Accounting for clustering is still not routinely undertaken in orthodontic studies

Abstract

Summary Background The aim of this study was to identify studies with clustering effects published in orthodontic journals and to assess the frequency by which clustered designs are correctly accounted for in the statistical analysis. Factors associated with appropriate management of clustering effects during the statistical analysis were explored. Materials and method A search of three leading orthodontic journals was undertaken to identify studies with clustering effects published between 1 January 2019 and 31 December 2021. Descriptive statistics and frequency distributions were calculated. Associations between the correct statistical handling of clustering effects and study characteristics were explored via univariable and multivariable analyses. Results Three hundred and sixty-two studies were considered to have clustering effects. Only 22.4 per cent of studies correctly accounted for clustering effects in the statistical analysis with no improvement compared to a previous study in 2012 using the same journals. An association between the use of the correct analysis to account for clustering and the following study characteristics was detected: involvement of statistician, study type, significance of the results, and accounting for clustering in the sample size calculations. In the univariable analysis, interventional studies, non-significant results, and the involvement of a statistician were associated with higher odds of the use of the correct analysis. Of the studies correctly accounting for clustering, the most used tests were repeated measures ANOVA (43.3 per cent) and mixed models (40.7 per cent). Conclusions Compared to previous research, there appears to be no improvement in accounting for clustering effects in studies published in orthodontic journals. To prevent incorrect inferences being drawn, clustering effects need to be recognised and accounted for in orthodontic studies. Recommendations to improve the accounting of clustering effects, at both the study level and during the statistical analysis are suggested.