Quantitative Assessment of Statistical Reviews of Patient Safety Research Articles

Abstract

Objectives For 8.5 consecutive years, all patient safety articles of a journal underwent statistical review before publication. We sought to establish the prevalence of statistical themes in the statistical reviews, consideration of contemporary statistical methods, and their associations with time to journal receipt of authors' revision. Methods An initial set of statistical themes was created using the statistical editor's notes. For example, for the statistical theme of "CONSORT checklist," the search term needed was "CONSORT." A complete (exhaustive) list of additional themes was obtained inductively. Results Among the 273 subsequent reviews for manuscripts that were ultimately accepted, the number of paragraphs that included a theme of a statistical method was only weakly associated with longer revision times (Kendall τ = 0.139 ± 0.039, P = 0.0004). Among the total 3274 paragraphs of statistical reviews, 72.2% did not include a theme of a statistical method (e.g., the editor instead asked the authors to clarify what statistical method had been used) (95% confidence interval [CI] = 70.6%-73.7%, P < 0.0001 versus 50%). Among the 207 manuscripts with a review that included a statistical method, 47.3% included a contemporary topic (e.g., generalized pivotal methods) (95% CI = 40.4%-54.4%). However, among the 911 corresponding paragraphs of statistical review comments, only 16.0% included a contemporary theme (95% CI = 13.7%-18.6%). Conclusions The revised versions of patient safety articles, which are eventually to be accepted for publication, have many statistical limitations especially in the reporting (writing) of basic statistical methods and results. The results suggest a need for education of patient safety investigators to include statistical writing.