Modeling disease progression in spinocerebellar ataxias

Background and objectivesThe most common autosomal-dominantly inherited spinocerebellar ataxias (SCA), SCA1, SCA2, SCA3 and SCA6, account for more than half of all SCA families. Disease course is characterized by progressive ataxia and additional neurological signs. Each of these SCAs is caused by a CAG repeat expansion, leading to an expanded polyglutamine stretch in the resulting type-specific protein. To comparatively investigate determinants of disease progression, we analyzed demographic and genetic data and three-year clinical time courses of neurological symptoms. The aim was to provide tailored marker candidates and prediction models to support type-specific clinical monitoring and trial design. MethodsTo analyze relationships among the different neurological symptoms, we examined co-occurrence patterns of deterioration events. Predicting disease progression was treated as a survival analysis problem. ResultsThe data set contained 1538 subjects from five different longitudinal cohorts and 3802 visits. The pattern of neurological symptoms that showed progression varied with the SCA type. Mining of the progression data revealed the Scale for the Assessment and Rating of Ataxia (SARA) sum score to be the most representative descriptor of disease progression, reflecting progression of the majority of the other included symptoms. We trained models for predicting the progression of each neurological symptom for each SCA type from genetic features, age and symptoms at the baseline visit. The most universal predictors included the SARA sum score, gait and the CAG repeat length of the expanded allele. Finally, deterioration in disease staging was studied in detail: For the milestones of deterioration, (i) the need to use walking aids and (ii) the requirement to use a wheelchair, we discovered common as well as diverging predictive markers. For clinical interpretability, a decision tree was built to indicate the probability of progression within 3 years in dependence of the top predictive features. DiscussionData-driven approaches are potent tools to identify the main contributing features of progression prediction. Progression events for the disease stage were predictable from the baseline neurological status. Remarkably, a limited number of features had predictive importance, and only few were shared among all four SCA types, including gait and the SARA sum score, confirming the need for type-specific models.