Chest X-ray AI analysis potentially predicts ASCVD risk for patients with missing EMR data
Click Here to Manage Email Alerts
Key takeaways:
- Artificial intelligence may predict 10-year heart disease risk from a single chest X-ray.
- For individuals with missing ASCVD risk calculator data, AI may assist risk stratification for primary prevention.
Artificial intelligence analysis of a single chest radiograph may predict 10-year CVD risk beyond the traditional atherosclerotic CVD risk calculator, even among individuals with missing electronic medical record data, researchers reported.
“Because the necessary input variables to calculate the ASCVD risk score are often not available in the EMR, other approaches for population-based screening are desirable to identify individuals at high risk who are likely to benefit from a statin. Opportunistic and automated risk assessment using data commonly available in the EMR (such as chest radiographs ) could complement the ASCVD risk score and may be helpful to improve statin uptake,” Jakob Weiss, MD, senior resident in radiology at Massachusetts General Hospital, and colleagues wrote. “Deep learning offers new possibilities for risk prognostication beyond currently established methods by using routine imaging, such as chest radiographs, which are among the most common tests in medicine.”
Weiss and colleagues developed a deep learning model to assess risk for CV death and future major adverse CV events from a routine chest radiograph image and compared the model’s performance with that of the 10-year ASCVD risk score for assessing statin eligibility, defined as an ASCVD risk score of 7.5% or higher.
The AI model was developed using chest radiographs from the PLCO Cancer Screening Trial and externally validated a cohort of 8,869 outpatients with unknown risk due to missing ASCVD risk score inputs (mean age, 60 years; 82% white; 45% men) and 2,132 outpatients whose ASCVD risk score could be calculated (mean age, 60 years; 76% white; 36% men).
The findings were published in the Annals of Internal Medicine.
Among individuals with unknown ASCVD risk, after adjustment for risk factors, those with a risk of 7.5% or higher as predicted by the deep learning model had higher 10-year risk for MACE compared with the ASCVD risk calculator (adjusted HR = 1.73; 95% CI, 1.47-2.03).
Among individuals with known ASCVD risk, the deep learning model predicted risk for MACE beyond the ASCVD risk calculator (aHR = 1.88; 95% CI, 1.24-2.85), according to the study.
In addition, the AI model and the traditional ASCVD risk calculator were concordant for statin eligibility in 69.7% of patients, and in the remainder that were discordant, the AI model up-classified half as statin-eligible.
Among participants with missing inputs for the CVD risk score, those with a 10-year ASCVD risk estimate of 7.5% or higher derived by the AI model had a 10-year incidence of MACE of 11.8%, which indicates this group may benefit from primary prevention, the researchers wrote.
“On the basis of a single routine chest radiograph image, a deep learning model can predict 10-year incident MACE,” the researchers wrote. “The model could be applied in most patients for whom lipids or other inputs to the ASCVD risk score are not available. Opportunistic screening of chest radiographs may help identify individuals at high risk for cardiovascular disease, prompting risk factor assessment and targeted prevention.”
Collapse
Click Here to Manage Email Alerts