Paolo Bettica2
1Certara, 2Italfarmaco SpA
Introduction: Duchenne muscular dystrophy (DMD) is a severe progressive muscular disease caused by mutations in the X-linked gene encoding for the dystrophin protein [1]. DMD affects both skeletal and smooth muscle and is associated with degenerative loss of function. The onset of symptoms occur at an early age and most children are no longer able to walk by 12 y and have a life expectancy of 20-30 y. The condition can be characterised by one of several different disease progression (DP) scales with the NorthStar Ambulatory Assessment (NSAA) being the most general composite score. NSAA comprises 17 questions rated 0-2 with a range of scores from 0 to 34 (normal activity). A DP model and Clinical Trial Simulation tool has been developed previously for DMD by by C-Path’s Duchenne Regulatory Science Consortium (D-RSC) [2] who used a Chapman-Richard growth model linked to a sigmoid Emax model. While the D-RSC model provided a good description of the available data, the DP was found to be delayed for patients with higher baseline score and for older patients which contrasts with the general heuristic understanding of DMD. Survivorship bias has been well documented for observational studies [3] and may also occur for clinical trials. It is associated with biased (often better) outcomes for patients on a treatment as those who are well enough to meet the entry criteria of a study may do better than those who do not. It has particular importance when considering DP models which are designed to quantify the time-course of a disease. Objectives: To investigate whether the delay in onset of DP for DMD in older children could be a function of survivorship bias. Methods: A simulation-estimation study was setup to address the question as to whether it is possible that the use of selection criteria for children entering clinical trials could affect inference about DP. A simulation model was constructed for NSAA that was identical to that developed by D-RSC[2] with the exception that the DP covariates resulting in a delay were removed so that the model no longer simulated a delay in DP for older children and with higher baseline score. This involved removal of 2 covariate effects, baseline age (BAGE) and baseline score (BSCORE). The estimation step was done using the original model (ie, including those covariate effects). Two data sets were simulated. For both datasets the minimum inclusion NSAA baseline score (BSCORE) was 10 and minimum age was 5y with the LLOQ of the BSCORE of 3. Simulation datasets: (S1) a single long longitudinal study where all patients were enrolled at the age of 5 years and followed for 20 y; (S2) 6x short longitudinal studies where eligible patients were between the ages of 5 and 20 y of age, were studied for 3 years with NSAA repeat scores every 6 months. The data were analysed using NONMEM v7.5. Results: The simulation model contained no influence of BAGE or BSCORE on the clinical trial data. The final estimation from S1 data revealed a negative influence of BSCORE on DP (low baseline scores were associated with more rapid disease progression). This effect is possibly due to LLOQ censoring. The final estimation model from the S2 dataset found a positive influence of both BAGE and BSCORE in delaying DP which aligns with the result of Lingineni [2] both indicating that higher BAGE and BSCORE delayed DP. Conclusions: This simulation exercise highlighted the potential for selection criteria to influence apparent covariate effects. The cause of this bias was the introduction of selection criteria for entry into the short duration longitudinal studies, where those patients who had slow disease progression were preferentially selected (e.g. a 20y old subject must have a NSAA score > 10). This did not affect the long-longitudinal study since few if any 5y old patients failed to meet the minimum BSCORE criteria of 10. These results do not formally state that the DP model has been affected by survivorship bias but rather such an effect cannot be ruled out.
[1] Angulski et al. Frontiers Physiol 2023;14:1183101 [2] Lingineni et al. CPT Pharmacometrics Syst Pharmacol. 2022;11:318–332 [3] Austin et al. J Eval Clin Pract 2006;12:601-612
Reference: PAGE 33 (2025) Abstr 11413 [www.page-meeting.org/?abstract=11413]
Poster: Drug/Disease Modelling - Paediatrics