142P A preliminary machine learning retrospective observational study to predict treatment response to nusinersen in non-sitter spinal muscular atrophy

Disease-modifying treatments, such as nusinersen, have significantly improved the disease trajectory of Spinal Muscular Atrophy (SMA) patients. Treatment response is heterogeneous, and predicting response is crucial for clinicians to give realistic guidance to patients and carers when counselling about treatment options. We collected data on 124 patients from the SMA Reach UK and Italian Network studies. Participants were included if they were treated under 4 years and were non-sitters at nusinersen initiation, were randomly allocated to training (80%) and testing cohorts (20%). Tree and regression-based machine learning for survival outcomes were compared, and oblique random forests were selected, with a testing C-Index of 0.74. The features were selected from items of the CHOP and HINE-2, respiratory symptoms and bulbar interventions using mutual information. In total, 62 (50%) have achieved sitting, at a median age of 2.4 years. The predicted median time-to-sitting in those requiring NG or G-Tubes at treatment was 4-months later on average than those without, and this was the most influential factor. The ability to kick at treatment, (≥1 on the HINE-2 item) was also influential in predicting the likelihood of sitting, as was antigravity shoulder movement in supine and ability to reach in side-lying (4 on items 1 and 8), and neck muscle activation in pull to sit (≥2 on item 14) and rolling fully prone from supine when elicited from arms (≥3 on item 7) on the CHOP at treatment. Most notably, those who could roll fully prone from supine had a median predicted time-to-sit 5.8 months earlier than an identical patient who could not. Nusinersen starting and symptom onset age were less influential, whilst SMN2 copy numbers did not improve the model. This work provides a tool to predict treatment response in nusinersen-treated patients and has the potential to be extended to other treatment options, moving clinicians towards personalised treatment plans for each patient.