Alternatives to MMRM for preclinical Alzheimer's clinical trials: Clinical trial design and implementation. (7th December 2020)

• Record Type: Journal Article
• Authors: Donohue, Michael C.; Sethuraman, Gopalan; Langford, Oliver; Lin, Wenyi; Insel, Philip; Thompson, Wesley K.; Raman, Rema; Sperling, Reisa A.; Aisen, Paul S.
• Is Part Of: Alzheimer's & dementia. Volume 16(2020)Supplement 9
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-12-07
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.044915
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: The Mixed Model for Repeated Measures (MMRM; Mallinckrodt, et al. 2001) is the most commonly used approach for assessing treatment effects in Alzheimer's clinical trials. An alternative nonlinear Disease Progression Model (DPM) which assumes the ratio of group means is fixed over time has been proposed for DIAN‐TU (e.g. Wang, et al. 2018). We assess nonlinear models and alternative linear models for Preclinical Alzheimer's clinical trials like the A4 Study (Sperling, et al. 2014). Method: Tables 1 and 2 describe the mean and correlation structures that we considered. Models were fit to data from cognitively normal ADNI participants (amyloid positive vs negative). We compared mean PACC trajectories over time and Akaike Information Criterion (AIC; Sakamoto, et al 1986). All models were fit by maximum likelihood using the R package nlme. Simulations were used to assess power and Type I error for clinical trials in Preclinical Alzheimer's (Figure 1). Results: Figure 2 demonstrates the various mean structures fit to ADNI. Models with stronger shape assumptions are smoother than the unstructured mean of MMRM (dashed lines). The DPM‐like nonlinear model with fixed group mean ratio ("NL0") provides a relative underestimate of the amyloid group difference at final visit. Comparing values of AIC (Figure 3), we find little evidence to support using alternatives to the unstructured mean and variance of the typical MMRM. Simulation studies (Table 3) suggest that power can be improved with simpler mean structures (Hybrid or Quadratic), while maintaining good Type I error control with an unstructured variance. It was difficult to fit the nonlinear models with the full suite of correlation structures. The NL0 could be reliably fit with random intercept and heterogeneous variance but showed no advantage over the simpler Hybrid or Quadratic mean structures. Conclusion: ADNI data support the use of the MMRM with unstructured mean and variance, but simulations suggest simpler mean structures might provide a modest improvement in the power from 85%, with an unstructured mean, to 90%, with Hybrid or Quadratic mean. Nonlinear DPM‐like models demonstrated no advantage over linear model alternatives in ADNI and in simulations of Preclinical Alzheimer's clinical trials.