Harnessing Technology to Improve Patient Care

2016 - 1 February - Technological Game Changers
Peter Laussen, MB, BS, FCICM; Mel Almodovar, MD; Christopher Horvat, MD; Robert S.B. Clark, MD; Leo Anthony Celi, MD, MS, MPH
What innovative technologies can critical care practitioners leverage to improve patient outcomes? Critical Connections asked a handful of experts to address this question. See what cutting-edge tools they identified as potential game changers.


T3 (Tracking, Trajectory and Trigger) Tool: Platform Toward Precision Critical Care
Peter Laussen, MB, BS, FCICM; Mel Almodovar, MD

The T3 (tracking, trajectory and trigger) tool is a flexible and scalable software platform deigned to iteratively utilize the huge amounts of continuous physiologic data streaming from devices and monitors at the bedside. Conceived and developed in the cardiac intensive care unit (ICU) at Boston Children’s Hospital in Boston, Massachusetts, USA, in 2010, and licensed by Etiometry LLC (Boston, Massachusetts) in 2013, it has also been deployed at the Hospital for Sick Children in Toronto, Ontario, Canada, since 2013, and more recently at Children’s National Hospital in Washington, DC, USA, and Great Ormond Street Hospital for Sick Children in London, United Kingdom. With other sites also considering implementation, an overarching goal is to establish a network for sharing and analyzing these large datasets.

We know that ICUs are dynamic, resource-intense and data-rich environments in which large streams of continuous physiologic data are available, and in which patients’ clinical courses can change rapidly and unexpectedly. Rapid changes in physiologic variables drive equally rapid changes in decisions and treatment. A vital aspect of critical care management is the ability to perceive an evolving clinical picture—to be predictive, rather than reactive and prescriptive, in our management practices. This requires the visibility, interpretation and analysis of all available clinical, physiologic and laboratory data streams. Yet in our ICUs, data is presented to clinicians from heterogeneous and independent sources at the bedside; the sheer volume and complexity of data can often be overwhelming, difficult to integrate and subject to variable interpretation among clinicians. T3 is a solution to this problem.

Here are some of the features of the T3 platform.

1. It allows the capture and storage of continuous physiologic data. Digital data streaming from bedside monitors and other devices, including the ventilator and near-infrared spectrometer, are captured at five-second increments via a Health Level Seven International (HL7) feed from the bedside monitor to production and research servers. The Web application is hosted in a Java Jetty Web server, and uses a data cache that is subscribed to the stream of patient data from the HL7 broker and the streaming calculators (where novel calculations such as coronary perfusion pressure are made), making for fast page loads and low database use.  

The T3 platform is vendor agnostic, can be used with different monitoring systems, and can be linked to other selected data sources, including laboratory and electronic medical record, to aggregate categorical data with the continuous physiologic data streams.
Waveform data (electrocardiography, capnography, plethysmography, and arterial) is being collected onto the T3 platform at the Hospital for Sick Children using the ViNES medical device integration software platform (True Process, Wisconsin).

2. The web-based representation of data enables direct clinician interaction. Multiple data streams can be visualized simultaneously and overlaid upon each other, and the time window can be expanded or contracted for rapid visualization of a patient’s current status or historical course. Annotations, events, and decision data may be added at specified time points, further augmenting the collected data.

3. T3 incorporates a unique analytics platform software, called the Quality Improvement (QI) Sandbox, that allows clinicians to directly analyze patient datasets and perform benchmarking without needing to learn a programming language or possess specialized raw data assembly and manipulation skills. It lowers the barriers to performing large-scale clinical and quality assurance studies.

4. T3 is a platform for the analysis and modeling of continuous physiologic data and for hosting algorithms and trajectory indices. This is a very important feature because it allows for algorithms and modeling to be displayed in real time to clinicians at the bedside. Our first hemodynamic trajectory index, called the stability index (SI), is undergoing evaluation at the Hospital for Sick Children. The SI is analogous to trajectory indices used in financial markets and is, in effect, an early warning signal for critical care. Current work is focused on validating a predictive model-based algorithm that determines the probability of inadequate oxygen delivery, as well as developing the probability of respiratory failure and the physiologic etiologies of both inadequate oxygen delivery and respiratory failure.

Taking Predictive Analytics to Point-of-Care
Christopher Horvat, MD; Robert S.B. Clark, MD
Clinicians caring for hospitalized patients face an ever-growing stream of data. Consultation notes, vital signs, laboratory values, radiology reports, and therapy recommendations are constantly added to a patient’s medical record. Sifting through this barrage of information to extract a clinically salient message can require significant time and skill. Easing this job is the emergence of electronic medical record (EMR) systems, which have regimented this massive load of health information into spreadsheets and text files. Patient information that was historically maintained at the bedside, stored outside a patient’s room and/or archived is now commonly accessed from a clinician’s office, workroom or home computer. 
“Meaningful use” mandates have spurred the adoption of EMRs nationally, and they are now commonplace in both rural and urban healthcare centers. The same computing power that has been leveraged to organize, present and store volumes of data has remained largely underutilized in aiding point-of-care clinical decision-making. Synthesizing the constantly growing load of information at the bedside has remained the task of the clinician. 
The Pediatric Rothman Index (pRI) (PeraHealth) is an EMR-based tool that we have adopted at the Children’s Hospital of Pittsburgh to facilitate clinician recognition of increasing acuity or impending clinical deterioration. The pRI is unique in its ability to automatically aggregate 26 clinically important variables housed in the EMR into a single, acuity-related score. Variables include vital signs, laboratory data, and a head-to-toe multisystem nursing assessment. The pRI is derived from the Rothman Index, a score developed for use in hospitalized adults that has been demonstrated to be predictive of mortality and hospital readmission. In hospitalized children, we performed a retrospective review of inpatient pRI scores surrounding medical emergency responses at our institution. We reported that the pRI recognized children at risk for critical deterioration with high specificity,(1) identifying a predictive analytic tool and setting the stage for point-of-care application.   
The final hurdle was practical, and required a design engineering solution. Simply put, how were we going to use the EMR to generate and send alerts for the right patients, at the right time, to the right care providers? Solving this hurdle required a cooperative effort between our hospital’s information technology (IT) group and the engineers at PeraHealth, but is now operational. In essence, the EMR is surveyed every 15 minutes for patients who fall below specified pRI thresholds. An electronic alert is then sent to networked cell phones, personal physician cell phones, and/or pagers of dedicated response team members. The implemented work flow calls for bedside assessments by the primary service and nursing supervisors with subsequent notification of pediatric intensive care unit (PICU) nursing staff and physicians for patients triggering “high alerts,” or deployment of a select group of the hospital’s medical emergency response team for patients triggering “very high alerts.” Moving forward, we are examining the impact of a pRI-guided work flow on the volume and nature of PICU admissions. 


A great deal of unharnessed computing power exists within the IT infrastructures that have sprouted as hospitals have adopted EMR systems. The pRI is likely to hold additional utility as a point-of-care predictive analytic tool beyond the recognition of ICU-worthy deterioration. As the complexity of care grows, other automated tools that merge tech-savvy ingenuity with the daily tasks of the busy clinician will be welcome.


1. da Silva YS, Hamilton MF, Horvat C, et al. Evaluation of electronic medical record vital sign data versus a commercially available acuity score in predicting need for critical intervention at a tertiary children’s hospital. Pediatr Crit Care Med. 2015 Sep;16(7):644-651.


The MIMIC Approach
Leo Anthony Celi, MD, MS, MPH

In 2011, the Institute of Medicine (IOM) published the report Clinical Practice Guidelines We Can Trust at the request of the U.S. Congress through the Medicare Improvements for Patients and Providers Act of 2008. Reporting on the state of existing clinical guidelines, IOM concluded that, for many clinical domains, high-quality evidence is lacking or nonexistent, and practice guidelines often rely on low-quality evidence or expert opinion.
In parallel, growing skepticism about the reliability of published studies has recently roused the scientific community. Contributing factors are numerous, and include the community’s emphasis on sensational discoveries rather than reproducibility, an academic incentive system that promotes careerism, lack of access to unpublished studies, and conflict of interest with industry. A movement calling for data sharing and continuous and transparent peer review was born to address some of these issues.  
Digitalization of health data may provide an opportunity to generate diagnostic and treatment recommendations when there are no existing guidelines, but it may also open the floodgates to more irreproducible studies unless safeguards are put in place. Clinicians who are most familiar with the questions and the context in which they exist should play a key role in health data analytics, but clinicians are typically lacking in expertise in data science. 
The Laboratory of Computational Physiology (LCP) at the Massachusetts Institute of Technology developed and maintains the Medical Information Mart for Intensive Care (MIMIC), an open-access de-identified database of patients admitted to the intensive care units (ICUs) at Beth Israel Deaconess Medical Center in Boston, Massachusetts, USA. MIMIC III spans the period from 2002 through 2011 and contains more than 60,000 ICU admissions. In addition to the database, the group organizes datathons, which bring together required experts from different fields in a venue that encourages constructive collaboration, group learning, error checking and methodological review during the initial design and subsequent phases of research. A software platform is provided to the participants that connects directly to the database, allows documentation and information sharing among the team members regarding the research question and study design and, most importantly, facilitates archival and sharing of codes and queries between the teams for cohort selection, variable extraction, data visualization and analysis.
Aware of the limitations of a single-center database, the group partnered with the Philips eICU Research Institute to help curate the Philips eICU database, consisting of close to three million ICU admissions from across the United States. A de-identified subset of 200,000 patients will be released to investigators free of charge via PhysioNet, a research resource also developed and maintained by LCP that offers well-characterized physiologic data and related open-source software to the biomedical community. Discussions are underway with hospitals in Belgium, Brazil, the United Kingdom and Greece to scale MIMIC to include ICU data from around the globe. An international database affords numerous benefits. The practice variations across ICUs that can be examined is much richer. Cross-validation of models among institutions will determine which findings are institution specific and which are generalizable. Most importantly, knowledge discovery is accelerated exponentially if more investigators participate in ICU data analysis.
But the value of large amounts of data hinges on the ability of researchers to share data, methodologies and findings in an open setting. The multi-expertise viewpoints inserted openly into the research process—the MIMIC approach—would aid in the conception, development, process and publication of research that is more reliable, while hopefully remaining as interesting, innovative and important as that produced by the current system.