J. Randall Moorman, M.D., is Professor of Medicine, Physiology and Biomedical Engineering at the University of Virginia, and Chief Medical Officer of Advanced Medical Devices, Diagnostics and Displays Inc. (AMP3D). A clinical cardiologist, he develops predictive analytics monitoring models for early detection of patient deterioration using advanced mathematical time series, machine learning, and other analytical methods. He is the founding Director of the UVa Center for Advanced Medical Analytics, and Editor-in-Chief of Physiological Measurement.
Sponsored Content Provided By Premier
Uncover the patterns. Put your clinical data to work.
Millions of patient data points are collected every day. Enable proactive care and improve outcomes.
Dr. Randall Moorman, M.D.
Professor of Medicine
University of Virginia
Can you share some insights on the concept of patterns in the data?
RM: By uncovering and connecting the millions of data points we begin to see “signatures” of potential serious events before the clinical signs are there. These patterns of data, along with venue- and event-specific models, can help us move to proactive, rather than reactive care. Without these techniques, patients can get sick right under our noses, and that can result in poor outcomes for patients, hospitals, and insurers:
- Increased morbidity and mortality
- Longer lengths of stay
- Substantial unreimbursed costs of care
Every day in healthcare we react to something that happens to the patient. A lot of times a patient is moving steadily down the track . . . headed toward a cliff . . . hours before a formal diagnosis occurs. Patterns in the data, presented to clinicians in the context of their normal workflow, can help care-givers apply the brakes before the train gets to the end of the track and thereby prevent unnecessary patient harm and increased costs of care.
How are patterns effectively surfaced to clinicians in real-time?
RM: By simple visualization that does not require interpretation. This is the key to adoption and practical use. Let’s take the surgical intensive care unit (ICU) as an example. At the nursing station is a large flat screen monitor mounted to the wall. On the monitor are visual data points. Each patient bed number is displayed on the monitor with a simple visualization that looks like a comet. The comet rises or falls based on the acuity of the patient.
The head of the comet gives us the patients risk of clinical adverse events in the next 6 to 12 hours. The length of the comet tail and the brightness tell us how rapidly the patient has deteriorated or how the patient has progressed over the prior 3 hours. Visual indications are simple, intuitive, and immediately actionable. There is no number to quantify or interpret. The comet simply goes up – if patient is at risk. When the comet goes down – the risk of a prospective bad event decreases.
Is this method of displaying data patterns effective?
RM: Yes. It can warn of patient deterioration 6 to 12 hours in advance for clinical events like sepsis, hemorrhage leading to large unplanned transfusion, respiratory failure leading to urgent unplanned intubation, and patient deterioration leading to ICU transfer—among other things.
In fact, in a recent publication, we demonstrated the use of this method reduced the diagnosis of sepsis shock by 52%.1
Is it difficult to implement this predictive analytics technology?
RM: As opposed to other predictive applications – there is no new data to assemble. Our method leverages data already collected in the EHR, as part of normal patient care, and adds the continuous hemodynamic monitoring data in real time.
The research to build the algorithms to uncover and present patterns is established and continues to grow.
- Over 150 patient years of monitoring data
- More than 2 million clinical observations
- Manual review of 10,000+ patient files
- 1,000+ clinician adjudicated events
It is not enough to identify the patterns in the data. Just connecting proverbial “dots” can often lead to bad analysis. Our more than 300 models start with clinician-validated events documented in the patient record. The data is analyzed—including continuous physiological monitoring, lab tests, vital signs, and more—leading up to the events at issue. Models are venue- and event-specific, presented in a visual interface, that a clinician—at a glance — can know what to do without breaking stride on the way to see a patient at risk.
The web-based technology allows healthcare organizations to choose the best way to grasp and surface the visualization based on their clinical workflow: large displays, computers or smart devices.
Sponsored By:
To learn more about predictive analytics, visit theradoc.com.
Sponsored Content