Also, if the main objective of the study (like reducing the number of death) is not successful, you can pick another criteria of successfulness like number of day in the hospital, viral load after a week, ... there are many to choice.
In particular in the fists study about hydroxychloroquine by Raoult, he claims that it's a success, but if you read the paper you can notice that there is no one death in the trial group and no deaths in the control group. (And the control group is not a randomized control group, just an unrelated bunch of patients in another city.)
About why early interventions with hydroxychloroquine get apparent better results:
That's a good question, but the publication bias is not the only source of problems.
It's difficult to be sure, but look at the study here in the OP. My guess is that the treatment with ivermectin includes implicitly phones call "Locatel" and that additional care from doctors improve the result. Other studies have similar additional care for the people under treatment.
And if you start the phone calls earlier, it's better that starting late. So you may measure it in the trials.
You may disagree with my explanation, and it's fine. The real problem is how to be sure if I'm wrong or you are wrong (or both). So the solution is to do double blind studies, were some people that get the medicine and other people that get a placebo that looks identical. But the important part is that both get all the additional health treatment. Both get phone calls that perhaps are useful. Both get automatic weakly nurse checks that are common in other treatments and perhaps are useful. ... So it's not necessary to guess if any of the other part of the treatment is useful, because the only difference is drug vs placebo.