Predicting Time to Death after the withdrawal of Life Supporting Therapies Using Variability Analysis of Vital Signs Waveform Data
CCCF ePoster library. Scales N. Nov 7, 2018; 234194
Dr. Nathan Scales
Dr. Nathan Scales
Login now to access Regular content available to all registered users.

You may also access this content "anytime, anywhere" with the Free MULTILEARNING App for iOS and Android
Abstract
Rate & Comment (0)

Background: Organ donation after circulatory determined death (DCD) accounts for 17% of all deceased organ donors in Canada, a fraction of its theoretical potential [1]. Up to half of all potential DCD patients are unable to donate because they do not die within the required time limits to minimize organ ischemic damage [2]. The inability to predict time to death results in additional stress for families, a reduced number of successful donors, and increased costs for the health system. The DePPaRT (Death Physiology and Prediction After Removal of Therapy) study captured vital signs waveform data from 654 patients undergoing withdrawal of life sustaining therapy (WLST) in 22 centres in Canada, the Czech Republic, and the Netherlands. As variability is associated with illness [4,6,7], we hypothesized that the loss of vital signs variability, specifically heart rate and blood pressure variability (HRV & BPV), in the hour prior to WLST might be useful in predicting time to death after WLST.



Objective:

To develop a model based on vital signs waveform variability capable of predicting whether a patient will die within the timelines required for organ donation (0.5, 1, and 2 hours after WLST).



Methods: Vital signs waveform data was available prior to WLST in 567 patients. This data was processed to obtain beat-to-beat event times series from the ECG and arterial blood pressure waveforms (R peak to R peak interval (RRI), systolic, diastolic, and pulse blood pressures), using CIMVA [3,4]. Each series was filtered to remove arrhythmias prior to the calculation of up to 53 measures of variability, using windows of 750 beats. We excluded patients that had prolonged WLST, namely patients with additional life supporting therapies removed after 10 minutes after the first act of WLST.  Patients enrolled prior to and after March 2017 were used in the derivation (210 patients) and validation (105 patients) cohorts, respectively (Figure 1).  A comprehensive suite of variability measures was calculated for each patient. All features were normalized to have a mean of 0 and a standard deviation of 1. We used random survival forests in R to develop a predictive model. The model yielded a probability of dying within each 15-minute interval up until 24 hours after WLST for each patient (Figure 2a). Probabilities at specific times on these curves (t = 0.5, 1, and 2 hours) were used as scores to predict if a given patient would die within these times. Features were ranked based on their importance over a 10-fold cross validation of the training set. The model was retrained using only the highest ranked variables, adding one variable at a time. We used all 210 patients from the training set and the 17 features required for the AUC to plateau (Figure 2b) in a final model tested on the validation set.



Results: The derived model employing 17 variability features achieved a median AUC of 0.74, 0.78, and 0.76 for death within 0.5, 1, or 2 hours, with sensitivities and specificities of (0.65, 0.81), (0.73,0.75), and (0.76,0.72) respectively, on the validation set (Figure 3a). Nearly 86% of the patients with the highest scores died within 2 hours of WLST (Figure 3b).



Conclusion: For the first time, a predictive model using only heart rate and blood pressure variability data has been shown to achieve similar performance to models derived using clinical data [5]. In the future, we will combine and evaluate models incorporating both variability and clinical data.

 


    This eLearning portal is powered by:
    This eLearning portal is powered by MULTIEPORTAL
Anonymous User Privacy Preferences

Strictly Necessary Cookies (Always Active)

MULTILEARNING platforms and tools hereinafter referred as “MLG SOFTWARE” are provided to you as pure educational platforms/services requiring cookies to operate. In the case of the MLG SOFTWARE, cookies are essential for the Platform to function properly for the provision of education. If these cookies are disabled, a large subset of the functionality provided by the Platform will either be unavailable or cease to work as expected. The MLG SOFTWARE do not capture non-essential activities such as menu items and listings you click on or pages viewed.


Performance Cookies

Performance cookies are used to analyse how visitors use a website in order to provide a better user experience.


Save Settings