Read on Twitter
1/ Our new paper about the minimal important difference (MID) of antidepressants (AD). Because there is a paywall, here the main findings in a thread. https://ebm.bmj.com/content/early/2021/02/15/bmjebm-2020-111600.full
2/ First of all, Michael was the one who did most of the work, so credit goes to him! We had great discussions and chats, and I learned so much from this collaboration. His overview on the literature is outstanding!
3/ The goal of the paper was to summarize different approaches/criteria to estimate the clinical importance or minimal important difference (MID) of antidepressants (AD), because there is still some controversy if the efficacy of AD is clinically meaningful or not.
4/ Some may be surprised that there is controversy bc the efficacy of 2 points AD-placebo difference on the HAMD-17 is obviously tiny & hardly clinical meaningful. However, some say the HAMD is inappropriate & there is no consensus about criteria for MID or clinical significance.
5/ There are two main approaches to estimate the MID. 1st, anchor-based estimates, where the treatment effect is compared with clinical judgment. 2nd, distribution based estimates where observed treatment effect is judged via their statistical properties (eg standard deviation).
6/ For anchor-based methods, changes in HAMD scores are compared with the clinical global impression scale (CGI). Here, 0-3 points improvment on the HAMD-17 corresponds with “no change” in the CGI, and...
7/ ... at least 7-9 HAMD-17 points are needed for a “minimal improvement” assessed with the CGI. Thus, the observed efficacy of AD is well below this threshold of minimal improvement.
8/ Counter-argument 1: the HAMD-17 is no reliable measure of depression and better ones exist (MADRS, HAMD-6, BDI), or using quality of life measures instead. As we show, the efficacy of AD with those measures is still below the threshold of clinical significance.
9/ Counter-argument 2: regression to the mean is likely a problem for pre-post measures on the HAMD, but less so for the CGI. Therefore, we assumed that the 0-3 points difference corresponding with “no change” on the CGI may actually reflect regression to the mean.
10/ However, even after correcting for potential regression to the mean, the efficacy of antidepressants is below these corrected estimates. This is true for the HAMD-17 but also for other, perhaps better depression rating scales.
11/ Counter-argument 3: it is inappropriate to use MID estimates based on within group pre-post change for between group comparisons (AD vs. Placebo).
12/ That is, if let's say 5-8 points on the HAMD-17 scale correspond with “minimal improvement” on the CGI, you cannot infer that improvement with AD needs to be 5-8 points larger than improvement with placebo.
13/ However, this is under debate. For example, it is common to use clinical significance estimates to plan the sample size of clinical trials. One reviewer also asked why not use the MID for between group comparisons.
14/ Alternative estimations of between group MID could be comparisons of pts with mild and moderate depression, or with partly and full remission. Using this approach for different depression scales, the efficacy of AD is still not exceeding these estimates of MID (see table).
15/ Distribution based estimates are the second main approach to estimate the MID. Across fields, the MID often corresponds with 0.5 standard deviation. Again, the observed efficacy of AD is well below 0.5 SD (see table).
16/ Another approach is to use 7-10% of the maximum scale value bc this was found to correspond with the MID for health related quality of life (this correlates substantially with depression scales). We did this in a previous version of our paper only. Results were comparable.
17/ Importantly, (corrected) anchor based and distribution based estimates seem to lead to comparable estimations of the MID. This is a main message of our paper.
And the observed average efficacy of AD does not exceed even the lower-bound of the different MID estimates.
And the observed average efficacy of AD does not exceed even the lower-bound of the different MID estimates.
18/ Also importantly, it is not sufficient that AD exceed the lower bound of the MID estimates. These lower bounds only make sense in the absence of any risks of a treatment, and this is clearly not the case for AD.
19/ AD have notable risks/side-effects, e.g. sexual dysfunctions for common ADs, or withdrawal after stopping treatment. Consequently, the suggested lower-bounds of the MID are most likely not appropriate estimates of the sufficiently important difference. https://twitter.com/PloederlM/status/1265555488655192065?s=20
20/ Closure: "In conclusion, the efficacy estimates of antidepressants, as assessed with either HDRS-17, HDRS-6, MADRS, BDI-II or SF-36 consistently fail to exceed the lower-bound MID estimates, indicating
that the clinical significance of antidepressants remains
uncertain...
that the clinical significance of antidepressants remains
uncertain...
21/ ...More research on patient-reported anchors, preferably based on QoL measures, is urgently required to evaluate whether patients consider the benefits of antidepressants over clinical management without antidepressants sufficiently worthwhile"
22/ PS: the sufficiently important difference, which also takes into account the risks, is something that has to be judged by patients, not by clinicians. Thus, there may be substantial individual differences.
23/ So, what we need are related studies with patients. Such studies were done for pain management etc, but not for depression treatment.
For example: https://www.meta.org/papers/using-benefit-harm-tradeoffs-to-estimate/15673581
It is striking that related research for AD is lacking (AFAIK), given their widespread use for decades.
For example: https://www.meta.org/papers/using-benefit-harm-tradeoffs-to-estimate/15673581
It is striking that related research for AD is lacking (AFAIK), given their widespread use for decades.
24/ Such a study is on my to do list. Imagine informing patients about the HAMD (of MADRs), the small average efficacy, and also the risks - should be done before prescribing, BTW. Hard to imagine that patients would be satisfied with 2 HAMD points difference.
