Death count altered in the TOGETHER trial

Table S6 in the supplementary appendix was altered on April 5th.

I post this to draw reference to an edit that may otherwise go unnoticed. My intent here is to understand what changes were made, and why. The backstory to this whole thing is that after six months of delay, the ‘big study’ on Ivermectin has arrived, and it has ‘big problems’.

***

As previously reported, the deaths in Table S6 of the TOGETHER paper do not tally to the same number of deaths in Table 3. In a question and answer thread, Edward Mills, the principal investigator for the TOGETHER trial, offered an explanation for why the death figures are different. Below is the relevant question from an email on April 3rd 2022. The full answers Mills gave are at Steve Kirsch’s substack.

Question: There is a discrepancy between "death" and "Grade 5 adverse events"
which should be identical.

Edward Mills: I will check on this, but if memory serves me correctly, it is COVID deaths in table 3 and all-cause deaths in Grade 5. This should have been labeled COVID deaths in table 3. We will have that inserted.

Then, as of April 5th 2022, Table S6 in the supplementary appendix of the paper was updated. An adverse event was added to the Grade 5 Ivermectin tally, and an adverse event was removed from Grade 5 Placebo. Are changes to data permitted after the publication of a paper? Is this normal? I’m leaving this as an open thread so please discuss the matter there. There were other changes, but I want to focus on these altered Grade 5 figures.

Table S6 is titled, “Adverse Events by Grade, MedDRA Type and Treatment Group”. We’re not given any further clues as to what exactly this table represents. The update to the table hasn’t clarified it either - perhaps ambiguity here is helpful? For clarity - remind yourself of the issue, above is Table S6, with Sunday 4th’s version on the left, and the recently altered version on the right. The ‘totals’ we’re looking at are in the first row, “Grade 5, N=x”. Below is Table 3, “Secondary Outcomes with Ivermectin as Compared with Placebo”.

Put simply, until April 5th, the deaths in Table 3 and Table S3 didn’t match, and they are now exactly the same.

Before I get into what I think is happening here, I want to go through a possible explanation. I asked around, and some researchers think Table 3 could represent data for ‘Intention to Treat’ patients, while Table S6 could be the ‘Modified Intention to Treat’ population. This difference in the ‘patient populations’ could explain the difference in deaths, but it renders the recent alteration inexplicable because the deaths are now the same.

I’m not fully convinced by these explanations, firstly because the differences between ITT and MITT populations are quite small, for Ivermectin it’s a change of just 5 patients. Stretching the explanation, if Table S6 is supposed to represent ‘Per Protocol’ then we would expect a difference because there are 55 fewer patients in the IVM ‘Per Protocol’ population. If we expect a difference, why have the tables been updated in such a way as to make them the same?

The second reason I’m not convinced is because we are having to infer what these changes mean after the fact. We don’t have a changelog of what was altered and why. We weren’t clear on why there should be any difference in deaths in the first place, and the one explanation I found doesn’t tell us why the numbers should now be the same. Is there another explanation? We have a tiny snippet from Mills that might help to illuminate things…

Mills’ email from April 3rd suggests Table S6 is all-cause deaths. Table 3, he says, “should have been labelled COVID deaths. We will have that inserted.” That could explain why there was a difference in deaths between Table S6 and Table 3. It’s also credible: the study protocol on page 14 specifically says that ‘All-cause, respiratory and cardiovascular death’ were being evaluated as secondary objectives. Let’s move forward with the all-cause death explanation because as we follow its logic, it reveals a critical issue.