Bedside Big Data, multiorgan variability and complex systems science

Physiological waveform data collection and bedside Big Data

Utilizing the currently untapped information contained in waveform data has the potential to reduce the diagnostic and prognostic uncertainty inherent in critical care. This uncertainty results in delayed diagnosis, unnecessary or inappropriate therapy and increased complications, mortality and cost of care. For over a decade, we have been facilitating and championing physiological waveform collection at hospital sites, within the context of multicenter clinical studies, in order to answer clearly defined, practical clinical problems. One of DAL research focus is the collection, cleaning, processing and extraction of key information from physiological waveform data.


Variability and complexity research

The patterns of variation of fiducial points extracted from physiological waveforms, such as inter-beat and inter-breath interval time series (i.e. HRV and RRV) have been shown to relate directly to underlying health. Health is characterized by a large degree of complex variations and decreased variability is associated with age and illness and correlates with illness severity, indicating reduced adaptability and/or increased stress. In order to uncover and characterize these complex patterns of variation, we have developed a tool called Continuous Individualized Multiorgan Variability Analysis (CIMVA). CIMVA is a software platform for performing a series of automated algorithms that assess the quality of physiological waveform files and provide a comprehensive multivariate characterization of the degree of variability and complexity of physiological signals, tracked over time. Each variability metric calculated provides a unique perspective on the data as no single technique offers a definitive characterization of biologic signals. CIMVA facilitates the reproducibility, internal and external validity of results across different studies and different groups of researchers. Custom drivers have been developed to parse a inputs from various bedside and ambulatory monitors. With the help of CIMVA, we have explored a range of clinical applications of variability monitoring.


Fractals and entropy production

Entropy production is universally present in nature. Thought to be unrelated, fractal structures demonstrating multiscale self-similarity are similarly ubiquitous. Combining data sets from the human organism, the study of the two and their hypothetical association may be evaluated. This research project seeks to fundamentally understand if nature's desire to optimize entropy production is the driving force for fractal structure, by rigorously evaluating multiscale self-similar spatial and temporal structures and their relationship with entropy production, focusing on the utility of this novel understanding. Understanding the association between fractals and entropy production has broad scientific implications for monitoring and impacting the system level properties of far-from thermodynamic equilibrium dynamical systems. This project is run in collaboration with Prof. André Longtin (Department of Physics, University of Ottawa, Ottawa, Canada).


Artificial Intelligence at the bedside: predicting patient outcomes

Predicting Extubation Outcomes

Expeditious, safe extubation is vitally important in the care of ICU patients as prolonged mechanical ventilation and failed extubation (i.e. re-intubation within 48 hrs) are associated with prolonged ICU and hospital length of stay as well as increased morbidity, mortality & costs. Ability to tolerate reduced ventilatory support during a spontaneous breathing trial (SBT) is the current recommended practice to assess extubation readiness. Despite this practice, several studies document an extubation failure rate of 15% and traditional measures (e.g. rapid shallow breathing index) and weaning indices are in general poor predictors of extubation failure. In a large prospective, observational, multicenter CIHR-funded study of 721 patients (Weaning and Variability Evaluation (WAVE) study), we found that while heart rate variability (HRV) and respiratory rate (RRV) during SBTs were significantly altered in association with extubation failure, a predictive model (called WAVE score) based on a few RRV measures demonstrated better predictive accuracy compared to current measures. Internally validated performance of the WAVE score suggest AUCs of 0.72 overall, with increased performance on patients with high RSBI or in patients clinically deemed high-risk. Extubation Advisor™ (EA) is a bedside decision support tool designed to help both standardize the process of SBT performance and reporting, and provide the clinician with optimal prediction of extubation outcomes. EA will provide a means for RTs to record their objective and subjective assessment of an individual patient's readiness for extubation, along with an estimate of the risk of extubation failure based on conventional measures (f/VT) and based on respiratory rate variability analysis, summarizing the information in a report delivered within minutes at the bedside, to help clinicians assess a patient's readiness for extubation. We believe this combination of standardized SBT performance and reporting and optimal prediction of extubation outcomes will improve outcomes and reduce costs, by both decreasing the incidence of extubation failure and expediting extubation of low risk patients, shortening length of stay We are currently analyzing data from a pilot phase I mixed methods observational study, where we introduced and evaluated Extubation Advisor within two academic critical care units and will publish results in early 2019.


Predicting deterioration in patients and risk stratification

Sepsis is a leading cause of morbidity and mortality with a large number of hospital admissions, deaths, and $24 billion in hospital costs per year in North America. Severe sepsis and septic shock are the most common causes of mortality in critically ill patients, accounting for 10% to 15% of intensive care unit (ICU) admissions and 2.9% of all hospital admissions. Early treatment improves outcomes and unexpected, late deterioration requiring ICU admission in patients with infection initially admitted to the ward are associated with higher costs, hospital length of stay, and greater mortality.

Sepsis risk stratification in the Emergency Department (ED)

Disposition decision-making for patients with infection in the ED is both complex and critical to patient outcomes. A standardized, repeatable and individualized tool to enable ED physicians to predict future deterioration does not yet exist. In a recently published study in collaboration with Dr. Doug Barnaby (Feinstein Institute for Medical Research, Northwell Health, Manhasset, NY), we have evaluated vital signs, laboratory values and HRV prospectively upon ED admission in 1247 patients (832 patients with high quality data) presenting with infection and SIRS. Using machine learning techniques, we have developed model based on a combination of HRV metrics and laboratory values to predict subsequent deterioration (vasopressors, intubation, non-invasive ventilation, ICU admission, death) within 72 hrs of ED presentation, and identified a high-risk cohort of patients. We are currently pursuing validation and are implementing our predictive model as part of a standardized assessment and clinical tool to assist disposition decision-making, deciding if patients should be discharged, admitted to wards or step-down units, and thereby better prevent costly and harmful unexpected deterioration on wards and improve care.


Early warning of sepsis in at risk populations

Bacterial infection leading to septic shock remains a major cause of suffering and death, particularly in patients with impaired host defense. Timely antibiotics and restoration of perfusion are proven to save lives and costs and therefore early identification of both presence and severity of infection is critical to prevent clinical deterioration, organ failure and death. We have demonstrated that HRV monitoring provides an early detection of infection in neutropenic patients, with a loss of heart rate variability across different metrics up to 48h prior to antibiotics escalation. We are now looking at combining the value of sensitive continuous variability monitoring with specific daily serum biomarker assessment, in order to provide early detection and prognostication of infection in high-risk patients with a daily risk stratification and early warning of infection. Over 10 years, we have collected up to 2 weeks of continuous ECG data from 81 ambulatory patients with hematologic malignancy undergoing myeloablation and bone marrow transplantation (BMT), who have a markedly elevated risk of infection. Using machine learning techniques and a small set of features extracted from a composite measure of variability over time, we trained a predictive model to estimate probabilities of deterioration over time for each patient. That model was further combined with key blood serum biomarkers data available at 24h intervals to refine and further improve the prediction. These results are expected to be published in early 2019.


Severity of illness in the ICU

In conjunction with RRV, HRV monitoring helps track the severity of illness and organ failure in critically ill patients and can improve the benefit, timing, and duration of sedation interruption in critically ill patients.


Predicting organ donor suitability and physiology at the end of life

Organ donation provides lifesaving and cost effective treatment for thousands of Canadians, yet hundreds of patients die every year waiting for transplant. Previously, deceased organ donation was a rare event that only occurred after a person was declared dead by neurologic criteria. Organ donation after circulatory determination of death (DCD), which allows donation in the event of a severe unrecoverable illness (but not requiring fulfillment of criteria for neurologic death), has expanded donation. DCD now accounts for a quarter of deceased organ donation in Canada. There is marked potential for DCD to grow as challenges are addressed. Over a third of consented DCD candidates do not proceed to donation after the withdrawal of life sustaining measures (WLSM); patients do not die within the time limits set to minimize ischemic organ damage. These unsuccessful DCD attempts increase family distress and consume scarce healthcare resources because the retrieval team is assembled prior to WLSM. DCD is often not offered to families or if offered, many do not consent to DCD because of the uncertainty regarding likelihood of success. There is a critical need to improve the prediction of time to circulatory death and DCD suitability, and to objectively characterize ischemia time during the dying process. As part of the Death Prediction and Physiology after Removal of Therapy (DePPaRT) Study led by Dr. Sonny Dhanani (University of Ottawa), we have been investigating these questions. Applying machine learning techniques to biomedical waveform data, we have derived and validated a predictive model, based on reduced heart rate and blood pressure variability, demonstrating 80% accuracy in identifying which DCD candidates are likely to become successful donors. Assessment of biomedical waveforms also enables objective characterization of organ ischemia during the DCD process. We hypothesize that this pioneering approach will provide DCD clinical decision support that will enhance the efficiency and success of DCD donation attempts. Ongoing efforts and future plans include pursuing the development of a prototype clinical decision support tool (Donation Advisor, or DA) that will also accurately quantify organ ischemia during DCD and conduct a pilot randomized controlled trial (RCT) to provide a standardized, personalized assessment of donor and organ suitability. The inability to predict time to death of DCD eligible patients results in uncertainty, inhibition of DCD initiation, and in frequent unsuccessful DCD attempts that harm families and waste scarce resources. Our pioneering bedside application of biomedical engineering combines a novel waveform-derived predictive model rooted in variability analysis along with personalized assessment of organ ischemia during WLSM, within a feasibility RCT. A world-class, multidisciplinary team is poised to complete this study in Ontario, implementing a tool that is automated, adaptive, easy to understand, and timely. The study outcomes have the capacity to markedly improve DCD efficiency, donor effectiveness, family experience, and reduce costs throughout Canada.


Variability monitoring

Autonomic Nervous System function and ability to dissipate heat during environmental and physical stresses

WHO estimates that heat exposure will cause >100,000 additional deaths per year during the 2050s. To address this vital public health issue, experts have urged the development of criteria to identify and, ultimately, protect individuals who are at susceptible to heat stress, such as the rapidly aging population and workforce. A heat stress monitoring system capable of identifying individuals at risk and providing a probability of developing heat related injuries within a given time frame can play an important role in preventing and mitigating the public health risks caused by increased ambient temperatures. HRV monitoring can be an important tool for assessing the impact of heat stress and exercise on vulnerable individuals, with implications for environmental health policies. In collaboration with Dr. Glen Kenny, we are investigating the short- and long-term changes in ANS function and in the ability to dissipate heat, caused by exercise-induced or passive heat stress and related to age, sex, chronic conditions (diabetes, hypertension), environmental conditions (e.g. cooling, protective clothing) and physical characteristics in adults. Using HRV and RRV as indirect measures of cardiopulmonary health and function during exercise or heat stress, we are looking at key metrics that could provide a repeatable and non-invasive assessment of heat-stress susceptibility, predict and prevent incoming heat-related injuries. This could generate appropriate heat management activities (health warning system, emergency response plan, heat adaptation strategies) to protect health during EHE, based on the identification of age- and disease-specific high risk ambient and exercise conditions.


Assessment of Cerebral Autoregulation after Sub-Arachnoid Hemorrhage: Variability in Transcranial Doppler signals

The brain is one of the most metabolically active organs in the human body, accounting for approximately 20% of the body resting energy consumption despite weighting only 2% of the total body mass. While a reduction in perfusion leads to ischemic injury, overabundant cerebral blood flow is also problematic as it may lead to breakdowns in the blood-brain barrier resulting in severe neurologic sequelae. Thus, stringent regulation of cerebral blood flow is crucial to the maintenance of brain function. This physiologic mechanism called cerebral autoregulation adjusts the resistance of the cerebral vasculature with the purpose of maintaining a constant cerebral blood flow regardless of changes in systemic blood pressure. Aneurysmal subarachnoid hemorrhage is a life threatening condition with an estimated annual incidence of 9.1 patients per 100,000 and case-fatality rates close to 40%. Surviving an episode of subarachnoid hemorrhage is often associated with the early loss of the ability to regulate cerebral blood flow. This is critical, as approximately 80% of patients lacking cerebral autoregulation are at risk of developing delayed cerebral ischemia; a condition that is the leading cause of morbidity, mortality and disability after surviving the rupture of an intracranial aneurysm. Consequently, accurate detection of impairment in cerebral autoregulation can allow early identification of patients at risk of delayed cerebral ischemia, so effective preventative measures could be established. Transcranial Doppler (TCD), a non-invasive ultrasound method that measures cerebral blood flow velocity in large intracranial vessels has been used to study cerebral autoregulation as it can reflect changes in cerebral blood flow with acceptable accuracy. Traditionally, cerebral autoregulation is evaluated by linear correlations between blood flow velocity and pressure, whereby the presence of correlation suggests impaired cerebral autoregulation. However, cerebral blood flow is thought to be a non-linear and non-stationary phenomenon and regulation of cerebral blood flow requires short and long-term adjustments to match substrate delivery with metabolic demands. Preliminary studies suggest that changes in cerebral blood flow variability may increase the likelihood of adverse events. Therefore, nonlinear variability metrics are better suited to identify patients at risk of losing their ability to regulate cerebral blood flow. In this pilot study led by Dr. Rosendo Rodriguez (Ottawa Hospital Research Institute) we are assessing the variability of cerebral blood flow velocity, blood pressure and heart rate over time looking a fractal and chaotic metrics, by using TCD in age- and gender-matched healthy subjects and patients with aneurysmal sub-arachnoid hemorrhage.


Non-invasive fetal health monitoring

In collaboration with Dr. Martin Frasch (University of Washington, Seattle, WA), we are researching the applications of continuous fetal heart rate variability (fHRV) analysis for early detection of hypoxic-acidemia near term as well as the mechanisms and the manipulation of the fetal neuroimmune response to inflammation. Fetal HRV can be used as a non-invasive tool to detect foetal asphyxia near-term better than currently-available FHR monitoring. Multidimensional fetal HRV analyses can also provide a non-invasive approach to identify fetuses developing inflammation and acidemia and thus reduce risk of lasting neurological deficit. We and others have successfully linked fHRV metrics with inflammatory markers in the blood, gut and brain in animal models of fetal development, with potentially important applications for the monitoring of adverse events (acidemia, prenatal stress) in human fetal development. In collaboration with Drs. Helena Soares and Lavinia Schuler-Faccini (INAGEMP - Departamento de Genética - Instituto de Biociências, Universidade Federal do Rio Grande do Sul, Brazil), we have also been investigating the use of fHRV to help identify toddlers exposed to Zika virus during pregnancy.


Other Projects

We regularly support collaborators and provide assistance within our areas of research expertise. Please check our publications page for DAL publications from other collaborative projects.