According to a study reported in the online journal BMJ Innovations, an artificial intelligence (AI) system created from the details of individual heartbeats captured on an ECG (electrocardiogram) can effectively identify diabetes and pre-diabetes.
Image Credit: Shutterstock.com/ Lightspring
The technique might be used to test for the disease in settings with limited resources, according to the researchers, if it is shown effective in broader investigations.
In 2019, it was predicted that 463 million individuals worldwide had diabetes. Moreover, detecting the disease in its early stages is essential for avoiding subsequent significant health issues. But the measurement of blood glucose plays a significant role in diagnosis.
This is not only invasive but also hard to roll out as a mass screening test in low-resource settings, the researchers point out.
Even before obvious blood glucose changes, the cardiovascular system experiences early structural and functional changes that are visible on an ECG heart trace.
The objective of the study was to see if machine learning (AI) techniques could be utilized to maximize the screening capability of the ECG to identify high-risk individuals for type 2 diabetes and pre-diabetes.
They drew from those who took part in the Diabetes in Sindhi Families in Nagpur (DISFIN) study, which investigated the genetics of type 2 diabetes and other metabolic features in Sindhi families in Nagpur, India, who were at high risk of developing the condition.
Families residing in Nagpur, which has a high population of Sindhi people, and who had at least one known instance of type 2 diabetes, were included in the research.
Participants gave specifics about their own and their families' medical histories, described their usual diets, and underwent a full range of clinical evaluations and blood testing. 61% of them were women, with an average age of 48.
The diagnostic standards outlined by the American Diabetes Association allowed for the identification of pre-diabetes and diabetes.
Pre-diabetes and type 2 diabetes were both quite common, with prevalence rates of 30% and 14%, respectively. Additionally, the prevalence of other significant comorbid diseases, including high blood pressure (51%), obesity (about 40%), and disordered blood fats (36%), as well as insulin resistance (35%), was similarly considerable.
Each of the 1262 subjects had a conventional 10-second 12-lead ECG cardiac tracing. And for each of the 10,461 single heartbeats that were captured, 100 distinct structural and functional characteristics for each lead were merged to create a prediction algorithm (DiaBeats).
The DiaBeats algorithm swiftly identified diabetes and pre-diabetes based on the size and shape of individual heartbeats with an overall accuracy of 97% and a precision of 97%, regardless of important variables, such as age, gender, and underlying metabolic diseases.
Significant ECG parameters consistently matched the biological mechanisms known to underlie the cardiac alterations linked to diabetes and pre-diabetes.
The trial participants were all at high risk of diabetes and other metabolic problems, making them unlikely to reflect the broader population, according to the researchers. Those who were taking prescription medications for diabetes, high blood pressure, high cholesterol, etc., found DiaBeats to be a little less accurate.
Furthermore, data on individuals who developed pre-diabetes or diabetes was not available, making it impossible to assess the effects of early screening.
The study team concluded, “In theory, our study provides a relatively inexpensive, non-invasive, and accurate alternative [to current diagnostic methods] which can be used as a gatekeeper to effectively detect diabetes and pre-diabetes early in its course.”
They added, “Nevertheless, adoption of this algorithm into routine practice will need robust validation on external, independent datasets.”
Journal Reference:
Kulkarni, A. R., et al. (2022) Machine-learning algorithm to non-invasively detect diabetes and pre-diabetes from electrocardiogram. BMJ Innovations. doi: 10.1136/bmjinnov-2021-000759.
Source: https://www.bmj.com/