St Cross College
CDT Postgraduate Studentship in Healthcare Innovation (RCUK Digital Economy Programme grant number EP/G036861/1)
Vital-sign data fusion models for post-operative patients
The current standard of clinical practice for patient monitoring on surgical wards in the UK is regular manual observations of the main vital signs by nursing staff and the usage of early-warning-score systems, in which scores are assigned to the manual observations, and care escalated if the scores exceed some predefined threshold. We propose the introduction of a physiological 24-hour variability index and the use of machine learning techniques to improve such systems.
A dataset comprising observational vital-sign data from 200 post-operative patients taking part in a clinical trial at the Oxford Cancer Centre, was used to explore the trajectory of patients’ vital-sign changes during their stay on the post-operative ward. Models of normality based on data with and without the variability index from patients who had a “normal recovery” are constructed and compared using probabilistic and discriminative methods, and then tested with “abnormal” data from patients who deteriorate sufficiently after surgery to be admitted to the Intensive Care Unit (ICU). Results show that the 24-hour variability of respiratory rate and systolic blood pressure after surgery improve the performance of the different machine learning approaches in identifying “abnormal” patients, and may be good predictors of ICU admission for surgical patients.