Three reassuring-looking nulls, and a methods section that explains why the word “pooled” is doing more work here than usual — the same distinction between what different designs can be asked that decides how much any average is worth.
What went into the pool
The eligible studies were randomized controlled trials, observational studies, large database analyses and pharmacovigilance disproportionality studies. [1] All four were combined into single risk ratios.
Those are not the same measurement. A randomized trial compares what happened to two groups assigned by chance. A disproportionality analysis divides one kind of report by other reports inside a spontaneous database, with no count of how many people took anything. The first estimates a risk; the second estimates how a database looks, which is why the cohort evidence on this question is worth separating from the report counts rather than averaged with them.
What the numbers say
Depression 1.25, 95% CI 0.95 to 1.65. Anxiety 1.22, 95% CI 0.93 to 1.60. Suicidal ideation or attempt 1.20, 95% CI 0.90 to 1.62. Every interval crosses 1, so none is significant.
All three point estimates sit above 1, which is worth saying plainly and worth not over-reading. It is not evidence of harm. It is also not the same picture as three estimates landing on either side of 1, and with intervals this wide the data cannot distinguish between the two.
The sentence the authors close on
They describe their own findings as reassuring, and then write that they should be interpreted as the absence of a detected increased risk in the available evidence rather than definitive proof of no psychiatric risk.
That is a precise and unusually honest closing line, and it is the whole of why not finding something is not finding nothing written by the people who did the work.
What to carry away
That regulators on both sides of the Atlantic have looked at this question and not confirmed a causal link, and that this analysis does not contradict them.
And that when a meta-analysis reports heterogeneity in the nineties, the first thing to check is what it pooled. An average of incompatible designs is a summary of a literature, not a measurement of a risk — the same caution that applies to any pooled figure no individual study reported. Anyone experiencing thoughts of harming themselves should speak to a doctor now, not weigh an interval.