multiparameter, multipatient physiologic data harvest
TRANSCRIPT
E6 Wakefield Roundtable Discussion
submodels into the decision model to allow more information to beextracted from multiparameter, multipatient digitized data.
doi:10.1016/j.jcrc.2010.05.016
Methodologies and applications of continuous variability analysisCarolyn McGregor
A century ago, a new baby was just as much a cause for concernas it was a cause for joy. That was because more than 1 in 10 infantsdied at birth or shortly thereafter. Although that number has droppedsignificantly in the past 100 years, the number of premature births hasnot decreased as much—in fact, preterm births may actually be on therise in Canada. Today, 1 out of every 14 Canadian mothers will givebirth prematurely (ie, 7.1%). In New SouthWales, Australia, in 2005,7.2% were born prematurely, whereas 8.1% were nationally. Theseearly births, which happen in the seventh and eighth month ofpregnancy, are responsible for three quarters of all infant deaths inCanada. Worse still, even when infants survive, premature babiesmay develop lifelong problems if they are not properly cared for inthe crucial days and weeks after birth. Neonatal intensive care units(ICUs) boast state-of-the-art medical devices to monitor and supportpremature babies; however, neonatologists are increasingly weigheddown by vast quantities of charted data and 86% false alarms frommedical devices. In fact, all intensive care units face similar dataissues. Although neonatal ICU admissions accounted for only 15% ofICU admissions in Canada in 2003, they represent the second majorcohort (other than patients older than 62 years), as the average lengthof stay for a neonate is 2 to 3 times as long. Significant opportunitiesexist for infrastructures, methodologies, and application that performcontinuous variability analysis to assist with the real-time support andclinical research of complex, nonlinear body systems.
As Canada Research Chair in Health Informatics, Dr CarolynMcGregor is pioneering new ways and approaches for suchinfrastructures, methodologies, and applications. This presentationoverviews several collaborative research projects that aim toincrease survival rates and quality of life rates for neonates throughimproved technological approaches for real-time clinical manage-ment and clinical research. Key contribution areas includeinfrastructures, methodologies, and applications for (1) standardsfor the secure collection, transmission, and storage of heterogeneousbody system data; (2) the first on-demand virtual neonatal ICUsupporting rural, remote, and urban neonatal care, known as bushbabies; (3) the real-time cross-correlation of physiologic data to usecomplex condition predictors to generate complex neonatal medicalalerts; and (4) new approaches to data mining techniques to supportnull hypothesis–based clinical research to determine new trends andpatterns that could predict the onset of critical medical conditions.
doi:10.1016/j.jcrc.2010.05.017
Multiparameter, multipatient physiologic data harvestPatrick Norris
Fundamental clinical approaches for assessing vital signs havechanged little since 1903, when Cushing defined the importance ofperiodically measuring a patient's heart rate and blood pressure.
Although physiologic parameters have been added to the milieu,assessment remains largely a manual, periodic process. A growingbody of research suggests value in automated analysis of continuouslysampled data. For example, heart rate variability has been associatedwith onset of sepsis, multiple organ dysfunction syndrome,myocardial infarction, and elevated intracranial pressure. Relation-ships between arterial and intracranial pressure waveforms maypredict loss of cerebral autoregulation. Pulmonary emboli might bedetected early based on characteristics of the pulmonary arterypressure waveform. Despite convincing research results from animalexperiments and small clinical studies, these and similar observationsremain largely unrealized in patient care.
Clinical adoption requires not only compelling scientific evidencefrom relevant patient populations, but also affordable data captureand processing solutions that can be practically implemented at thebedside. To date, technical, economic, and social barriers have madesuch advances difficult. Commercial monitoring systems do notadequately analyze physiologic signals or store data for extendedperiods of time for off-line analysis without extensive customization.As such, even relatively simple trends in vital signs clearly requiringclinical attention may go unnoticed because of human inability toprocess information in busy clinical settings, especially in theintensive care unit where many automated alerts based onphysiologic trends or thresholds are of dubious value.
A number of custom-built systems collect and process bedsidemedical device data with the ultimate goal of improving medicaldecision-making. Furthermore, generic decision-support modelsand monitoring frameworks have been defined specifically withdense physiologic data in mind. However, few systems usingcomplex, generic decision support models have reached routineclinical use. Other systems, without extensive decision supportmodels, have successfully delivered clinical alerts beyond thoseavailable via bedside physiologic monitors, sampled and archiveddense physiologic data to support clinical research, or made medicalmonitor data available remotely. Academic efforts have demon-strated many of these advancements, and commercial solutions arebeginning to offer many similar capabilities.
Unfortunately, commercial solutions are not widely realized untilyears after technology becomes available for purchase, often decadesafter research first suggests clinical value. Regulatory requirementsand market forces limit vendors' ability or willingness to rapidlyprovide new features. Fiscal as well as human-factors constraintsprevent customers from adopting new monitoring technology as itbecomes available. Clinicians and researchers wishing to leverage thisresource are faced with building custom systems for physiologic datacapture and management. Such systems have been technicallydescribed in the literature, but few of these reports detail functionalrequirements needed to support clinical and/or research use of densephysiologic data.
Given ongoing changes in technology, these functional, asopposed to technical, details may be more relevant to those buildingor buying similar systems. This presentation will cover functionalrequirements encountered over the past decade in our work on theSignal Interpretation and MONitoring project and current andplanned methods to meet the evolving requirements of multipa-rameter, multipatient physiologic data harvest.
doi:10.1016/j.jcrc.2010.05.018