Predicting Treatment Effect on the Loss of Ability to Rise From Floor in Duchenne Muscular Dystrophy Using Individual Trajectories: An Application of Joint Modeling in FOR-DMD
Author(s)
Tasnim Hamza, PhD1, Keith R. Abrams, BSc, MSc, PhD2, Rossella Belleli, MA, MPH1.
1F. Hoffmann-La Roche, Ltd, Basel, Switzerland, 2Visible Analytics, Reading, United Kingdom.
1F. Hoffmann-La Roche, Ltd, Basel, Switzerland, 2Visible Analytics, Reading, United Kingdom.
OBJECTIVES: This analysis aimed to develop a model connecting longitudinal time to rise from floor (TTR) trajectories with time-to-event (TTE) data for loss of ability to rise from the floor (LoR) to characterize Duchenne Muscular Dystrophy (DMD) progression and predict future treatment effects. The FOR-DMD trial (NCT01603407) included 196 boys with DMD aged 4-7, randomized to daily prednisone, daily deflazacort, or intermittent prednisone. TTR was collected at baseline, 3, 6, and every 6 months until 60 months.
METHODS: A Bayesian joint model was developed, combining a longitudinal sub-model for TTR trajectories and a TTE sub-model for LoR. The longitudinal sub-model, a linear mixed-effects model, incorporated corticosteroid regimen-specific slopes and a quadratic function for log-transformed TTR over time. It included random effects for intercept and slope and was adjusted for baseline age, North Star Ambulatory Assessment (NSAA) total score, and corticosteroid regimen. The TTE sub-model was a parametric proportional hazards regression model that modeled time from randomization assuming a Weibull baseline hazard. It was adjusted for the same covariates, and longitudinal trajectories were incorporated using a current value association structure. The joint model was estimated using the rstanarm R package and JAGS.
RESULTS: The joint model estimated the association between TTR trajectory and LoR (4.7; 95% CrI: 3.4-6.4). A 16% annual decrease in TTR for daily vs. intermittent corticosteroids resulted in a LoR hazard ratio of 0.43 (95% CrI: 0.32, 0.55). These results were robust to various model adjustments, including removing covariates, adding interactions, using cubic splines for baseline hazard, and altering priors.
CONCLUSIONS: A joint model, developed using the FOR-DMD dataset, provides a robust framework for predicting treatment effects on the LoR in DMD patients, utilizing individual TTR trajectories and other covariates. This approach can predict the effect of new treatments on LoR based on observed TTR trajectories.
METHODS: A Bayesian joint model was developed, combining a longitudinal sub-model for TTR trajectories and a TTE sub-model for LoR. The longitudinal sub-model, a linear mixed-effects model, incorporated corticosteroid regimen-specific slopes and a quadratic function for log-transformed TTR over time. It included random effects for intercept and slope and was adjusted for baseline age, North Star Ambulatory Assessment (NSAA) total score, and corticosteroid regimen. The TTE sub-model was a parametric proportional hazards regression model that modeled time from randomization assuming a Weibull baseline hazard. It was adjusted for the same covariates, and longitudinal trajectories were incorporated using a current value association structure. The joint model was estimated using the rstanarm R package and JAGS.
RESULTS: The joint model estimated the association between TTR trajectory and LoR (4.7; 95% CrI: 3.4-6.4). A 16% annual decrease in TTR for daily vs. intermittent corticosteroids resulted in a LoR hazard ratio of 0.43 (95% CrI: 0.32, 0.55). These results were robust to various model adjustments, including removing covariates, adding interactions, using cubic splines for baseline hazard, and altering priors.
CONCLUSIONS: A joint model, developed using the FOR-DMD dataset, provides a robust framework for predicting treatment effects on the LoR in DMD patients, utilizing individual TTR trajectories and other covariates. This approach can predict the effect of new treatments on LoR based on observed TTR trajectories.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
P27
Topic
Health Technology Assessment, Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
Pediatrics, Rare & Orphan Diseases