Statistical analysis in orthodontic journals: are we ignoring confounding?

AIM

To assess the prevalence of adjustment for confounding within statistical analysis and matching at the design stage in leading orthodontic journals and to explore potential associations between accounting for confounding and publication characteristics.

MATERIALS AND METHODS

Twenty-four issues of four leading orthodontic journals with the highest impact factor were searched from July 2014 backwards. Confounding adjustment through statistical analysis and study characteristics including journal, study design, region of origin, number of authors, number of centres, involvement of a statistician, significance of results, and type of analysis were recorded. Reporting of matching at the design stage was also recorded.

RESULTS

Of 426 studies identified, only 71 (17 per cent) accounted for confounding in the statistical analysis. There was evidence that journal, country of authorship, and involvement of a statistician (odds ratio = 3.91, 95 per cent confidence interval: 2.16-7.10; P < 0.001) were significant predictors of accounting for confounding at the analysis level. Reporting of matching at the design stage was identified in 111 of 426 (26 per cent) studies in which 9 studies adjusted for confounding at the analysis level.

CONCLUSIONS

Appropriate adjustment for confounding in orthodontic literature either at the design or at the analysis stage was identified in less than half of studies overall (41 per cent), suggesting lack of expertise and awareness in design, conduct, analysis, and reporting of non-randomized studies in this field. This is a critical limitation that can potentially result in biased estimates and associations between examined exposures and outcomes.