Epistemic Asymmetry Between Abstracts and Discussion Sections in the Orthopaedic Literature: A Corpus Analysis of 201 Publications.

BACKGROUND: Abstracts and Discussion sections serve distinct functions: Abstracts are optimized for brevity and impact, while Discussion sections contain authors' most explicit engagement with uncertainty and the scope of inference. Whether these sections differ systematically in epistemic stance-the level of confidence or uncertainty with which findings are stated and qualified-has not been formally examined in the orthopaedic literature.

METHODS: All original research articles published in The Journal of Bone and Joint Surgery (JBJS) and Clinical Orthopaedics and Related Research (CORR) during the first half of 2025 were included (total, n = 201; JBJS, n = 153; CORR, n = 48). Editorials, reviews, letters, and case reports were excluded. The densities of hedges (markers of uncertainty and qualification) and boosters (markers of confidence and certainty) were measured per 1,000 words using an automated 233-item lexicon. The epistemic shift index (ESI) quantified the net difference in epistemic stance between sections, with a positive value indicating greater caution in the Discussion. Abstract limitation surfacing (ALS) indicated explicit acknowledgment of limitations in the Abstract.

RESULTS: Discussion sections contained 2.8 times the hedge density of Abstracts, and the ESI was positive in 176 (87.6%) of the 201 articles. The mean ESI was 7.8 (95% confidence interval, 6.9 to 8.8; t[200] = 16.39, p < 0.001; Cohen d = 1.15). Discussions carried more than double the hedge-to-booster ratio of Abstracts (3.58 versus 1.69, p < 0.001). The asymmetry did not differ by open-access status (p = 0.807) or journal (p = 0.142) but varied significantly by study design (Kruskal-Wallis p = 0.034), with the largest asymmetry in diagnostic studies and randomized controlled trials. Only 12 (6.0%) of 201 Abstracts acknowledged study limitations.

CONCLUSIONS: The Discussion sections of the orthopaedic research articles were substantially more epistemically cautious than their Abstracts across journals and access models. In most articles, the Abstract presented findings without caveat, with corresponding limitations appearing only in the Discussion. A limitation of this analysis is that epistemic stance was measured at the word and phrase level only. These findings have implications for clinical decision-making and for artificial intelligence systems trained on freely accessible Abstract text.