Thread by @BadreNicolas

STAR*D a 🧵

If you don’t read the thread - super short: they claimed that antidepressants as used in common practice worked 67%, when the result actually was 35%.

Problem #1

#1 The biggest problem, the extra 931. We were told that STAR*D was about 4,041 patients with depression when in reality 931 (99+508+324) did not score high enough on a depression scale (HSRD of less than 14), so it should have been 3,110 patients.

Imagine a trial of disease X (e.g. cancer), where patients are recruited from all over the country. They are given medications Z (e.g. chemotherapy) as part of the trial. What if we later found out that 931 (almost ¼) of those 4,041 recruited did not have the disorder?

#2 the missing 370. As you can see in this picture of Step 1, it suggests that only 3,671 patients received citalopram. This is misleading, in reality all 4,041 received a prescription of citalopram and should have been included in the results for intent-to-treat analysis.

#3 HAM-D v QIDS. The study reported outcomes using the QIDS rather than HSRD. Fundamentally, the scales are equivalent. However, they should have made it much clearer that the QIDS was not blinded. So they inappropriately used a non-blinded QIDS when a blinded HSRD was available.

#4 The definition of healed in STAR*D is less than ideal. A patient denies depression for just 1 visit, it is considered a remission. Sadly, this is common in depression trials. With ketamine, you can even be considered in remission for denying depression for just a few hours.

#5 The STAR*D trial also looked at folks who recovered and how long they stayed well. The results were less than ideal. Only 108 (out of 4,041) did not relapse. To be fair to the author, this was included in the paper (though not well outlined or explained).

#6 “Theoretical” results. The conclusion in the abstract and in the conclusion has an interesting word change - “theoretical.” When you read the paper, you may not realize its significance, but it meant that individuals who dropped out were assumed to have also been able to obtain remission (the opposite of intent-to-treat analysis). Without this statistical voodoo (not including the other problems) the remission rate would have been 51%.

#7 Pharma. The disclosure as described by the authors was 850 words (that is 850 words more than my disclosures).

When you put all those problems together you get - they claimed that antidepressants as used in common practice worked 67%, when the result actually was 35%.

A few more thoughts on STAR*D. This is mostly a critic of the main paper of STAR*D: Rush et al. (2006). Acute and longer-term outcomes in depressed outpatients requiring one or several treatment steps: a STAR* D report. American Journal of Psychiatry

There are many others paper written about STAR*D, which do not have those mistakes. Sometimes people believe that the study should be retracted, when it would be more appropriate to say that only papers that wrongly describe the study should be.

This thread is based on the important work of Piggott et al. (2023) A reanalysis of the STAR* D study’s patient-level data with fidelity to the original research protocol. BMJ open

Read both Rush et al and Piggott et al then make your own conclusions. This is important. STAR*D is highly cited and only getting more cited.

Also consider reading my article on the story.

psychiatrictimes.com/view/star-d-it…