shouldn't lump these trials together in meta-regression. Demographic variables are very consistent with nothing found as far as age, education, income, comorbid illness, etc. Well manualized therapy, good adherence, and the correlation is absurdly strong.

There is reason to expect the positive correlation based on some of the ancillary findings from included trials that I didn't include in my analysis because there was too much heterogeneity in reporting. I expected the finding and it was in my a priori analysis plan, but as I said, it was stronger than i expected. And, as I said, I'm very surprised that I haven't found another therapy that shows the same thing. i've got suspicions on what this therapy affects to produce these results, and I expected other therapies to also affect the same variable, thus producing a similar effect on effect sizes.