New Delhi: Amid speculation about Artificial Intelligence’s role in medicine and the dangers of potential hallucinations, researchers have developed AI models that can pore through electronic health records and electrocardiograms to identify people who are at risk of a sudden cardiac arrest.
The study was published on May 11 in JACC: Advances, a journal of the American College of Cardiology. It was conducted by researchers at the University of Washington School of Medicine, in collaboration with scientists from Massachusetts General Hospital and the Broad Institute of MIT and Harvard.
A sudden cardiac arrest continues to claim millions of lives every year, and has a survival rate lower than 10 per cent in most parts of the world.
“Using artificial intelligence applications and health records data, the prediction of cardiac arrest in the general population is feasible,” Neal Chatterjee, a cardiologist at the University of Washington School of Medicine, and the study’s lead investigator, said in a press release.
Three data sets
Using data from a test population of nearly 1.7 million patients in the US healthcare system, the researchers built three AI models trained on three different datasets. The first dataset looked only through electrocardiogram (EKGs), the other exclusively analysed electronic health records (EHR) and the third AI model was built on a combined set of EKG and EHR data sets.
To validate and verify the AI models, researchers divided the patients into three groups. The first, the training cohort, was made up of 993 individuals who had experienced an out-of-hospital heart attack between 2013 and 2021, and 5,479 similar patients who had not.
Through the training cohort, the AI model learned to recognise the kind of EHR and EKG data entries which were associated with a higher cardiac arrest risk.
The researchers then gave the AI model a new set of patients—463 cardiac arrest cases from 2022-2023 and 2,979 control patients. The team found that the AI model could successfully identify high-risk and low-risk patients in the new group, too.
Eventually, the AI model was tested on a real-world cohort, including 39,911 people who had received EKGs in 2021. Researchers then looked at the records of patients who experienced heart attacks over the next two years to check whether they matched the AI model’s predictions.
As per the team’s findings, the AI model correctly predicted 153 of 228 high-risk individuals who then went on to have a heart attack.
Also read: PCOS to PMOS. What the name change mean for patients
Limitations of the research
While the findings suggested that AI models could predict the risk of cardiac arrests, Chatterjee highlighted that more research is needed to determine what would be the best course of action after an AI model identifies a high-risk patient.
“We need to figure out which follow-on studies to pursue to understand what we do with this patient information. What screening, what surveillance, what intervention is warranted?” he said.
However, one of the key limitations of the study is that the data were sourced from one health care system, and whether the findings can be generalised for the larger public is not known. In the third stage of its training, the AI model’s predictions were based on data about individuals who received an EKG. Therefore, the model’s predictions may not hold for patients who have not received an EKG.
(Edited by Saptak Datta)