Some concerns about imputation methods for missing data

ABSTRACT

Comments on an article by RJ Van Lieshout et al. (see record 2022-41504-002). Van Lieshout and colleagues1 conducted a randomized clinical trial of an online 1-day cognitive behavioral therapy workshop vs waiting list in a population at risk of postpartum depression and evaluated the effect using the Edinburgh Postnatal Depression Scale, a patient-reported outcome measure of depressive symptom severity. Second, although they used LMM, we have concerns about whether this method adequately imputed the missing outcomes. In general, repeated outcome measurements are required to correctly estimate the missing values using LMM. However, this study had only a single follow-up point, and in LMM the baseline measure must be used as a covariate and not as an outcome. We wonder how such a model could have accounted for missing outcomes. More details about the specific LMM methods used would be helpful. (PsycInfo Database Record (c) 2023 APA, all rights reserved)