The SMART-LV Pilot Study

Pilot Evaluation for SMartphone-adaptable Artificial Intelligence for PRediction and DeTection of Left Ventricular Systolic Dysfunction

The goal of this pilot study is to evaluate the prospective performance of an image-based, smartphone-adaptable artificial intelligence electrocardiogram (AI-ECG) strategy to predict and detect left ventricular systolic dysfunction (LVSD) in a real-world setting.

The SMART-LV pilot study will be a prospective cohort study in outpatient clinics at the Yale New Haven Hospital. Participants who have undergone a 12-lead electrocardiogram (ECGs) with either a high (≥80%) or low (<10%) probability of LVSD on AI-ECG algorithm, but without an echocardiogram done in the clinical setting for at least 90 days after the ECG, will be identified by electronic health record (EHR) and invited for a limited echocardiogram/cardiac ultrasonogram for assessing LV ejection fraction. The goal of the study is to evaluate the feasibility of recruiting patients and performing the study after pursuing a screening on 12-lead ECGs. The procedure currently used for detection of LVSD, echocardiograms, are inaccessible and expensive. Therefore, while AI-ECG-based algorithms using a smartphone- or web-based application can broaden access to screening, a thorough evaluation for this indication is needed before clinical adoption. The investigators intend to use the results as pilot data for sample size and drop-off rate estimation for a subsequent larger prospective cohort study aimed at validating the performance characteristics of the model in a screening setting. The validation of this accessible ECG-based screening strategy, that can be directly used by clinicians using a smartphone or web-based application, can transform the early identification of LVSD before the development of symptoms, thereby allowing broader utilization of evidence-based therapies to prevent symptomatic heart failure and premature death.

In the ECG repository of Yale New Haven Hospital, all patients undergoing a 12-lead screen in an outpatient setting, from whom 20 individuals, 10 each with high and low predicted probability of LVSD, will be invited for a limited echocardiogram to definitively evaluate for LVSD. The investigators will assess whether the AI-ECG model continues to have the reported discrimination and sensitivity of >90% for LVSD diagnosis in a screening setting in outpatient routine clinical care.

A novel AI-ECG model developed at the Cardiovascular Data Science (CarDS) lab will be used as Software as Medical Device (SaMD) on ECG images for detection of LVSD.The AI-ECG model will be used on all participants undergoing a 12-lead ECG.

A novel AI-ECG model developed at the Cardiovascular Data Science (CarDS) lab will be used as Software as Medical Device (SaMD) on ECG images for detection of LVSD.

Device feasibility of AI-ECG will be evaluated by comparing the proportion of patients with LVSD on echocardiography among those with a high predicted probability of LVSD on an AI-ECG screen compared with the proportion of patients with LVSD on echocardiography in those with a negative AI-ECG screen. Higher proportions indicate successful detection of asymptomatic LVSD compared with routine clinical care.

Inclusion Criteria: Provision of signed and dated informed consent form. Stated willingness to comply with all study procedures and availability for the duration of the study Exclusion Criteria: Patients who have undergone a prior echocardiogram. Patients with a prior diagnosis of left ventricular dysfunction, based on a documented low ejection fraction (EF) in the medical record. Patients with an intermediate predicted probability of low EF (10 to 80%) Patients with a prior diagnosis of heart failure as determined by International Classification of Diseases-10 diagnosis code for heart failure. Research opt-out patients