In an intensive care unit (ICU), every second matters. In this fast-moving, high-stakes environment, clinicians must routinely make sense of a constant stream of “time-series data” – the physiologic data generated from monitors, ventilators and other bedside devices – to accurately diagnose and treat patients as well as intervene when problems arise.
In the ICU tens of thousands of data points are generated every second by patient-monitoring equipment, creating an enormous set of patient data. A typical patient in an ICU generates 200 to 300 megabytes of data per day. A patient with an average ICU stay of five to seven days will generate a volume of data similar to what is required to store their entire genome!
The care of critically ill children depends on exquisite attention to detail and the rapid assimilation of clinical, physiological and laboratory data. However, the sheer volumes of complex, real-time data can make it challenging for clinicians to filter the ‘noise’ and hone in on the subtle relationships between precise physiologic data and changes in the clinical status of their patients. All too often, clinicians have little warning of impending patient deterioration, and by the time overt symptoms appear – the ideal window to intervene may have passed.
Applying ‘big data’ to improve patient care and save lives is what motivates the work of Craig Rusin, PhD, Associate Professor of Pediatrics-Cardiology and Principal Investigator of the Predictive Analytics Lab at Baylor College of Medicine.
Dr. Rusin, an engineer and data scientist, was recruited to the Division of Cardiology and Texas Children’s Heart Center in 2011. Dr. Rusin’s arrival marked the beginning of a new era for the Heart Center – with him being positioned at the interface of patient care, predictive analytics and data science.
The Predictive Analytics Lab, comprising a group of PhD-level mathematicians, engineers and data scientists, as well as undergraduate and graduate students, evolved from Dr. Rusin’s work within the cardiology division.
Dr. Rusin and his team develop algorithms, or what he refers to as “virtual patient monitors,” to automate clinical surveillance of time-series data in critical-care and high acuity clinical environments. These algorithms constantly process physiologic data in real time, identify patterns and warn of life-threatening events in advance.
The development of algorithms is possible because even a seemingly sudden or acute clinical deterioration actually builds up over a short time, changing the physiologic dynamics of the patient in the process. These precursors are often too subtle for human clinicians to identify, but a computer can be trained to detect such patterns. By using technology to anticipate adverse events, the health care team can render care proactively instead of reactively, with tremendous potential for improving survival rates in critically ill patients.
In this way, the team at the Predictive Analytics Lab aims to deliver the right information into the hands of the right clinicians at the right time so they can better navigate complex cases with large-volume data loads and make the best, most informed care decisions possible.
“Algorithms complement the abilities of existing clinicians by providing a secondary layer of scalable patient surveillance. They do not get sick, take vacation or get distracted by other patients,” Dr. Rusin said. “This advancement in technology represents a fundamental shift in the way that traditional patient monitoring and surveillance has been conducted over the last 50 years.”
“Software development moves much faster than other traditional medical technology spaces, such as drug development or medical device manufacturing. Biochemists and geneticists may wait decades to bring a medication to market. If they develop something at the beginning of their career, they may not ever see it come to fruition,” Dr. Rusin said. “Algorithms, on the other hand, have a much more compressed development and deployment timeline, allowing them to get into the hands of providers who need them sooner.”
Currently, Dr. Rusin’s team at the Predictive Analytics Lab is studying the validity of a new computer algorithm for newborns and infants with hypoplastic left heart syndrome and related lesions.
Hypoplastic left heart syndrome (HLHS) is a congenital heart defect in which the left side of the heart does not develop correctly in utero. Compared to people with healthy, normal-size hearts, patients with HLHS have small or completely closed mitral and aortic valves and a small left ventricle that is unable to adequately support the circulation of blood through the heart.
If left untreated, HLHS is a uniformly fatal congenital heart defect, so these infants must undergo a series of three operations in order to survive. The first surgery (stage 1) takes place soon after birth; the second surgery between 3 and 6 months (stage 2) and the final surgery (stage 3) between 2 and 5 years of age. Nowadays, these patients have a 10 to 15 percent mortality rate during the first year of life, but the early postoperative course after the first stage of surgery is typically the highest risk period both for mortality and for acute life-threatening deteriorations.
Dr. Rusin and his team discovered that a physiologic fingerprint precedes acute or seemingly sudden patient deterioration – and it can be detected one to two hours in advance, a vital window of time for doctors to stabilize a patient.
The team first tested and validated the algorithm on a cohort of 238 patients at Texas Children’s. The Journal of the American College of Cardiology prominently featured this work.
Following the study’s initial success, Dr. Rusin received grants from the National Institutes of Health and American Heart Association to refine the algorithm and expand the study to three other institutions: Children’s Hospital of Philadelphia, Children’s Hospital Colorado and Nationwide Children’s Hospital. The Journal of Cardio and Thoracic Surgery accepted the results of this most recent multi-institutional study for publication.
The last upcoming step is obtaining Food and Drug Administration approval so that the algorithm can get to market within a few months and start benefiting a wider group of patients.
“As an engineer in the medical space, I get excited about solving tough problems and building technologies that help to save lives,” Dr. Rusin said. “For all of us in the lab, there is no greater reward than seeing our work in action, making a difference for critically ill children and the clinicians who treat them.”