Cardiovascular disease (CVD) is a leading global cause of morbidity and mortality and excessive healthcare expenditures. Pulmonary hypertension (PH) represents an insidious and progressive subset of CVD affecting an estimated 1% of the general population, increasing to up to 10% in the population over the age of 65. Recent advancements in artificial intelligence (AI) have shown promise in transforming PH diagnosis by enabling the analysis of complex physiological data. Specifically, AI algorithms applied to electrocardiography (ECG) and phonocardiography (PCG) waveforms captured through novel medical devices, such as smart stethoscopes, have demonstrated potential in detecting PH and other cardiovascular conditions with high sensitivity and specificity.

Despite the promising capabilities of AI algorithms, a significant barrier to their clinical implementation is the lack of high-quality, prospectively collected datasets for validation. Many existing AI algorithms have been trained on retrospective data, which may not capture the variability and complexity of real-world clinical scenarios. This limitation raises concerns about the generalisability and reliability of AI predictions across diverse patient populations.

Therefore, there is a critical need for prospective validation studies to assess the performance of AI algorithms in realworld settings, ensuring their accuracy and applicability before widespread clinical deployment. Imperial College London's Health Impact Lab (Hi Lab) and collaborators continue to develop artificial intelligence (AI) algorithms that use cardiac waveforms to predict cardiovascular disease (CVD), including pulmonary hypertension (PH). The performance of these algorithms requires validation on prospectively collected patient data (waveforms) - where the ground truth for the algorithms under investigation is recorded during routine echocardiography as part of clinical care. This study aims to prospectively collect a large dataset of cardiovascular ECG and PCG data, along with corresponding gold-standard echocardiography findings. This dataset will be used to validate AI algorithms for important CVD, such as pulmonary hypertension enhancing their reliability and clinical applicability.

This prospective observational cohort and validation study aims to create a patient dataset of point-of-care cardiovascular waveforms for validation of Imperial College London artificial intelligence (AI) algorithms trained for the detection of cardiovascular disease (CVD), using said waveforms as input, with a focus on pulmonary hypertension (PH).

The study will recruit 1,000 unselected patients attending Imperial College Healthcare NHS Trust for routine echocardiography. Each patient will undergo a non-invasive examination using a smart stethoscope that records 3- lead electrocardiogram (ECG) and phonocardiogram (PCG) waveforms, in addition to the standard echocardiography parameters. Baseline demographic data and medical history will also be collected, and a chart review will be performed at 24 months to capture any subsequent morbidity or mortality. The study will validate the AI algorithms by comparing their performance to echocardiography results, the current gold standard for CVD diagnosis. The primary outcome measures will include performance characteristics - sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F-1 Score of the AI algorithm in detecting CVD.

Inclusion Criteria:

• Age 18 years or above
• Able to give informed consent
• Attending for echocardiography at Imperial College NHS Trust

Exclusion Criteria:

• Any chest wound, skin pathology or other feature that would prohibit routine stethoscope examination

All-comers to echocardiography (n=1,000) will be approached for participation by a member of the research team i.e. identification by virtue of presence in the Imperial College NHS echocardiography department whilst attending routine care, gold-standard investigation.