Lack of confidence interval reporting in dermatology: a call to action.

Abstract

Confidence intervals (CIs) offer a complete and intuitive presentation of results and are more informative than P-values. The current prevalence of CI reporting in the dermatology literature has not been discussed. To evaluate CI reporting prevalence in the dermatology literature, factors predicting reporting and to compare CI reporting in the dermatology literature with dermatology research in the New England Journal of Medicine (NEJM). MEDLINE was queried for trials and observational studies published from 2007 to 2017 in six dermatology journals with varying impact factors: British Journal of Dermatology, Journal of Investigative Dermatology, JAMA Dermatology, Journal of the American Academy of Dermatology, European Journal of Dermatology and Journal of Cosmetic Dermatology. Randomly selected studies were reviewed to calculate CI reporting prevalence, and logistic regression identified factors predictive of CI reporting. Of 97 studies meeting the inclusion criteria, only 21 (22%; 95% CI 14-31%) reported CIs for the primary outcome. Of the remaining 76 studies, 43 reported a measure of variance (57%; 95% CI 45-67%). More recent year of publication was the strongest predictor for CI reporting [odds ratio 1·44; 95% CI 1·10-1·89 (P=0·01)] in multivariable analysis. CI reporting in NEJM was significantly higher than in the dermatology literature (64% vs. 22%; P<0·001). CI reporting in the dermatology literature is low. We urge both dermatology journals and researchers to improve clinical interpretation of study results by taking steps to increase CI reporting.