Can a medical record predict the risk of cardiac arrest?

June 14, 2024 – Until now, known risk factors for cardiac arrest were limited to cardiovascular disease, such as high blood pressure and high cholesterol. Research published on June 11 identifies a number of new factors – such as being single and alcohol and drug abuse – that are also strongly associated with out-of-hospital cardiac arrest (OHCA).

“Our current understanding of the risk of cardiac arrest is based on data collected decades ago in populations that do not necessarily represent our diverse societies today. There is a scientific need to better understand the factors underlying this public health problem,” said Dr. Neal Chatterjeea cardiologist at the UW Medicine Heart Institute in Seattle.

Chatterjee is the lead author of a study in which machine learning models searched the electronic health records of more than 1.5 million patients to identify factors associated with OHCA. The The results appear in the journal Circulation.

“Over 400,000 people in the United States experience cardiac arrest each year,” said Chatterjee. “We have invested millions of dollars in research over the past decade to improve survival after such events, and the results have been modest. With this work, we hope to advance prediction and prevention.”

The researchers began with the King County Out-of-Hospital Cardiac Arrest Registry. The names of these people were matched to the electronic health records of 1.5 million patients treated between 2010 and 2021 at UW Medicine, a large academic health system in Seattle.

This search identified 2,366 patients (64% male) who had experienced sudden cardiac arrest. Their anonymized medical records were analyzed by three machine learning models for common medical and non-medical factors. A control population of 23,660 other patients in the system was developed at a 1:10 ratio and matched to the OHCA patients by age and sex.

The researchers reported “significant differences in demographics, vital signs, electrocardiographic parameters, comorbidities, and medication distribution between OHCA cases and controls.”

The consensus predictors of OHCA in the three machine learning models included:

  • Substance abuse disorder
  • Fluid and electrolyte disturbance
  • Alcohol abuse
  • no longer corrected QT interval
  • higher heart rate
  • Demographics (minority race, single marital status)

“When it comes to the causes of cardiac arrest risk in our community, it turns out that there are several noncardiac factors, such as substance abuse disorders, that are just as important as the cardiac factors we have known about for years,” Chatterjee said.

The researchers then adjusted the machine learning model's metrics to better understand how frequently it would identify OHCA risk if used in UW Medicine's 1.5 million patient population. The positive predictive value for OHCA was calculated to be about 1.5% per year – 1-2 people out of 100 per year – in contrast to the most recent risk estimate based solely on cardiovascular factors: 1 in 1,000 per year.

Being told that the probability of suffering a sudden cardiac arrest in the next year is 1 to 2 percent is a more meaningful conceptualization of risk than saying 1 in 1,000, says Chatterjee.

“For the first time in half a century, we are seeing the true complexity of the risk factors underlying the risk of cardiac arrest. For each individual patient, the interventions required to reduce this risk must be individualized and tailored,” he said.

Hypothetically, Chatterjee continued, a machine learning model could traverse a health system's electronic medical records to identify patients at highest risk. During an office visit with a primary care physician, those patients could be made aware of these personal factors, some of which would likely be modifiable.

The model could also help influence public health agencies' investments in tackling the root causes of population risk, he said. For example, if you know that drug and alcohol addiction are the leading causes of cardiac arrest risk, you can potentially get a higher return on investment in initiatives that address those risks.

For more informations: