In ST‐segment elevation myocardial infarction (STEMI) mortality risk rises with each passing minute, making medical intervention a race against time.
STEMI occurs with a total blockage of a coronary artery, with the subsequent death of heart muscle tissue. Diagnosis is made with an electrocardiogram (ECG).
When emergency departments (EDs) are evaluated on median door-to-ECG time, the goal is to acquire an ECG in under 10 minutes. If ECG results confirm the STEMI diagnosis, the patient is immediately taken for percutaneous coronary intervention – angioplasty and stents – to avert a serious and possibly fatal outcome.
To explore the potential impact of an artificial intelligence intervention to enable improvements in patient care and risk assessment, researchers at Vanderbilt University Medical Center were part of a 10-center study to identify the types of patients experiencing delays in STEMI diagnosis. The study was recently published in the Journal of the American Heart Association.
Even though the overall median door-to-ECG was 7 minutes, a sizeable subgroup – 37.9 percent – waited more than 10 minutes to receive their ECG after arrival in the ED. Patients in this delayed group were more likely to be female, Black, non‐English speaking, or diabetic.
Dandan Liu, Ph.D., an associate professor of biostatistics at Vanderbilt, is senior author of the study. Liu cautioned that the investigative team was not searching for a “causal relationship” in the discrepancies.
“We’re not saying, ‘because you’re female, the ECG is delayed,’” Liu said. “We’re trying to explore the patterns and identify areas of health care improvement for patients with the disease.”
Maame Yaa A. B. Yiadom, M.D., formerly of Vanderbilt and now an associate professor and vice chair for research in the Department of Emergency Medicine at Stanford University, was the principal investigator for the study team.
“I primarily study variation in clinical outcomes for standardized care delivery processes,” Yiadom said. “When you start looking at outcome differences, you find health disparities that require understanding to improve care equity for all with the disease.”
Limitations of Median Door-to-ECG Time
For at least 30 years, cardiologists have pushed to improve the timeliness of screening, diagnosis and treatment of STEMI in the ED, and median times to intervention have been reduced in many cases.
“A principal opportunity for improvement is moving the ECGs completed during triage up to the arrival intake phase of ED care as frequently as possible.”
“As an emergency physician working at the front lines, I feel it when we don’t hit the mark for patients,” Yiadom said. “Sharing research that highlights opportunities for care improvement is the focus of my work.”
“We wanted to look at the percentage of patients who don’t hit the median. That’s a missed opportunity for process improvement,” she added.
A 10-minute median intervention time is not always evenly achieved, Liu noted. “On the macro level, the time may be below 10 minutes, but when you look at individual patients it can be a completely different story. With individual patients, the median is no longer meaningful.”
The research team suggests looking at the percentage of patients compliant with the care target of 10 minutes as an alternative.
The three‐year retrospective study included 676 ED‐diagnosed patients with STEMI from geographically diverse facilities across the United States. ECGs were performed during ED intake in 62.1 percent of visits, ED triage in 25.3 percent, and main ED care in 12.6 percent.
Chances of being in the delayed ECG group were 51 percent greater for Black patients, 36 percent greater for females, and 7 percent greater for non-English speakers. Non-white Latinos and people with diabetes also had greater likelihood of a delay.
“A principal opportunity for improvement is moving the ECGs completed during triage up to the arrival intake phase of ED care as frequently as possible,” said Liu.
AI and Predictive Models: A Possible Path Forward
Screening all arriving patients for a low-incidence condition is a “needle in a haystack” problem. During the timeframe of this study, the 10 participating EDs collectively cared for approximately 2 million patients and reported a STEMI incidence of 0.1 percent.
Still, a majority of ED patients presenting with possible STEMI symptoms will have a negative ECG. Machine learning predictive models could potentially help balance diagnostic precision and optimal sensitivity, said Liu.
The authors report they have identified some risk factors associated with door-to-ECG delay that should be further confirmed in future studies, highlighting a need to screen differently for separate cohorts of patients.
Machine learning may not replace human efforts, but it has promise in augmenting human performance to improve patient care and timely diagnosis, the authors conclude. The next step is to develop more automated approaches, such as artificial intelligence for EHR systems that will assist in patient evaluation.
