[DavidSJ]

But this study doesn't say the opposite. It fails to show an effect. That's different from proving the absence of an effect.

This was my first thought, jaded as I am from bad scientific reporting (such as the linked article) which doesn’t distinguish between these two cases, so I had a look at the actual study.

In this case, it looks like the 95% confidence interval just barely overlaps the null hypothesis, however the mean effect favors placebo:

The mixed-effects model showed no evidence of effect of group assignment on post-infusion MADRS scores at 1 to 3 days post-infusion (-5.82, 95% CI -13.3 to 1.64, p=0.13).

(See also figure 2, which clarifies the direction of effect.)

So, potential methodological issues aside, I’d actually consider this evidence against a strong benefit relative to placebo, and possibly very weak evidence of harm.