One more comment here. So... by nature... the subgroup analysis should average out in each category to an effect size of ~0.9. Which holds for all of them except for the time since symptom onset, where the average is >1. This doesn't make sense and suggests dropped events biased.