An emerging deep learning model may facilitate more efficient diagnosis of life-threatening conditions and appropriate triage of patients with acute chest pain (ACP) in the ER.

In a retrospective study involving data from over 50,000 patients who presented to emergency departments with acute chest pain, researchers assessed the merits of deep learning analysis of chest X-rays to predict 30-day risk estimates for acute coronary syndrome (ACS), pulmonary embolism (PE), aortic dissection and all-cause mortality. The study authors compared three models, which included model 1 (age and sex), model 2 (age, sex, and biomarker positivity (troponin positivity or d-dimer positivity), and model 3 (model 2 plus deep learning).

The deep learning model demonstrated an 83 percent area under the curve (AUC) after fine-tuning in the external test set for aortic dissection and all-cause mortality in comparison to model 2’s 66 percent AUC for aortic dissection and 68 percent AUC for all-cause mortality, according to the study. The researchers noted the deep learning model had a more modest AUC for PE (69 percent), which was eight percent higher than the AUC for model 2 (61 percent). For ACS, the study authors noted comparable AUCs for the deep learning model (86 percent) and model 2 (87 percent).

In contrast to commonly used chest pain scores that provide an estimate for ACS, the study authors pointed out that the deep learning model offers a more comprehensive risk assessment with its ability to consider risks for ACS, aortic dissection, and PE.

“Compared with other chest radiography-based DL models, which identify specific abnormalities, our model provides individualized risk assessment for ACP syndrome, which may help deferral of additional cardiovascular or pulmonary tests in up to four to eight times as many individuals, potentially aiding rapid triage of patients in the (emergency department),” wrote Michael T. Lu, M.D., M.P.H., the director of AI and co-director of the Cardiovascular Imaging Research Center at Massachusettes General Hospital (MGH), and colleagues.

(Editor’s note: For related articles, see “Deep Learning Model Predicts 10-Year Cardiovascular Disease Risk from Chest X-Rays” and “New AI Modality Gets FDA Nod for Detecting Aortic Atherosclerosis and Ectasia on Chest X-Rays.”)

Employing a sensitivity threshold of 99 percent, the researchers found that the deep learning model indicated deferral of additional pulmonary and cardiovascular testing in 14 percent of patients in contrast to model 2, which indicated deferral of the additional testing in two percent of patients.

In an accompanying editorial, Jin Mo Goo, M.D. Ph.D. noted that chest imaging has increased in emergency rooms in the United States over the past two decades despite “often negative” results for patients presenting with ACP. While acknowledging that deferring additional imaging in low-risk patients is “a clinically and economically crucial issue,” future studies are necessary to develop and refine deep learning applications for clinical use in this patient population, according to Dr. Goo, a professor in the Department of Radiology at the Seoul National University College of Medicine.

In regard to study limitations, the authors noted a potential selection bias with the retrospective data being drawn from electronic health records (EHRs) at university teaching hospitals. The researchers acknowledged the EHR database lacked detail on leading symptoms, initial triage information and patient-level care data. While the patient inclusion criteria emphasized those in whom additional pulmonary or cardiovascular tests could be deferred, the study authors conceded that the patients in the EHR database may have included those without acute chest pain.