EHR-based model predicts serious decline in ICU

Flagging patients earlier provides more time for intervention
Tools

An EHR-based model was more effective in predicting intensive care unit cardiopulmonary arrest or death than prior risk models, including one based on human judgment, according to a study published at BMC Medical Informatics and Decision Making.

The study, from Parkland Hospital in Dallas, looked at predictors for cardiopulmonary arrest, acute respiratory compromise and unexpected death, known as RED events, using data from the previous 24 hours. That data included vital signs, laboratory data, physician orders, medications, high-risk floor assignment, and the Modified Early Warning Score (MEWS), among other treatment variables.

The researchers found the automated model outperformed the MEWS model on sensitivity and specificity, and identified patients destined to have RED event on average 16 hours beforehand.

It had twice the sensitivity of the human-based Rapid Response Team (RRT) model (51.6 percent vs. 25.8 percent) with only a small tradeoff in specificity (94.3 percent vs. 98.8 percent), the authors wrote. It also flagged patients likely to undergo this serious deterioration 5.7 hours sooner than the RRT.

The study was retroactive, so it had no effect on clinical outcomes, a topic for further research, the authors noted.

University of Michigan researchers say EHRs should be used more frequently to calculate the risk of death of patients in the ICU to improve patient flow and triage. After looking at records of 100,000 patients admitted to VA acute care hospitals, it found ICUs admitted an overly high number of patients who had cardiac illness but were relatively well.

Meanwhile, a study from the Massachusetts Institute of Technology looking at lab tests for intensive-care patients with gastrointestinal bleeding found that a computer model based on 11 measurements could accurately classify more than 80 percent of both necessary and unnecessary tests. The researchers achieved an average reduction of 50 percent of eight common gastrointestinal lab tests.

FierceHealthIT is sponsoring an executive breakfast at HIMSS13: Using Predictive Analytics to Improve Care and Efficiencies on Wednesday, March 6. Sign up to join usCardiologist Eric Topol, a professor of genomics at The Scripps Research Institute in San Diego and author of "The Creative Destruction of Medicine: How the Digital Revolution Will Create Better Health Care," is slated to give the keynote address at the conference.

To learn more:
- read the research

Related Articles:
Researchers: EHRs can improve ICU patient flow
Predictive analytics can slash unnecessary GI lab tests by half
Eric Topol: Medical technology revolution needs validation to move forward