Heart Rate and Patient Mortality: Machine Learning's Role in Healthcare
Presenter Type
UNO Undergraduate Student
Major/Field of Study
Bioinformatics
Advisor Information
Associate Professor School of Interdisciplinary Informatics
Location
MBSC Ballroom Poster # 1305 - U
Presentation Type
Poster
Start Date
24-3-2023 1:00 PM
End Date
24-3-2023 2:15 PM
Abstract
Introduction: In a clinical setting, methods for predicting patient mortality are often used to enhance the decision-making of health-care professionals and better manage patient care. Further refinement of these methods is an important step in improving patient outcomes and overall quality of care. Heart rate variability (HRV) is a measure of the heart’s complex beating patterns, giving medical professionals additional insight into patient health. Previous research has demonstrated the potential use of heart rate variability as a metric for patient mortality prediction for various conditions. This study aims to establish this link in a broader and more diverse set of patients.
Methods: Two measures of HRV, SDNN and PNN50, were examined along with heart rate to investigate the role of HRV in patient mortality. The MIMIC-III clinical database, hosted by PhysioNet.org, was combined with the MIMIC-III waveform database to provide a set of patient records with accompanying ECG data. ECG data was converted into a list of inter-beat intervals, and this list was used to generate HRV statistics for each patient. The statistical difference between the patient population that passed away within 90 days of their ECG recording and those that survived was calculated with a Mann-Whitney U rank test. Additionally, a subset of patients with several recordings within a short timeframe of their death were chosen to examine how HRV trends change preceding patient death.
Results: Differences in heart rate and PNN50 were statistically significant between patients that expired within 90 days of recording and those that survived. When examining these differences in relation to patient gender, only male patients continued to show these differences with statistical significance. Logistic regression analysis showed little to no success in differentiating between patients based solely on HRV data.
Discussion/Conclusion: Heart rate and PNN50 are statistically linked to 90-day patient mortality. While unable to predict patient mortality on their own, including these measures into a pre-existing patient mortality model may improve overall accuracy. Computational restraints limited the amount of analysis that was done for each patient; future work should handle all possible ECG data to provide a more complete picture. Future work in this area should also examine recording context, as HRV while unconscious can differ significantly from HRV while awake and at rest. Overall, the use of machine learning techniques and publicly available data is important for the continued development of healthcare models.
Scheduling
10:45 a.m.-Noon, 1-2:15 p.m., 2:30 -3:45 p.m.
Heart Rate and Patient Mortality: Machine Learning's Role in Healthcare
MBSC Ballroom Poster # 1305 - U
Introduction: In a clinical setting, methods for predicting patient mortality are often used to enhance the decision-making of health-care professionals and better manage patient care. Further refinement of these methods is an important step in improving patient outcomes and overall quality of care. Heart rate variability (HRV) is a measure of the heart’s complex beating patterns, giving medical professionals additional insight into patient health. Previous research has demonstrated the potential use of heart rate variability as a metric for patient mortality prediction for various conditions. This study aims to establish this link in a broader and more diverse set of patients.
Methods: Two measures of HRV, SDNN and PNN50, were examined along with heart rate to investigate the role of HRV in patient mortality. The MIMIC-III clinical database, hosted by PhysioNet.org, was combined with the MIMIC-III waveform database to provide a set of patient records with accompanying ECG data. ECG data was converted into a list of inter-beat intervals, and this list was used to generate HRV statistics for each patient. The statistical difference between the patient population that passed away within 90 days of their ECG recording and those that survived was calculated with a Mann-Whitney U rank test. Additionally, a subset of patients with several recordings within a short timeframe of their death were chosen to examine how HRV trends change preceding patient death.
Results: Differences in heart rate and PNN50 were statistically significant between patients that expired within 90 days of recording and those that survived. When examining these differences in relation to patient gender, only male patients continued to show these differences with statistical significance. Logistic regression analysis showed little to no success in differentiating between patients based solely on HRV data.
Discussion/Conclusion: Heart rate and PNN50 are statistically linked to 90-day patient mortality. While unable to predict patient mortality on their own, including these measures into a pre-existing patient mortality model may improve overall accuracy. Computational restraints limited the amount of analysis that was done for each patient; future work should handle all possible ECG data to provide a more complete picture. Future work in this area should also examine recording context, as HRV while unconscious can differ significantly from HRV while awake and at rest. Overall, the use of machine learning techniques and publicly available data is important for the continued development of healthcare models.