Among 541 abstracts presented at the 2018 American Diabetes Association Scientific Sessions, only 59.3% were subsequently published in indexed journals within five years. Critically, abstracts reporting statistically non-significant findings faced a 29% lower likelihood of publication compared to significant ones (adjusted RR 0.71, 95% CI 0.54–0.93), with raw publication rates of 42.3% versus 61.9% respectively.
This finding lands squarely within a well-documented crisis in biomedical publishing. Publication bias — the selective suppression of null or inconclusive results — distorts the evidence base clinicians and policymakers rely on, systematically inflating apparent treatment effects. In diabetology, where meta-analyses inform glucose-lowering drug approvals and guideline thresholds, this selective dissemination carries direct patient-care consequences. The 40% non-publication rate means a substantial fraction of human trial effort and participant exposure simply vanishes from the scientific record.
Limitations are notable: the study examines a single conference year (2018), which may not reflect current trends, particularly post-pandemic shifts toward open-science practices. Conference abstracts also vary widely in methodological rigor, and the search strategy, though dual-database, may have missed publications in non-indexed journals. Causal directionality between significance and publication cannot be fully established from this design.
As a preprint not yet peer-reviewed, these figures should be interpreted cautiously. Nevertheless, the analysis is methodologically straightforward and the signal is consistent with prior publication-bias literature across specialties — confirmatory rather than paradigm-shifting, but a timely reminder that effect sizes and confidence intervals must displace p-value gatekeeping.