Taking a PTSD meta-analysis to Budapest

Our PTSD meta-analysis was accepted to the International Conference on Addiction and Psychiatry in Budapest. It’s the second conference for this paper, after the Southern California Conference for Undergraduate Research in 2025, and the first one that requires a passport.

I led this project in the Jahanikia NeuroLab, with a team of five.

The question

A meta-analysis combines the results of many studies to get one bigger answer. Ours asked how well PTSD treatments work, using 538 effect sizes from 328 randomized controlled trials.

The catch is that “how well does it work” depends on what you compare it to. A treatment can look great next to a waitlist, where the control group gets nothing, and look ordinary next to another real treatment. So instead of reporting one number, we looked at what drives the number.

We used a three-level meta-regression with robust variance estimation, which handles the fact that many trials report more than one effect size. We cross-checked it with MetaForest, a machine learning method that screens which study features matter most.

What we found

Comparator type was the biggest driver. Against inactive controls, treatments had an effect size of g = −0.64, which is a meaningful effect. Head-to-head against other treatments, the effect was near zero.

Then we looked at publication bias, the problem that studies with exciting results are more likely to get published. We re-estimated the field-wide effect with three different corrections: PET–PEESE, selection models, and RoBMA. After correcting for bias, the effect came out near zero.

That’s not a fun result to report, but I think it’s an important one. It doesn’t mean PTSD treatment doesn’t help people. It means the way trials are designed and published can make the evidence look stronger than it is, and the comparison group matters more than most headlines suggest.

The paper is under review at a journal. Before that, it’s going to Budapest.