Objectives: Amyotrophic lateral sclerosis (ALS) is a rare disease with heterogenous progression rates ranging from slow to rapid progression. Studies aiming to characterize the factors associated with the progression rate have focused on survival but few concerned the functional decline trajectory [1]. Methods: Using PROACT database (8,569 patients), we select spinal and bulbar patients with at least a baseline and a second follow-up visit. We randomly select spinal patients to get two balanced groups of 1,380 patients.The following steps were performed: 1) We built a multimodal ALS course map that grasped long-term disease progression in a mixed-effects fashion [2] with Leaspy. We used 6 features: the four subscores of ALSFRS-R, forced vital capacity (FVC) and BMI. 2) We extracted the progression rate and onset age, and the relative progression of each feature of each patient from the parameters of the model. 3) We also computed from the disease course, the conversion age to a progression threshold for each feature: 8 points for ALSFRS-R subscores in reference to FT9 score, 18.5 for the BMI and 2.43 litres for FVC.Finally, we compared the distributions of the extracted parameters and conversion ages using independent t-test. Results: We found that ALS starts 2.84 years later for bulbar patients, but progresses 1.36 times faster. We observed, for bulbar patients, that ALSFRS-R fine and gross motor progress 3.7 and 4.4 months later than the spinal patients but ALSFRS-R bulbar progression starts almost 8 months before. Bulbar patients reached endpoint thresholds define above, in average after spinal patients for: ALSFRS-R fine motor (8 points, 49.7 months), gross motor (44 months), bulbar (7.5 months) and respiratory (23.3 months), FVC (24 months). Conclusions: We build a modelling framework for describing ALS subtypes effect on functional endpoints from multimodal screening assessments. This model also allows describing and predicting individual progression which can pave the way to discriminate fast and slow progressor for stratification of clinical trial. This methodology could also be applied in clinical trial to compare treated and placebo arms. References [1] Grollemund, Vincent, et al. Frontiers in neuroscience 13 (2019): 135.[2] Schiratti, Jean- Baptiste, et al. The Journal of Machine Learning Research 18.1 (2017): 4840-4872.