Participant followup rate can bias structural imaging measures in longitudinal studies

Abstract

Longitudinal MRI analysis is essential to accurately describe neuroanatomical changes over time. Loss of participants to followup (dropout) in longitudinal studies is inevitable and can lead to great difficulty in interpretation of statistical results if dropout is correlated with a study outcome or exposure. Beyond this, technical aspects of longitudinal MRI analysis require specialised processing pipelines to improve reliability while avoiding bias towards individual timepoints. In this article we test whether there is an additional problem that must be considered in longitudinal imaging studies, namely whether dropout has an impact on the function of FreeSurfer, a popular software pipeline used to estimate important structural brain metrics. We find that the number of acquisitions available per individual can impact the estimation of cortical thickness and brain volume using the FreeSurfer longitudinal pipeline, and can induce group differences in brain metrics. The effect on trajectories of brain metrics is smaller than the effect on brain metrics. HighlightsO_LILongitudinal MRI analysis is essential to accurately track neuroanatomical changes over time C_LIO_LILongitudinal MRI analysis requires specialised processing pipelines to reduce bias towards single timepoints C_LIO_LIParticipant drop out or loss can bias neuroanatomical measures derived from longitudinal pipelines C_LIO_LIWe find that group differences in the number of acquisitions available to analyse can cause group differences in estimated cortical thickness and brain volume C_LIO_LIThis bias appears to be due to the number of scans used to create individualised templates in the Freesurfer longitudinal pipeline C_LIO_LIThe effect on estimates of brain metric trajectories appears smaller than the effect on the estimates of brain metrics C_LI