Before I submit a report, I now compare the effect size and confidence interval in the abstract to the primary results table; last week it caught a CI that would have flipped a “significant” claim. What’s your go-to quick accuracy/ethics check that doesn’t slow the review to a crawl?
A 60-second denominator sweep: do the Ns stay consistent from abstract to flow diagram to Table 1, and do the percentages actually sum to about 100 within rounding. If an arm mysteriously loses people or a subgroup total can’t be reconciled, I stop and ask before reading further.
Nice catch on the CI flip — I’ve started doing a one‑minute prereg check: grab the registry ID and make sure the declared primary outcome and timepoint match the main table and abstract. If they quietly swapped outcomes or timepoints, the “significant” claim usually evaporates, ugh. A 10‑second multiplicity sanity check is usually the only extra thing I add.
I do a 30‑second units/scale sanity check: are ORs labeled as RRs, are log outcomes back‑transformed, and do mg/dL vs mmol/L stay consistent across abstract, methods, and tables — it’s saved me from more than one mg/dL→mmol/L facepalm. Do you also peek for sign flips on betas between text and the main table?