For a more in depth evaluation of various strategies, we could encode the full PRISMA chart from each meta-analysis, to separately evaluate the peformance of automated strategies in distinct PRISMA steps separately (e.g. is automated screening similar to LLM screening)?