A recent meta-analysis of meta-analyses (seen via Less Wrong and Metamodern) found that most meta-analyses were of very low quality:
Of the 145 systematic reviews we found, fewer than half met each quality criterion; 49% reported study flow, 27% assessed gray literature, 2% abstracted sponsorship of individual studies, and none abstracted the disclosure of conflict of interest by the authors of individual studies. Planned, formal internal quality evaluation of included studies was reported in 37% of systematic reviews. The journal of publication, topic of review, sponsorship, and conflict of interest were not associated with better quality. Odds of formal internal quality evaluation (odds ratio [OR], 1.10 per year; 95% confidence interval [CI], 1.02-1.19) and either planned, formal internal quality evaluation or abstraction of quality criteria of included studies (OR, 1.17 per year; 95% CI, 1.08-1.26) increased over time, without positive trends in other quality criteria from 1990 through June 2008. Systematic reviews with internal quality evaluation did not meet other quality criteria more often than those that ignored the quality of included studies.
If this is true, we can't trust it.