A carregar...
Publicação
Relapse Prediction in Multiple Sclerosis: a Supervised Learning Approach
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 961.32 KB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Multiple sclerosis (MS) is an autoimmune disease of the central nervous system affecting, approximately,
2.8 million people around the world. Its onset may be abrupt or insidious and its evolution, clinical
manifestation, and treatment response may vary widely across patients, making it difficult to predict
and manage the disease outcomes. The majority of MS patients (around 85%-90%) are diagnosed with
relapse-remitting MS (RRMS), a form of the disease that is characterised by alternating periods of relapses
and of total or partial recovery (remission). These relapses last at least 24h, develop acutely or subacutely,
are mostly monosymptomatic, and are documented to affect patients’ disability levels, at least in the early
stages of the disease, and their social, working, and household activities. Despite all the existing research
focused on understanding the characteristics and impact of relapses, these remain largely unpredictable
both in time and location. Moreover, the fact that this disease manifestation affects most MS patients and
that the anticipation of a future relapse may help clinicians to provide more timely and effective treatments
to MS patients, emphasizes the need to develop models that are able to predict when a next relapse may
occur. Using a dataset containing 3,679 relapses from 859 patients, this study proposes developing a
prediction model for when a future relapse may occur based on the clinical profile of RRMS patients at
the time of relapse. To this end, three classifiers were learnt (Logistic Regression, Decision Tree, and
Random Forest), considering as target variables if a relapse happened within one or two years, and as
a clinical profile the combination of demographic, symptoms, disability, treatment, and MRI data. The
best-performing model was the Random Forest, with an AUC of 68%, when evaluating if a relapse may
happen within two years from the previous relapse. The most relevant variables for predicting the timing
of a future relapse include the time elapsed between the two most recent relapses, the number of relapses
already endured by a patient, the number of different medicines the patient has tried since the onset of the
disease, disease duration, the number of symptoms experienced in a relapse, and the age at onset. Overall,
the findings reported in this study can help clinicians to assess the future prospects of the disease, for a
given patient, and to tailor the healthcare provided accordingly.
Descrição
Tese de mestrado, Ciência de Dados, 2022, Universidade de Lisboa, Faculdade de Ciências
Palavras-chave
Esclerose Múltipla Esclerose Múltipla Recidivante Remitente Surtos Aprendizagem Automática Análise preditiva Teses de mestrado - 2023