Updates in Emergency Syncope Research

The year is 2026. A 60-year-old female with a past medical history of smoking, obesity and hypertension presents to the Emergency Department (ED) after losing consciousness while waiting for the bus. She remembers feeling light-headed prior to fainting, but had no chest pain, palpitations, headache or shortness of breath before or after the episode. She ate a normal breakfast this morning and has not changed any of her hypertension medications. There was no head or neck trauma, as she was slowly lowered to the floor by a kind bystander. This has never happened to her before. In the ED, her vital signs are normal, except for mildly elevated blood pressure (142/82) and her physical exam is entirely unremarkable. Her ECG reveals a normal sinus rhythm without acute ischemic changes and normal intervals. You ordered a chest x-ray and blood work, including a high-sensitivity troponin and these have all returned within normal limits. She has never had an echocardiogram performed. She is worried about this episode of syncope and is asking you why she passed out. Does this patient need admission to the hospital for further testing and monitoring?

Challenges in acute syncope care

Syncope continues to be a common reason for patients to present to the ED. As the US population ages, we can expect to see an increased number of patients with this complaint. Finding the exact etiology of the syncopal event remains a challenge for emergency physicians; though patients frequently ask us why they fainted, it is too often a question we cannot answer. There are two primary objectives in the emergency department evaluation of syncope: diagnostic and prognostic. The first objective is to uncover an acute serious diagnosis that may be associated with or have caused, the syncopal event, such as pulmonary embolism or cardiac dysrhythmia. The second objective is to accurately risk-stratify the patient to determine the intensity of care which would be appropriate in the short-term. Additional investigations, such as in-patient cardiac monitoring and echocardiography are often ordered, but are generally low yield. Risk stratification of patients with unexplained syncope has been a focus of clinical research for over two decades and many clinical decision tools have been published. Despite these efforts, no clear winner among these many tools has emerged, though there is ongoing research that may shed light on this issue. Further, the management of patients with syncope in the ED will be the focus of the next GRACE clinical practice guidelines from the Society of Academic Emergency Medicine (SAEM) that will be published in 2025. These guidelines will explicitly focus on, among other topics, the validated risk-stratification tools currently available. Meanwhile, new analytic approaches using artificial intelligence could provide the next frontier in clinical risk-stratification for ED patients with syncope.

Risk-stratification tools still need external validation

A recent scoping review found 19 different risk-stratification tools for patients with syncope.¹ The majority of these had neither been externally validated nor compared with unstructured physician gestalt. Perhaps the most recognizable of these, the San Francisco Syncope Rule, was found to have inadequate sensitivity in subsequent validation studies.^2,3 The more rigorous Canadian Syncope Risk Score has been validated in a multicenter study in Canada, but not has not been studied in the US.⁴ The FAINT Score, which was derived in a large, multicenter cohort of older adults in the US is pending external validation.⁵ Fortunately, there is an ongoing NIHfunded, multicenter study named “PACES: Practical Approaches to Care in Emergency Syncope” which is nearing completion and could significantly advance this field of research. Data collection will be completed in fall 2024. Results from the PACES study will provide external validation for and compare performance of, the Canadian and FAINT scores, as well as comparing both to unstructured physician gestalt.

Potential applications of Articificial Intelligence

Researchers have begun to apply various artificial intelligence (AI) techniques, such as machine learning and deep learning, to clinical riskstratification problems. With a large and accurately labeled dataset, machine learning could theoretically provide highly accurate risk-stratification for patients with unexplained syncope. One key challenge to this is that retrospective clinical data, while plentiful, often lacks complete and accurate information on clinical outcomes post-discharge, while prospective datasets generally do capture follow-up, but lack the scale required to properly train such AI models.

However, there is exciting work emerging that hones in on a more specific clinical question in the management of ED patients with syncope – whether the patient would benefit from an echocardiogram. A deep learning model, named EchoNext, was recently developed to predict the results of an echocardiogram using a 12-lead ECG alone.⁶ This model was trained on over 425,000 ECG-echocardiogram pairs using significant structural heart disease as the primary outcome. It is currently undergoing external validation in a cohort of adult ED patients with unexplained syncope and if validated, could help optimize echocardiography utilization, a test that is often ordered but rarely changes clinical management.

Need for implementation research

Both conventional and AI-based tools will need to be thoughtfully implemented to have a beneficial real-world impact on patients. Some specific challenges are foreseeable. One is acceptance. Physicians are known to have “algorithm aversion”, that is, “the tendency of humans to shy away from using algorithms even when algorithms observably outperform their human counterpart”.⁷ Another is workflow integration. Significant cooperation between clinicians, bioinformaticists and organizational leaders will be necessary to make sure tools built into electronic health record systems can be meaningfully used within the context of actual clinical care. And finally, even if these tools are successfully implemented, we will need properly conducted clinical trials to demonstrate the sustainable and positive impact on patient-oriented outcomes of these new tools.

You apply a well-validated syncope risk-stratification tool to the case and determine that the patient is at low risk (under 2%) for serious adverse events at 30 days. The AI model automatically analyzes her ECG and suggests that she is at very low risk of having clinically significant structural heart disease. After discussing all of the findings in her workup and your overall assessment, the patient feels reassured. You and the patient reached a shared decision to discharge her home to follow-up with her primary care doctor in the next week.

Disclosure

Dr. Probst is currently supported by an R01 grant from the NIH/NHLBI (R01HL149680) and received a one-time research donation from Roche Diagnostics in 2023.

References

• Uit Het Broek LG, Ort BBA, Vermeulen H, Pelgrim T, Vloet LCM, Berben SAA. Risk stratification tools for patients with syncope in emergency medical services and emergency departments: a scoping review. Scand J Trauma Resusc Emerg Med 2023;31:48.
• Thiruganasambandamoorthy V, Hess EP, Alreesi A, Perry JJ, Wells GA, Stiell IG. External validation of the San Francisco Syncope Rule in the Canadian setting. Annals of emergency medicine 2010;55:464-72.
• Sun BC, Mangione CM, Merchant G, et al. External validation of the San Francisco Syncope Rule. Annals of emergency medicine 2007;49:420-7, 7 e1-4.
• Thiruganasambandamoorthy V, Sivilotti MLA, Le Sage N, et al. Multicenter Emergency Department Validation of the Canadian Syncope Risk Score. JAMA internal medicine 2020;180:737-44.
• Probst MA, Gibson T, Weiss RE, et al. Risk Stratification of Older Adults Who Present to the Emergency Department With Syncope: The FAINT Score. Annals of emergency medicine 2020;75:147-58.
• Jing L, Finer J, Hartzel D, et al. Abstract 14647: EchoNext: An ECG-Based Deep Learning Model to Detect Structural Heart Disease. Circulation 2023;148:A14647-A.
• Filiz I, Judek JR, Lorenz M, Spiwoks M. The extent of algorithm aversion in decision-making situations with varying gravity. PloS one 2023;18:e0278751.