Be careful about interpreting these estimates. Addiction is a good example: Although beset by selection problems, it looks like many (possibly most) substance-dependent people will eventually recover, despite poor evidence for any specific intervention (relative to just encouraging someone to quit), and the low odds of recovery for a single attempt. But you wouldn't say that these recoveries were somehow accidental or not self-directed! Slatestarcodex had an interesting review of this for alcoholism here. Also note that several of the other areas will tend to change regardless of a specific intervention: * Many (major) depressive episodes will resolve themselves over time. So a positive effect due to treatment could mean several different things: Faster recovery; increased odds of recovery; greater magnitude of recovery; less chance of relapse; or even something like less functional impairment despite no subjective relief. I've read less about other conditions. * Weight tends to flatten out and later decrease in mid-life. * Absolute personality trait measures drift with aggregate predictability over time. (Relatives measures are more stable, but it's not shocking to see large a large percentile change in an individual.) Finally: When people talk about weight-loss, they're usually talking about dieting. But bariatric surgery has good evidence for large, long-term weight loss, and is not a rare procedure, (~100,000 a year: link)

I wonder whether the classification of PTSD takes account of the apparently miraculous effects of MDMA shown in some studies.