ACoP 2022: Statistics and PMx e.g. MBMA, Bayesian application/method, trial design, optimal design, machine learning, data mining
Varun Aggarwal

Multivariate Joint Models to Predict Clinically Meaningful Decline in Duchenne Muscular Dystrophy as measured by NSAA Using Timed Function Test Trajectories.

Objective: Duchenne muscular dystrophy (DMD) severity is measured using several endpoints which are relevant at the various stages of the disease. The North Star Ambulatory Assessment (NSAA) is one such scale used in ambulatory DMD patients. NSAA is comprised of 17 items which capture various aspects of ambulation, each scored on a 0 (unable to achieve goal), 1 (modified method but achieves goal), 2 (achieves goal normally) scale. We previously developed a disease progression model for NSAA1, predicting the trajectory of NSAA using baseline covariates such as age, steroid use, study type, and genetic mutation. This model did not account for the fact that items composing NSAA could be measured on a continuum by timed function tests (TFT) such as time to stand (velocity stand), time to climb (velocity climb) and time to walk and run (velocity walk and run). A change in the TFT score which might not be captured by the discrete 0, 1, 2 item scores on NSAA might be important in predicting the NSAA trajectory. Thus, we developed a joint model that predicts a meaningful change (30%) from baseline in the NSAA using predicted TFT trajectories.  

Methods: A training dataset consisting of 475 subjects with NSAA measurement was created from the Duchenne Regulatory Science Consortium (D-RSC) database. Further filtering of subjects with only one longitudinal measurement resulted in 443 patients. This dataset resulted in three datasets which described overlap of NSAA with each of the TFTs of interest: velocity climb, velocity walk and run, and velocity stand. Time to event (30% decline in NSAA score) was defined for each of these datasets. Similarly, a validation dataset consisting of 312 subjects with NSAA measurements (289 with >1 longitudinal measurement) was created. For the joint modeling, first a survival sub model describing time to 30% decline was created for these three datasets. Next, following the approach in Lingineni et al1, longitudinal sub models that described and predicted the trajectory of the TFTs were developed and validated. Finally, three separate joint models were created which incorporated effect of the TFTs on the time to NSAA decline. R version 4.0.1 was used for modeling. 

Results: The joint modeling showed that the association factor for each of the TFTs was significantly different from zero, suggesting that TFTs affect the time to decline in NSAA score. A unit increase in the history of velocity climb increased this time by a factor of 3.83, a unit increase in the history of predicted velocity stand resulted in an increase by a factor of 10,000, and a unit increase in predicted value of velocity walk and run reduced the risk of decline by 15%.   

Conclusion: Time to a clinically meaningful decline in NSAA is significantly dependent on the trajectories of TFTs. Incorporating this information in the clinical trial simulators could provide a method to better design clinical trials and select trial populations. 

1 Lingineni K, Aggarwal V, et al. Development of a model-based clinical trial simulation platform to optimize the design of clinical trials for Duchenne muscular dystrophy. CPT Pharmacometrics Syst Pharmacol. 2022 Mar;11(3):318-332. doi: 10.1002/psp4.12753 PMID: 34877803; PMCID: PMC8923721. 


  • Varun Aggarwal, Critical Path Institute (Presenting Author)
  • Rhoda Muse, critical path institute (CoAuthor)
  • Sarah Kim, University of Florida (CoAuthor)
  • Jackson Burton, Critical path (CoAuthor)
  • Diane Corey, critical path institute (CoAuthor)
  • Lauren Quinlan, critical path institute (CoAuthor)
  • Klaus Romero, Critical (CoAuthor)
  • Ramona Belfiore-Oshan, critical path institute (CoAuthor)
  • Terina Martinez, critical path institute (CoAuthor)

Reference: ACoP13 (2022) STPM-328 []
Statistics and PMx e.g. MBMA, Bayesian application/method, trial design, optimal design, machine learning, data mining