Titulo Estágio
Risk prediction algorithm for timely identification of post-operative deterioration
Áreas de especialidade
Sistemas Inteligentes
Local do Estágio
DEI + Eindhoven University of Technology
Enquadramento
Despite improvements in anesthesia and post-operative care, about 25% of patients undergoing surgery suffer from serious post-operative complications [1]. Prediction and early identification of patients at risk of deterioration is thus crucial to guide clinical decisions before, during, and after surgery. In fact, after surgery the majority of patients is transferred to the ward, where 40% of unanticipated deaths occur, while approximately 1% of the patients is transferred back to the intensive care unit (ICU) for management of adverse events. Methods for risk stratification could thus be useful for postoperative care, potentially leading to direct admission to the ICU of those patients at risk of post-operative deterioration. Moreover, timely detection and management of deterioration in the ward is crucial to improve patient outcomes such as post-surgery deaths, unplanned ICU admissions and cardiorespiratory arrest. In this context, a retrospective cohort study of post-operative ward patients is proposed as part of the joint collaboration between the Eindhoven University of Technology, the Catharina Hospital, and Philips (Impulse II flagship).
Objetivo
Risk prediction tools and early warning systems have been implemented for timely identification of patients at risk of postoperative deterioration [2, 3]. However, current warning scores have several limitations: (i) they are designed for specific patient populations, and often unable to cope with co-morbidities; (ii) they are based on population averages, ignoring inter-patient variability; (ii) they are unable to incorporate new factors and domain knowledge; (iv) they have limited clinical interpretability; (v) they typically use only dichotomous and/or sparse, non-real-time data sampled periodically from the patient. The aim of this project is to investigate novel strategies for prediction of post-operative deterioration risk. Starting by the combination of multiple risk scores, new predictive factors will be investigated by e.g. machine learning approaches and incorporated in the risk prediction algorithm.
Plano de Trabalhos - Semestre 1
- Literature study / state-of-the-art: risk prediction models, multi-feature classification, machine learning, signal processing.
- Preliminary data analysis and extraction of predictive factors.
Plano de Trabalhos - Semestre 2
- Design, implementation and testing of an algorithm for risk prediction.
- Validation against state-of-the-art methods.
- Report and documentation.
Condições
The work requires a basic understanding of physiological processes, interest and knowledge in data-driven approaches for classification and feature selection, as well as in basic statistics and data analytics. A solid background in signal processing and machine learning is needed, with good control of at least one programming environments among MATLAB, Python, and R.
The Master-thesis includes a period to be spent at least 6 months at the Eindhoven University of Technology, with possible visits at the Catharina Hospital (Eindhoven).
Coordinator:
dr. Simona Turco (s.turco@tue.nl)
Prof Paulo de Carvalho (DEI)
Observações
1] Petit, Clemence, Rick Bezemer, and Louis Atallah. "A review of recent advances in data analytics for post-operative patient deterioration detection." Journal of clinical monitoring and computing (2017): 1-12.
[2] Mendes, D., et al. "Integration of Current Clinical Knowledge with a Data Driven Approach: An Innovative Perspective." International Journal of Information Technology & Decision Making 17.01 (2018): 133-153.
[3] Paredes, Simão, et al. "New approaches for improving cardiovascular risk assessment." Revista Portuguesa de Cardiologia 35.1 (2016): 5-13.
Orientador
Jorge Manuel Oliveira Henriques
jh@dei.uc.pt 📩