Reporting of Statistical Inference in Abstracts of Major Cancer Journals, 1990 to 2020

IMPORTANCE: Since the 1990s, reporting guidelines have developed that uniformly require authors to report a measure of precision (confidence intervals [CIs]) in addition to effect size. OBJECTIVE: To investigate the time trend of statistical inference and statistical reporting style in abstracts of major cancer journals. DESIGN, SETTING, AND PARTICIPANTS: This cross-sectional study reviewed all abstracts published between January 1, 1990, and December 31, 2020, in 10 high-ranking cancer journals (Lancet Oncology, Journal of Clinical Oncology, Cancer Discovery, Cancer Cell, JAMA Oncology, Annals of Oncology, Molecular Cancer, Journal of Thoracic Oncology, Journal of the National Cancer Institute, and Trends in Cancer) using a previously validated computerized algorithm to search the PubMed database. For the time trend analyses, 2 journals with only a few years of existence (JAMA Oncology and Trends in Cancer) were excluded. EXPOSURES: Calendar year, journal, and type of abstract (randomized clinical trial or other). MAIN OUTCOMES AND MEASURES: Proportions of abstracts containing CIs, P values without CIs, and qualitative expressions of statistical significance only were compared over time among journals. RESULTS: Overall, 24 034 of 42 509 abstracts (56.5%) contained statistical inference. Reporting of CIs increased over time in 5 of 8 journals. From 2016 to 2020, the most prevailing statistical reporting style was the presentation of CIs (3070 of 4895 [62.7%]). However, the proportion of abstracts reporting statistical inference based solely on the terms significant or nonsignificant was still 1195 of 4895 (24.4%) during this period and was most prevalent among basic science–oriented cancer journals (eg, 63 of 66 [95.5%] in Cancer Cell). A higher prevalence of CI reporting was associated with reporting of results from randomized clinical trials and the requirement to report according to guidelines (eg, 522 of 574 [90.9%] in Lancet Oncology). CONCLUSIONS AND RELEVANCE: These findings suggest that the reporting style of statistical inference in abstracts of major cancer journals has improved over time. A requirement in journals’ instructions for authors to present statistical inference in accordance with reporting guidelines and the implementation of these guidelines in submitted manuscripts on the part of journal editors may improve reporting.