Elevated concentrations of CSF corticotropin-releasing factor-like immunoreactivity in depressed patients.by Nemeroff CB, Widerlöv E, Bissette G, Walléus H, Karlsson I, Eklund K, Kilts CD, Loosen PT, Vale W.Science. 1984 Dec 14;226(4680):1342-4.
The possibility that hypersecretion of corticotropin-releasing factor (CRF) contributes to the hyperactivity of the hypothalamo-pituitary-adrenal axis observed in patients with major depression was investigated by measuring the concentration of this peptide in cerebrospinal fluid of normal healthy volunteers and in drug-free patients with DSM-III diagnoses of major depression, schizophrenia, or dementia. When compared to the controls and the other diagnostic groups, the patients with major depression showed significantly increased cerebrospinal fluid concentrations of CRF-like immunoreactivity; in 11 of the 23 depressed patients this immunoreactivity was greater than the highest value in the normal controls. These findings are concordant with the hypothesis that CRF hypersecretion is, at least in part, responsible for the hyperactivity of the hypothalamo-pituitary-adrenal axis characteristic of major depression.
"The results (see Fig. 1) were statistically analyzed by both parametric [analysis of variance (ANOVA) and Student-Newman-Keuls test] and nonparametric (Mann-Whitney U test) methods."
One of the most frustrating things about papers like this is that the raw data isn’t available, even if one has the time to go over it in detail. Once again, I find myself looking at a graph that I’m told is meaningful, significant, has something to say important about a major psychiatric syndrome. And what I see looks like a trivial difference that is probably meaningless, and I even doubt significant. So I did something that I’ve been tempted to do many times. I opened it in a graphics program and reconstituted the data by measuring the pixel count to the center of each data point and using that table, the baseline, and the ordinate scale to reproduce the data. I wouldn’t recommend this on a Nobel Prize application or even in a paper, but I thought I’d give it a shot because I don’t believe the analysis is correct, or correctly done [the next paragraphs is only for the hardy].
The CSF concentration of CRF-LI was significantly increased (by both methods of statistical analysis) in patients with major depression compared to either the normal controls or the patients with schizophrenia or senile dementia."
As you may recall, when we looked at Dr. Nemeroff’s NYU Grand Rounds and London lecture to the Institute of Psychiatry, we were alerted to a study reported as positive that Dr. Nemeroff, himself, had reported as based on an error so the significance disappeared, yet he presented it as a valid study in those presentations [see has to stop…]. So the best predictor of future behavior is past behavior. Now we have GSK, the VAH, and the NIMH chasing some new drug as a treatment for PTSD based on the very shakiest of speculations. Shame on him. Shame on them. And shame on journals that don’t vet questionable studies like this.
Maybe we ought to say shame on me too for using a pixel count to get my numbers. But instead of that – why not support Data Transparency so I don’t have to resort to extreme measures to confirm my reaction to that graph. Like I said, this kind of silliness has to stop…
Whoops: [for the even more hardy] I left out this plot from the R package. The upper and lower borders of the "boxes" represent 25% and 75% of the points. The fact that the Means [bold horizontal lines] aren’t centered in the box points to a skewing of the data [not normally distributed], suggesting that the ANOVA is not the best choice of statistics, and that the non-parametric test is a more appropriate choice [Kruskal-Wallis]. My method of data capture is also more likely to be accurate using only the rank order.