Interesting articles: A preliminary read of

“Comparative efficacy and acceptability of

21 antidepressant drugs for the acute
treatment of adults with major depressive
disorder: a systematic review and network
meta-analysis” (Cipriani et al)
First published at https://thingsididntknowinpsychiatry.wordpress.com/
March 8th, 2018
I’ve recently had the chance to read over an interesting article on antidepressants, namely
“Comparative efficacy and acceptability of 21 antidepressant drugs for the acute treatment of adults
with major depressive disorder: a systematic review and network meta-analysis”. I’m only just
starting to mull it over, but I thought I’d put out my initial thoughts.
Whilst it looks like a fairly solidly put together paper, I’m not overly excited by its conclusions.
Firstly, it’s a network meta analysis, which means its quality is heavily limited by whatever trials
get put into it. There’s nothing magical about a network meta analysis which makes it better- it is
merely an analysis where multiple treatment are being compared both directly within RCTs, and
indirectly across trials with a common comparator.
That’s not a complex concept. If A is compared to B directly, and B compared to C directly, then A
can be compared to C (albeit indirectly). As long as you have something standardised to compare
with, it’s quite workable.
That said, in these types of studies, the eligibility of studies is incredibly important. Because
anything’s placebo arm can be used as a comparator, a single study that is incorrectly admitted can
have notable ripple effects on the rest of the study. Likewise, in antidepressant efficacy studies,
questions like “what level of medication is considered the same intervention” become incredibly
important.
Reading over the paper itself, I note that there is a significant amount of heterogeneity in the criteria
for the trials accepted. In particular, excluding only trials with more than 20% of patients having
bipolar, psychotic depression, treatment resistant depression or medical comorbidities could
significantly confound results. Twenty percent is a large number of people who (potentially) are
suffering from an entirely different disorder.
My second concern regards the sourcing of the studies. 78% of the studies were funded by
pharmaceutical companies. This is somewhat concerning. Obviously, this is not a criticism of the
authors of the papers, who essentially had to work with that which was available. However, I must
make note of Anderson’s old meta-analysis, which suggests that merely being sponsored by a drug
company has a definite impact on outcome.
My third concern is regarding the dosing ranges. The limits that were put in place for efficacy was
any trial which were within the relevant licensed dose range. However, there is a significant
difference between a medication being tested at the bottom and upper parts of the range; the
efficacy of these two points may be quite different. I am thinking, for example, of venlafaxine,
whose noradrenergic effects only come into effect at higher doses. (I will have to read the specific
papers within the study to establish how much this is an issue; it may be that the doses are within
only a slight variation)
My fourth concern is regarding the use of unpublished material. Whilst it’s useful to have such
material in order to have the widest body of information, it also means that the data may not have
gone through particularly rigorous error checking. Equally, it isn’t easily available for me to review
the data directly. That makes me cautious.
So until I’ve had a chance to manually look over the referenced papers, and really get a good grip
on what’s going on in there, I’m going to have to hold off on being particularly influence in practise
by the results. Expect a fuller post on this topic a bit later.