PUL 2.0 is a key functional measure for assessing disease progression in DMD, especially after loss of ambulation, and has been used as an outcome in clinical trials. Predictive models for PUL trajectories can be used to contextualize trial outcomes, especially when there are no placebo controls. Additionally, understanding which patient characteristics are predictive of future changes in PUL can help inform clinical trial design. We developed a multivariable prognostic model for multi-year changes in PUL based on single-center real-world data (RWD) from patients with DMD.
PUL data were modelled at site with RWD as input and without the need for sharing RWD. Change in PUL total score (ΔPUL) was modeled over time as a function of baseline age, PUL, and ambulatory status. Baseline was defined as the first visit with PUL entry item score of 1-5. Predictive performance was evaluated using 5-fold cross-validation (5-CV).
Sixty-nine patients were included with a mean age of 14.3 y (SD 5.9) at baseline, of which 52 (75.4%) were non-ambulant. At baseline, mean PUL was 25.9 (SD 11.1) and 37 patients (53.6%) had a PUL entry item score of 5, while 3 (4.4%), 10 (14.5%), 4 (5.8%) and 15 patients (21.7%) had an entry item score of 4, 3, 2, and 1, respectively. Annualized ΔPUL was -1.07 and the average patient had 3.5 PUL assessments post-baseline spanning a mean of 3.7 years of follow-up. The model including baseline age explained 55% of variation in ΔPUL with mean 5-CV prediction errors of ±4.5 overall and ±3.2, 4.3, 4.0 and 5.8 units at 1, 2, 3 and 4 years, respectively. A model excluding baseline age explained 52% of variation in ΔPUL. In both models, baseline PUL and ambulatory status were significant predictors of long-term ΔPUL trajectories. The use of widely available baseline characteristics to predict multi-year ΔPUL supports the potential for broad application of these models. Further validation is warranted to assess generalizability and additional prognostic factors.