Project Details -

Hospitalized patients with cardiac disease are at risk for sudden deterioration from complicating subacute illnesses such as sepsis, hemorrhage, and respiratory decompensation, resulting in significant increases in length of stay and mortality. These subacute illnesses can be mitigated with early recognition and intervention. The use of continuous predictive analytics has the potential to warn providers of impending deterioration, allowing for an earlier window of treatment, when it is more effective and not too late to intervene effectively, thus allowing care to become proactive rather than reactive. By recognizing earlier in the clinical course that a patient is imminently at risk for respiratory failure, sepsis, or hemorrhage, appropriate care may be escalated resulting in better outcomes.

This research will evaluate the impact of CoMET—continuous monitoring of event trajectories—an artificial intelligence tool with visual analytics that displays risk estimates for multiple adverse outcomes. The tool integrates data streams of rapidly changing health information to predict and communicate risk of impending decompensation.

The specific aims of the research are as follows:

Evaluate the impact of predictive analytics monitoring on patient outcomes.
Evaluate the impact of predictive analytics monitoring on clinical action.
Evaluate the impact of predictive analytics monitoring on costs to the health system.

The tool will use artificial intelligence models to incorporate cardiovascular data from patient monitors, as well as vital signs and lab test data from electronic health records. The vast quantity of these data is difficult for clinicians to absorb in aggregate, an issue compounded with the sickest patients who generate thousands of datapoints in short periods of time. The creation of visual tools allows for predictive analytics monitoring, while reducing the cognitive load for providers taking care of intensive care patients. The impact of the tool will be evaluated with a cluster randomized controlled trial—the Predictive Monitoring: IMPact of Real-time Predictive Monitoring in Acute Care Cardiology Trial (PM-IMPACCT)—comparing patient outcomes, clinical action, and costs for patients in the intervention group versus usual care. It is anticipated that this tool will draw clinicians’ attention to those patients that warrant earlier interventions. Following the completion of this work, the researchers plan to expand their analysis by conducting a multi-centered randomized control trial testing the effectiveness of Co-MET.

Continuous Predictive Analytics Monitoring to Improve Care for At-Risk Patients with Cardiac Disease

Theme:

Optimizing Care Delivery for Clinicians

Subtheme:

Using Real-Time Digital Healthcare Data to Improve Timely Treatment or Diagnosis

An artificial intelligence digital health tool that identifies patients on the verge of clinical deterioration may allow for faster intervention and a reduction in morbidity and mortality.

Recognizing complications and intervening quickly for hospitalized patients with cardiac disease can improve outcomes

Hospitalized patients with cardiac disease can quickly deteriorate and suffer from sudden complications that can lead to increased morbidity and length of stay, and in some cases, death. These complications can be lessened by early recognition and timely intervention, which is a challenge for healthcare providers who need to be able to identify these patients and allow for earlier clinical action. “These are patients on acute care floors, not in the ICU, and they unexpectedly get sick. And, so any clinician would tell you that we would want a crystal ball to be able to detect what patients are going [to] get sick unexpectedly,” said Dr. Jessica Keim-Malpass. “And as a nurse, I can tell you firsthand that these patients have very insidious symptoms. Often they get sick incredibly quickly before noticeable clinical symptoms and it can have devastating outcomes.”

Continuous monitoring using artificial intelligence and data visualization provides warning signs

While hospitals monitor patients and data, including vitals and labs, Dr. Keim-Malpass wanted to see if continuous predictive analytics monitoring at the bedside—including signals from EKGs captured every 2 seconds to get individual heartbeats and breaths—could warn providers of patients at risk for deteriorating health. The monitoring system gives providers an earlier window of treatment when intervention is most effective.

Drs. Keim-Malpass, Jamieson Bourque, and their University of Virginia team are testing continuous monitoring of event trajectories (CoMET)—an artificial intelligence tool with visual analytics that displays risk estimates for multiple adverse outcomes. The tool integrates data streams of rapidly changing health information to predict and communicate risk of impending decompensation. By shifting the care from reactive to proactive, providers can recognize earlier in the clinical course that a patient is imminently at risk for respiratory failure, sepsis, or hemorrhage, and intervene with appropriate care, resulting in better outcomes.

Screenshot of CoMET artificial intelligence tool

In a demo of CoMET, Dr. Keim-Malpass explains, “This tail-to-head, like a comet, represents how the patient's doing in the past 3 hours. You can see for one patient, they started off with moderate risk, and in over 3 hours, they have increased their risk. With the data visualization, your eyes are really meant to be drawn to this patient.”

The clinical information automatically gathered includes readily extracted information in the electronic medical record from the Clinical Data Warehouse, such as numerical values in flow sheets, labs, blood cultures, plus information that is not in the medical record or data warehouse, such as continuous cardiorespiratory monitoring data. A CoMET score is calculated, then displayed (or not displayed) on monitors to draw the clinician’s attention to patients warranting early or extra consideration.

Often, the vast quantity of data are difficult for clinicians to absorb in aggregate, an issue compounded with the sickest patients, who generate thousands of datapoints in short periods of time. That is why the creation of visual tools allows for predictive analytics monitoring, while reducing the cognitive load for providers taking care of these patients.

“All these data are available to clinicians, but anyone would tell you that the level to which algorithms can discern the data and develop risk predictions are certainly way beyond what any human can mentally process and take in.”- Dr. Keim-Malpass

Earlier recognition of deterioration is beneficial to all

Dr. Keim-Malpass and her team are evaluating the impact of CoMET in a cluster randomized controlled trial to see if its use will draw clinicians’ attention to those patients that warrant earlier interventions. While she hopes that the patients will see the biggest benefits, clinicians should also benefit from improved clinical decision making and reduced cognitive burden, as well as reduced costs to the healthcare system overall.