Can Artificial Intelligence Enhance Syncope Management?

Syncope, a form of transient loss of consciousness, remains a complex medical condition for which adverse cardiovascular outcomes, including death, are of major concern but rarely occur. Current risk stratification algorithms have not completely delineated which patients benefit from hospitalization and specific interventions. Patients are often admitted unnecessarily and at high cost. Artificial intelligence (AI) and machine learning may help define the transient loss of consciousness event, diagnose the cause, assess short- and long-term risks, predict recurrence, and determine need for hospitalization and therapeutic intervention; however, several challenges remain, including medicolegal and ethical concerns. This collaborative statement, from a multidisciplinary group of clinicians, investigators, and scientists, focuses on the potential role of AI in syncope management with a goal to inspire creation of AI-derived clinical decision support tools that may improve patient outcomes, streamline diagnostics, and reduce health-care costs.

Names and affiliations of the clinicians and investigators from around the world who participated in a virtual meeting on September 25, 2021, that highlighted the potential application of artificial intelligence to the management of syncope and inspired this manuscript.Among them were 25 syncope experts from 12 countries across 4 continents.A small few were not able to attend the meeting but contributed written statements on the subject of artificial intelligence and syncope.
Supplemental Table 2. Select Viewpoints on AI a from an International Panel of Syncope Experts

Question Expert Viewpoint
Should we be afraid of AI? Dr. Mauro Gatti, Italy: We should not be afraid of AI.It has been around since the fifties and human beings are still here.The key point is to study "the human in the loop," namely, the interaction between the clinician and the AI tool.AI can be used for automation or augmentation; medicine is no exception.It may be reasonably expected that automation will ultimately be applied to works that are simple for humans, while augmentation will be applied to more complex activities.
Several aspects of current AI models may make them more suitable for augmentation than automation, notably robustness, interpretability, and accountability.
The use of an AI model for augmentation has its own specific challenges.The two most critical requirements of a clinical decision support system to enable effective utilization are: Establishing whether TLOC is syncope, stratifying its risk and defining the etiology to assure the best treatment remains a great challenge.
Having time for long consultations with a team of dedicated multidisciplinary experts may be supplanted by AI/ML.However, patients with syncope need to be heard and physically examined, which can be time-consuming.Our experience in a dedicated syncope unit over the last 30 years, with a team of committed professionals, has shown that the more you practice, the more time you spend, and the more diverse the team is, the better the outcome for the patient.Further, public medical services, especially in less developed countries, have very poor conditions of care and this model is difficult to reproduce. 33AI is a promising tool to accelerate the process of algorithmic application to normalize clinical care worldwide, reducing diagnostic errors and improving patient outcomes.
At a time when the quality of doctor-patient relationships are assessed, the lack of physical examination is criticized, and a more humanized approach is requested.On the contrary, the importance of the computer in medicine and public health is increasing.Not only because of the accelerated growth of medical knowledge, but especially the development of technologies to support clinical decisions, which should be used cautiously to provide more diagnostic accuracy. 34tegrative medicine has benefited greatly from data storage, which has increased exponentially in recent years, creating the concept of big data.Supercomputers can create true neural networks and data processing algorithms in several areas, providing increasingly accurate diagnostic hypotheses. 34,35This data can be reviewed by specialists all over the world, with unprecedented advantages for health services and patient care.
TLOC and syncope are among the symptoms that would most benefit from this new tool, considering the vast possibilities of diagnosis and the need of faster and more accurate solutions. 36 will make it possible to spread the knowledge we have learned over the years and enable the development of new routes for better patient care.
How can AI/ML make the right clinical decision?
Dr. Fabrizio Ricci, Italy: Leveraging unsupervised ML-based cluster analysis and phenomapping techniques 37,38 might help recognize covert disease heterogeneity in patients with unexplained syncope and identify subsets -phenogroups -with distinct pathophysiological profiles and differential outcomes.Clustering is the most common unsupervised learning technique for exploratory data analysis to find hidden patterns or groupings in data according to similarity. 39P h is a high-throughput technology that can be leveraged to automatically extract, classify, and label elements from massive volumes of raw, unstructured, and unlabeled datasets.Large-scale NLP-driven analysis of unstructured EMR in combination with AI-powered deep phenotyping techniques would be a desirable approach to identify unique patterns of association among phenotypic variables at scale and discover subgroups of patients with clinically distinct phenotypic and prognostic profiles.Ultimately, whether the identification of distinct phenogroups h NLP, natural language processing can impact clinical decision making or deliver true patient benefit will have to be critically appraised. 38,40at are the future directions of personalized medicine and AI ?
Dr. Franca Dipaola, Italy: In the age of precision medicine, clinicians' decision-making is asked to rely upon personalized risk predictions.For syncope, risk stratification tools based on traditional statistical methods failed to be superior to simple clinical judgment, 41,42 thus requiring the use of alternative methodologies 36 and innovative technologies. 27e possibility to automatically classify and analyze risk factors with many different diagnoses and with no gold standard every one of us would agree on, we're going to have difficulties.We need that gold standard.How to get that is a major challenge.We have to be realistic in our limitations in AI development without a solid gold standard.
We need to develop detailed smart questionnaires and then test them to ascertain the predictive accuracy for various diagnoses.This is a complex task.Perhaps weaning out the most common diagnostic categories would be a feasible first step.Accomplishing this will necessitate time and substantial funding.Perhaps an NIH i grant should be embarked upon.
Dr. Tamara Lyubimtseva, Russia: To date, for this team, it is necessary to collect big clinical material by way of a special cloud with mathematical data processing.
How should we approach data acquisition?
Dr. Robert Sheldon, Canada: What are some unique applications for AI in syncope?
Additional solutions from the University of Iowa multidisciplinary team: Dr. Milena Gebska, USA: One potential solution may be electronic "syncope check-in booths," involving a custom set of questions created based on supervised learning experiences that the patient, family member, or witness would be asked to complete in the ED.This could potentially help uncover "the truth," and streamline the initial triage, evaluation, and assessment of these patients while simultaneously contributing to an augmented intelligence platform.
Avinash Mudireddy, USA: AI-derived data collection processes could be created via speech, vision, or text.They could be refined over time through re-training on new data, leading to highly developed syncopespecific AI.
This table contains select written and quoted viewpoints from the international panel of syncope experts in regard to specific objectives and opportunities for AI in syncope research.

Supplemental Figure 1 .
An International Panel of Syncope Experts: World Map A world map highlighting countries represented at the virtual meeting on September 25, 2021, which aimed to discuss the application of artificial intelligence in the field of syncope.The international panel included a total of 39 clinicians and investigators: including 25 syncope experts from 12 countries around the globe.1. Emergency Medicine Research Group Edinburgh (EMERGE), Department of Emergency Medicine, Royal Infirmary of Edinburgh, Edinburgh, UK. 2. Acute Care Group, Usher Institute of Population Health Sciences and Informatics, College of Medicine and Veterinary Medicine, University of Edinburgh, Edinburgh, UK Jayaprakash Shenthar MBBS, MD Sri Jayadeva Institute of Cardiovascular Sciences and Research, Bangalore, Karnataka, India South America Denise Hachul, MD, PhD Heart Institute, University of Sao Paulo Medical School, Sao Paulo, Brazil from large patient populations by ML algorithms could generate significant progress in individual risk stratification.In particular, since most of the prognostic information is available in textual form, the use of NLP techniques could allow the automatic extraction of relevant phenotypes from EMR or ad hoc repositories.Implementation of these techniques, through data sharing, can promote the optimization of syncope patients' diagnostic work up and personalization of care. 27What are the most important first steps in developing an AIbased syncope project?Dr. Benditt, USA: The use of AI is so critically dependent on the training set.And the training set has to be as good as it can possibly be.The difficulty we have is we don't have a training set where all 20 of us would all agree on the diagnosis.These datasets have all the limitations of the most component of us as syncope people.Consequently, AI can never be better than that because the training set is limited by our own ability to know the diagnosis.We can use it for individual things like identifying the vasovagal faint.But in the complex world of syncope,

# 1 :# 2 : 3 :# 4 :
Data acquisition and analysis efforts should focus on cause of syncope as well as risks incurred from syncope and independent comorbidities.The first tier of effort should target common and problematic syncope mimics: convulsive syncope versus epileptic convulsions, falls versus faints in the older patient, or collapse without apparent biological cause.iNIH, National Institute of Health #The second tier of effort should target risk stratification.This should start with agreement on whether the primary outcome should be related to the cause of syncope, or simply be a composite of poor outcomes due to the range of comorbidity factors and frailty.This effort should include sufficient power to be statistically robust.Scrupulous care must be taken to collect patients with gold standard diagnoses to initiate a valid AI assessment.This will require a large prospectively collected population and not administrative data.#6: AI/ML modelling should be done in sequential tiers or using a prior stratified patient populations of adequate size.How can we apply AI to the ECG in syncope evaluation?Dr. Artur Fedorowski, Sweden: How AI-ECG interpretation is used depends on the clinical setting-in a busy ED where time is short and risks should be minimized, in primary care where physicians deal with an array of medical conditions and may lack an experienced eye, or in cardiology where a syncope specialist may need a fine-tuned ECG assessment to detect an underlying channelopathy or conduction disorder.Each clinical context and its available resources would dictate the use of this technology.