Confidence interval reporting for measures of association in multivariable regression models in observational studies

Abstract

Background/objectives: To assess the adherence to reporting confidence intervals (CI) for measures of association in multivariable regression models (MRM) in articles with observational design indexed in MEDLINE.

Material and methods: A literature search was conducted using the MEDLINE bibliographic database to obtain a representative sample of studies with observational design and applying MRM (logistic, linear, and Cox regression) (n=428). Proportions and 95% CI of articles reporting CI for measures of association in MRM were calculated. Percentage ratios (PRs) were also calculated to describe the change in CI reporting before and after the publication of the STROBE statement.

Results: 188 of the 236 abstracts with measures of association (79.7%; 95% CI 74.5, 84.8) and 360 of the 428 main texts (84.1%; 80.6, 87.6) were provided with CI. A non-significant increase of 1% in the abstract, PR=1.01 (0.77, 1.29), and 7% in the main text, PR=1.07 (0.87, 1.28), occurred in the CI reporting after the publication of the STROBE guideline.

Conclusions: The STROBE guideline recommendation on reporting CI should be more thoroughly followed.

Keywords: Confidence interval; Guía STROBE; Intervalo de confianza; P-valor; P-value; STROBE guideline.