Longitudinally collected outcomes are often analysed using repeated hypothesis tests as if they had been cross-sectional observations. This does not account for the correlation among measurements taken on the same experimental unit, which if incorrectly addressed in the analysis, may lead to an underestimation of the variability of the analysis units, which in turn may lead to false statistical significance when treatment group means are compared.
A survey of papers published in Molecular Therapy during the last 10 years identified 67 longitudinal studies. The majority (93%) conducted statistical analyses, but only 18% addressed within analysis unit correlation and conducted longitudinal analyses. Cross-sectional analyses were performed in 75% of the studies. The most common approach was analyzing each time point data separately using ANOVA.
The authors conclude that the results imply that longitudinal data are not adequately analyzed in the fields of cell biology and gene therapy.
1. Liu C, Cripe TP, Kim MO. Statistical Issues in Longitudinal Data Analysis for Treatment Efficacy Studies in the Biomedical Sciences. Mol Ther 2010;18:1724–1730.