Non-invasive vital-sign monitoring and data fusion in acute care
<p>Post-operative patients can deteriorate physiologically, leading to adverse events such as cardiac arrest, unexpected intensive care unit (ICU) admission or death. Clinical staff monitor patients by observation of the vital signs. Early warning scores (EWS) are computed from vital-sign valu...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
2018
|
| Subjects: |
| _version_ | 1826273370218430464 |
|---|---|
| author | Tomlinson, H |
| author2 | Tarassenko, L |
| author_facet | Tarassenko, L Tomlinson, H |
| author_sort | Tomlinson, H |
| collection | OXFORD |
| description | <p>Post-operative patients can deteriorate physiologically, leading to adverse events such as cardiac arrest, unexpected intensive care unit (ICU) admission or death. Clinical staff monitor patients by observation of the vital signs. Early warning scores (EWS) are computed from vital-sign values to indicate when a patient requires an escalation in care. The investigations in this thesis assess the use of machine learning and continuous vital-sign monitoring to improve detection of physiological deterioration.</p> <p>Two post-operative datasets are introduced; vital-sign data collected from 407 patients using manual and continuous methods, and vital-sign data collected from 447 patients using manual methods. A machine learning model is trained on the first manual dataset using vital sign and diurnal vital-sign variability values, and is tested on the second dataset. The model achieves an Area Under the Receiver-Operator Characteristic curve (AUROC) of 0.886 for predicting adverse events within 24 hours, with the National Early Warning Score achieving an AUROC of 0.848. Respiratory rate variability, blood oxygen saturation variability and systolic blood pressure variability are more predictive of adverse events than the corresponding vital signs.</p> <p>A method comparison study is conducted for manual and continuous vital-sign measurements. There is no evidence of a bias towards normal readings in manual observations, or an arousal effect due to the vital-sign taker. However, differences between manual and continuous data are shown to be frequently large. EWS models should be designed specifically for continuous vital-sign data. Methods for estimating diurnal vital-sign variability from continuous data are described, and are shown to discriminate between physiological stability and deterioration.</p> <p>A clinical study of video camera-based non-contact monitoring of 15 post-operative patients is described. Heart rate and respiratory rate estimation methods are developed with mean average errors of 0.8 to 2.8 heart-beats/minute and 0.6 to 2.0 breaths/minute during 5 periods of physiological deterioration. Spatial maps of respiratory signal amplitude are shown for a patient with abnormal respiratory physiology.</p> <p>In summary, EWSs should incorporate features of vital-sign variability, with future work on continuous monitoring implementations. Camera-based methods may provide continuous vital-sign monitoring and spatial physiological maps in post-operative care.</p> |
| first_indexed | 2024-03-06T22:27:09Z |
| format | Thesis |
| id | oxford-uuid:570ab6ec-684b-48ef-9a29-5e41b80f2a1e |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-06T22:27:09Z |
| publishDate | 2018 |
| record_format | dspace |
| spelling | oxford-uuid:570ab6ec-684b-48ef-9a29-5e41b80f2a1e2022-03-26T16:54:11ZNon-invasive vital-sign monitoring and data fusion in acute careThesishttp://purl.org/coar/resource_type/c_db06uuid:570ab6ec-684b-48ef-9a29-5e41b80f2a1ePost-operative careBiomedical engineeringPhysiological monitoringEnglishORA Deposit2018Tomlinson, HTarassenko, LYoung, DPimentel, MMcleod, C<p>Post-operative patients can deteriorate physiologically, leading to adverse events such as cardiac arrest, unexpected intensive care unit (ICU) admission or death. Clinical staff monitor patients by observation of the vital signs. Early warning scores (EWS) are computed from vital-sign values to indicate when a patient requires an escalation in care. The investigations in this thesis assess the use of machine learning and continuous vital-sign monitoring to improve detection of physiological deterioration.</p> <p>Two post-operative datasets are introduced; vital-sign data collected from 407 patients using manual and continuous methods, and vital-sign data collected from 447 patients using manual methods. A machine learning model is trained on the first manual dataset using vital sign and diurnal vital-sign variability values, and is tested on the second dataset. The model achieves an Area Under the Receiver-Operator Characteristic curve (AUROC) of 0.886 for predicting adverse events within 24 hours, with the National Early Warning Score achieving an AUROC of 0.848. Respiratory rate variability, blood oxygen saturation variability and systolic blood pressure variability are more predictive of adverse events than the corresponding vital signs.</p> <p>A method comparison study is conducted for manual and continuous vital-sign measurements. There is no evidence of a bias towards normal readings in manual observations, or an arousal effect due to the vital-sign taker. However, differences between manual and continuous data are shown to be frequently large. EWS models should be designed specifically for continuous vital-sign data. Methods for estimating diurnal vital-sign variability from continuous data are described, and are shown to discriminate between physiological stability and deterioration.</p> <p>A clinical study of video camera-based non-contact monitoring of 15 post-operative patients is described. Heart rate and respiratory rate estimation methods are developed with mean average errors of 0.8 to 2.8 heart-beats/minute and 0.6 to 2.0 breaths/minute during 5 periods of physiological deterioration. Spatial maps of respiratory signal amplitude are shown for a patient with abnormal respiratory physiology.</p> <p>In summary, EWSs should incorporate features of vital-sign variability, with future work on continuous monitoring implementations. Camera-based methods may provide continuous vital-sign monitoring and spatial physiological maps in post-operative care.</p> |
| spellingShingle | Post-operative care Biomedical engineering Physiological monitoring Tomlinson, H Non-invasive vital-sign monitoring and data fusion in acute care |
| title | Non-invasive vital-sign monitoring and data fusion in acute care |
| title_full | Non-invasive vital-sign monitoring and data fusion in acute care |
| title_fullStr | Non-invasive vital-sign monitoring and data fusion in acute care |
| title_full_unstemmed | Non-invasive vital-sign monitoring and data fusion in acute care |
| title_short | Non-invasive vital-sign monitoring and data fusion in acute care |
| title_sort | non invasive vital sign monitoring and data fusion in acute care |
| topic | Post-operative care Biomedical engineering Physiological monitoring |
| work_keys_str_mv | AT tomlinsonh noninvasivevitalsignmonitoringanddatafusioninacutecare |