ACoP13 Will Be Live! 13

ACoP 2021: Statistics and PMx e.g. MBMA, Bayesian application/method, trial design, optimal design, machine learning, data mining
Rhoda Muse

Modeling informative drop out in a disease progression model of Duchenne Muscular Dystrophy

Objectives:

Duchenne Muscular Dystrophy (DMD) is a genetic disorder that predominantly affects boys, which causes progressive muscle loss resulting in loss of ambulation, difficulty breathing, cardiomyopathy and ultimately death. Therapy development for DMD has progressed, but there remains no cure or treatment that significantly delays disease progression, so better therapies are needed. Accurately modeling the progression of the disease is critical for optimizing clinical trial design to test such novel therapies. There are multiple endpoints that have been used in clinical trials, including functional tests such as the time to climb 4 stairs, modeled here as a velocity and termed “velocity climb”.  It has been observed that dropout from DMD clinical trials does not occur at the same rate across the population.  Subjects whose disease progresses at different rates also drop out of trials at different rates.  Thus, the objective of this work is to develop a joint model to account for this informative drop out and disease progression of the of DMD patients using velocity climb.

Methods:

Patient level data from 14 studies (clinical trial placebo arms and natural history studies) had been previously standardized, curated, and integrated.  A subset of data was extracted comprising 841 individuals with 5,358 longitudinal measurements for velocity climb (4 studies reserved for external validation).  The longitudinal marker was best modeled using a non-linear mixed effects model (the product of the Chapman-Richards growth function and a sigmoid Imax model) and the survival subcomponent for informative dropout was modelled using a Weibull baseline hazard. Two shared random effects  were used to link dropout to the longitudinal markers. Covariates included in the dropout sub-model were: baseline score and study type. Covariates included in the longitudinal sub-model included: baseline age, baseline score and mutation group.  The analysis was conducted using the Bayesian ‘rstan’ package in R using non-informative priors.

Results

The coefficient on the two shared random effects in the drop out sub model shows evidence of informative dropout with 95% credible interval (0.2358, 4.104) while there was no evidence that the second random effect was associated with dropout with 95% credible interval (-1.3441, 1.6242).  This suggests that there is sufficient evidence that the data are missing not at random. Patients with higher baseline score for velocity climb also had a longer time to dropout while older patients at baseline had a higher dropout rate.

Conclusions

Modelling the informative dropout reduces biases in the parameter estimates and predicted drop out times in clinical trial simulations.  The model developed, which allows for the prediction of individualized trajectories accounting for inter-subject variability, is intended to serve as the basis for a clinical trial simulation tool to help optimize efficacy studies for DMD. 

 

 

 

 

 





Author(s)
  • Rhoda Muse, Critical Path Institute (Presenting Author)
  • Jane Larkindale, Critical Path Institute (CoAuthor)
  • Sarah Kim, University of Florida (CoAuthor)
  • Stephan Schmidt, University of Florida (CoAuthor)
  • Juan Francisco Morales, University of Florida (CoAuthor)
  • Sudhir Sivakumaran, Critical Path Institute (CoAuthor)
  • Ramona Belfiore-Oshan, Critical Path Institute (CoAuthor)
  • Varun Aggarwal, Critical Path Institute (CoAuthor)
  • Klaus Romero, Critical Path Institute (CoAuthor)
  • Jackson Burton, Critical Path Institute (CoAuthor)
  • Diane Corey, Critical Path Institute (CoAuthor)



Reference: ACoP12 (2021) STPM-198 [www.go-acop.org/?abstract=198]
Statistics and PMx e.g. MBMA, Bayesian application/method, trial design, optimal design, machine learning, data mining
Top