The evolution of reporting statistical inference in abstracts of obstetrical studies from 2013 to 2023

Selective reporting of research results based on statistical significance compromises research validity, potentially misleading clinical decision-making and future research. In obstetrics, the extent of this issue remains unclear. This study aimed to characterize the frequency, patterns, and temporal trends in the reporting of P-values, effect sizes, and statistically significant results in abstracts of obstetric studies from 2013 to 2023.

We retrieved abstracts in the field of obstetrics between January 1, 2013, and December 31, 2023, from Medline, Embase and Cochrane CENTRAL. Automated text-mining was performed to detect and extract reporting of statistical inference, including P-values, effect sizes, Bayesian-related statistics, confidence intervals, and textual descriptions. The extracted statistical inferences were analyzed to assess trends over time and overall distribution, as well as specific patterns across different study designs.

A total of 23 167 eligible obstetric studies were identified from 46 788 abstracts. The proportion of abstracts reporting only P-values remained relatively stable over time, from 28.5% (95% confidence interval [CI]: 26.5%-30.5%) in 2013 to 27.6% (95% CI: 25.7%-29.5%) in 2023. There was a consistent rise in the proportion of abstracts reporting effect sizes, whether alone or alongside P-values, increasing from 22.1% (95% CI: 20.3%-24.0%) in 2013 to 39.5% (95% CI: 37.4%-41.7%) in 2023. Abstracts that reported neither P-values nor effect sizes decreased from 49.4% (95% CI: 47.2%-51.6%) in 2013 to 32.9% (95% CI: 30.8%-34.9%) in 2023. Most reported P-values clustered around common cut-offs, with 30.7% at 0.001 and 31.5% at 0.05. Among abstracts that reported statistical significance, 89.0% (95% CI: 87.4%-90.5%) reported a statistically significant difference, and the trend has remained stable over the past decade. Randomized controlled trials reported a lower proportion of statistically significant statements (82.4%, 95% CI: 75.9%-88.8%) than other study types.

Although the reporting of effect sizes has gradually increased over time, the use of standalone P-values remains common. The consistently high proportion of abstracts presenting at least one statistically significant result might reflect entrenched reporting practices in the field.