Low-quality studies mean meta-analyses are frequently worse than single studies

One possible takeaway from the article: think of peer review like you think of code review.

Code review is not there to produce bug-free code. Every serious quantitative study of code review shows it has surprisingly low impact on defect rates (not zero impact, but also not much impact).

Code review is there to produce understandable and maintainable code.

If you expect peer review to stop all bad results, the time to complete the review goes to infinity, just like happens with code review, and you still don't succeed at the original goal.