Do longitudinal orthodontic trials use appropriate statistical analyses? A meta-epidemiological study

Abstract

Background: In orthodontic trials, longitudinal designs with multiple outcome measurements over time are common. The aim of this epidemiological study was to examine whether optimal statistical analysis approaches have been used in longitudinal orthodontic trials.

Methods: Pubmed was searched in August 2021 for longitudinal orthodontic trials with at least three time points of outcome assessment published in the 2017–20 period. Study selection and data extraction were done independently and in duplicate. The analysis approaches undertaken were tabulated and associations between study characteristics and the use of optimal analysis or not were assessed using Fisher’s exact test and logistic regression.

Results: One hundred forty-seven out of 563 unique records were deemed eligible for inclusion. Only 26.50% of these trials used an optimal statistical analysis for longitudinal data where the data structure is accounted for. None of the study characteristics except the statistical significance of the results were associated with the appropriateness of the statistical analysis. The odds of significant results in studies with suboptimal analyses were higher than that in studies with optimal longitudinal analyses (odds ratio: 3.48, 95% confidence interval: 1.62, 7.46, P = 0.001). For the studies with optimal analysis, the most frequent test was repeated-measure analysis of variance (RMANOVA). The reporting of the statistical analysis section was suboptimal in the majority of the trials.

Conclusion: Most longitudinal orthodontic trials are not analysed using optimal statistical approaches. Inferences and interpretation of their results are likely to be compromised.