The mixed model for repeated measures for cluster randomized trials: a simulation study investigating bias and type I error with missing continuous data

Abstract

Background: Cluster randomized trials (CRTs) are a design used to test interventions where individual randomization is not appropriate. The mixed model for repeated measures (MMRM) is a popular choice for individually randomized trials with longitudinal continuous outcomes. This model's appeal is due to avoidance of model misspecification and its unbiasedness for data missing completely at random or at random. Methods: We extended the MMRM to cluster randomized trials by adding a random intercept for the cluster and undertook a simulation experiment to investigate statistical properties when data are missing at random. We simulated cluster randomized trial data where the outcome was continuous and measured at baseline and three post-intervention time points. We varied the number of clusters, the cluster size, the intra-cluster correlation, missingness and the datageneration models. We demonstrate the MMRM-CRT with an example of a cluster randomized trial on cardiovascular disease prevention among diabetics. Results: When simulating a treatment effect at the final time point we found that estimates were unbiased when data were complete and when data were missing at random. Variance components were also largely unbiased. When simulating under the null, we found that type I error was largely nominal, although for a few specific cases it was as high as 0.081. Conclusions:Although there have been assertions that this model is inappropriate when there are more than two repeated measures on subjects, we found evidence to the contrary. We conclude that the MMRM for CRTs is a good analytic choice for cluster randomized trials with a continuous outcome measured longitudinally.Trial registration: ClinicalTrials.gov, ID: NCT02804698.

Episodes of Undernutrition and its Predictors among Clients on Antiretroviral Treatment in Southwest Ethiopia: A Record Review

“…It considers the repeated and dependent nature of the data for reliable parameter estimates. 31 Significant improvements in the BMI score of ART clients were observed after ART initiation (P<0.0001). Accordingly, the mean BMI score at 24 months is better than the previous follow-up periods.…”

“…Akaike's information criteria (AIC) was used to assess model fitness; with smaller values considered as complex in estimating parameter estimates. 31 , 32 P -value of less than 0.05 used as cut off point to declare statistical significance, the alpha level to reject the null hypothesis.…”

Effect of missing values in multi‐environmental trials on variance component estimates

“…• Variance component (VC) estimates can be biased due to the use of likelihood-based methods. (Bell & Rabe, 2020) are estimated using the restricted/residual maximum likelihood (REML) method (Patterson, 1997;Patterson & Thomson, 1971). As mixed models are commonly fitted with REML, a MAR data pattern can be ignored for this type of analysis.…”

“…This theoretical claim has been confirmed by several simulation studies in animal breeding (Appel et al, 1998;Van Tassel et al, 1995;Cantet et al, 2000;Duangjinda et al, 2001;Schenkel et al, 2002;Sorensen & Kennedy, 1984;Van der Werf and De Boer, 1990) and cultivar evaluation trials (Piepho & Möhring, 2006;Hartung et al, 2021). Note that this means that in case of a MAR data pattern, selection can be ignored in the analysis, if the model is not mis-specified and the variances as well as BLUPs and BLUEs (Bell & Rabe, 2020) are estimated using the restricted/residual maximum likelihood (REML) method (Patterson 1997;Patterson & Thomson, 1971). As mixed models are commonly fitted with REML, a MAR data pattern can be ignored for this type of analysis.…”

Effects of an Internet-Based and Teacher-Facilitated Sexuality Education Package: A Cluster-Randomized Trial

“…Second, as the proportion of the participants lost to follow-up was significantly higher in the control group, compared with the intervention group, there could have been attrition bias, decreasing the comparability of the two arms. Although our use of the analysis of repeated measurements model was reported to be robust when treating randomized lost to follow-ups [32], an additional sensitivity analysis on participants who participated in all three rounds of the survey was conducted and the results of sensitivity analyses indicate no statistical differences from the main analysis. Third, due to the low prevalence of sexual behaviors among the participants, we could not assess the package's effects on influencing adolescents' sexual behaviors and other related sexual risks such as STIs.…”