A carregar...
Publicação
Linking patient data to scientific knowledge to support contextualized mining
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 6.3 MB | Adobe PDF |
Orientador(es)
Resumo(s)
ICU readmissions are a critical problem associated with either serious conditions, ill nesses, or complications, representing a 4 times increase in mortality risk and a financial
burden to health institutions. In developed countries 1 in every 10 patients discharged
comes back to the ICU. As hospitals become more and more data-oriented with the adop tion of Electronic Health Records (EHR), there as been a rise in the development of com putational approaches to support clinical decision.
In recent years new efforts emerged, using machine learning approaches to make ICU
readmission predictions directly over EHR data. Despite these growing efforts, machine
learning approaches still explore EHR data directly without taking into account its mean ing or context. Medical knowledge is not accessible to these methods, who work blindly
over the data, without considering the meaning and relationships the data objects. Ontolo gies and knowledge graphs can help bridge this gap between data and scientific context,
since they are computational artefacts that represent the entities in a domain and how the
relate to each other in a formalized fashion.
This opportunity motivated the aim of this work: to investigate how enriching EHR
data with ontology-based semantic annotations and applying machine learning techniques
that explore them can impact the prediction of 30-day ICU readmission risk. To achieve
this, a number of contributions were developed, including: (1) An enrichment of the
MIMIC-III data set with annotations to several biomedical ontologies; (2) A novel ap proach to predict ICU readmission risk that explores knowledge graph embeddings to
represent patient data taking into account the semantic annotations; (3) A variant of the
predictive approach that targets different moments to support risk prediction throughout
the ICU stay.
The predictive approaches outperformed both state-of-the-art and a baseline achieving
a ROC-AUC of 0.815 (an increase of 0.2 over the state of the art). The positive results
achieved motivated the development of an entrepreneurial project, which placed in the
Top 5 of the H-INNOVA 2021 entrepreneurship award.
Descrição
Tese de mestrado, Bioinformática e Biologia Computacional, Universidade de Lisboa, Faculdade de Ciências, 2022
Palavras-chave
Anotação semantica Ontologias Biomédicas Readmissões em UCI Apredizagme automática Embedding de grafos de conhecimento Teses de mestrado - 2022