RH09
Improving Detection Disease Course in Multiple Sclerosis: An Alternative Pros' Based Strategy
Measuring Multiple Sclerosis (MS) is an hard challenge. The main difficulty is the variety of functional problems that may affect the individual patients, so that no “typical” progression can be established, and no satisfactory “composite” indicators have been identified. Following recent guidelines, a methodological reflection is needed in order to develop a reliable core outcomes set.
Objectives:
The ongoing Italian MS Foundation (FISM) initiative on patients reported outcomes (PRO) has the main objective of validating a “functional profile” of MS based on meaningful variables and measures, useful to improve the disease course detection, quantify disease progression and identify the best disease predictors. Here, a proof of concept analysis is shown with the aim of detecting disease course.
Methods:
The clinical variables collected in the present study were based on functions sufficient to encompass the patient’s disability and to represent whole-person behaviours. The set of PRO and clinical scales selected were related mainly to mobility, fatigue, cognitive performances, emotional status, bladder continence, quality of life. About 500 people with MS were enrolled in the study without any inclusion/exclusion criteria unless MS diagnosis. The collected data were analysed with machine learning methods, taking into account both unsupervised and supervised methods. Assuming that every measured variable does not have the same relevance, the main goal pursued was to find the real dimension of data, using techniques such as principal component analysis and k-means clustering. On the other hand, to tackle the problem of disease course detection, a supervised problem was solved , in order to identify optimal models for disease progress discrimination based on linear classifiers such as Support Vector Machines.
Results:
The applied unsupervised techniques showed that the group of patients diagnosed as relapsing-remitting could be isolated from other clinical courses. Hence, supervised learning algorithms were applied to discriminate this group from the others. The best model reached median Matthews Correlation Coefficient, precision and recall of 0.610, 0.874, 0.820, respectively.
Conclusions:
The preliminary results suggest that the disease course could be inferred from PRO with a reasonable level of confidence and further improvements may be achieved in this context.