Tempus AI, Inc. has secured U.S. Food and Drug Administration 510(k) clearance for Tempus ECG-PH, an artificial-intelligence software device designed to analyze standard 12-lead electrocardiograms and identify signals associated with pulmonary hypertension. The software is intended for patients aged 40 years and older with cardiovascular symptoms such as shortness of breath, fatigue, chest pain or edema who do not already have a pulmonary-hypertension diagnosis.
The device does not diagnose pulmonary hypertension independently. Instead, it analyzes an ECG and returns a binary output indicating whether features associated with elevated mean pulmonary artery pressure above 20 mmHg are present, potentially prompting physicians to consider additional diagnostic evaluation.
That distinction is crucial because pulmonary hypertension remains difficult to identify early. Symptoms are often nonspecific, echocardiography is commonly used during evaluation, and definitive hemodynamic diagnosis frequently relies on right-heart catheterization. Tempus AI is effectively trying to turn a ubiquitous, inexpensive and non-invasive test into an additional screening layer for deciding who deserves closer investigation.
How can artificial intelligence find pulmonary hypertension in an ECG?
An electrocardiogram records electrical activity generated as the heart contracts and relaxes. Physicians can already recognize certain abnormalities related to right-heart strain, rhythm disturbances and chamber changes, but subtle combinations of waveform characteristics may contain information that is difficult to interpret consistently by eye.
Machine-learning systems can evaluate a much larger number of relationships across those waveforms. Rather than looking for one classical abnormality, an algorithm can identify combinations of patterns statistically associated with elevated pulmonary pressures in the data on which it was developed and validated.
Tempus ECG-PH therefore does not directly measure pulmonary artery pressure. It infers whether the ECG contains a signal associated with pressure elevation and produces an output intended to support referral or further diagnostic workup.
The FDA classified the product as pulmonary-hypertension machine-learning-based notification software under 510(k) K253699 and determined it substantially equivalent on August 21, 2026.
Why could pulmonary hypertension benefit from an AI screening layer?
Pulmonary hypertension encompasses conditions in which pressure inside the pulmonary circulation becomes abnormally elevated. Narrowing or stiffening of pulmonary blood vessels can progressively increase the workload on the right side of the heart, potentially resulting in right-heart dysfunction or failure.
The diagnostic problem is that early symptoms overlap with numerous more common conditions. Dyspnea, fatigue, chest discomfort and swelling do not automatically point clinicians toward pulmonary hypertension, meaning patients can move through repeated evaluations before the condition is considered.
Tempus AI estimates that pulmonary hypertension affects roughly 1% of the overall population and as many as 10% of adults older than 65. Against that backdrop, a software layer capable of examining ECGs that are already being obtained for symptomatic patients could help surface cases that might otherwise remain lower on a physician’s differential diagnosis.
The economic logic is equally important. Healthcare systems do not need a specialized scanner or new laboratory assay for every initial analysis if the software can operate on existing digital ECG recordings. That creates the possibility of scaling algorithmic screening through infrastructure already present in hospitals and clinics.
What are the limits of Tempus ECG-PH?
FDA clearance does not mean clinicians can use ECG-PH as a replacement for pulmonary-hypertension diagnosis. Tempus AI explicitly states that the system is not intended as a standalone diagnostic device and that its results must be interpreted alongside the original ECG, symptoms, other testing and medical history.
The software is also not intended for serial monitoring and should not be applied to paced rhythms. Its authorized population is restricted to symptomatic adults aged 40 years and older who do not already have pulmonary hypertension.
These restrictions illustrate an important principle in medical artificial intelligence: the value of an algorithm can depend as much on where it sits in the clinical pathway as on its headline accuracy. A notification tool can be useful if it directs appropriate patients toward echocardiography or specialist evaluation, yet problematic if clinicians begin treating its output as definitive.
How does ECG-PH fit into Tempus AI’s cardiovascular strategy?
ECG-PH is the third FDA-cleared product in Tempus AI’s ECG artificial-intelligence portfolio. The company previously secured clearance for ECG-AF, which analyzes electrocardiograms for signals associated with the risk of atrial fibrillation within the following 12 months, and ECG-Low EF, which looks for signs associated with reduced left-ventricular ejection fraction.
The portfolio reveals a strategy that goes beyond automating traditional ECG interpretation. Instead of using artificial intelligence merely to label rhythm abnormalities that a cardiologist could already see, Tempus AI is attempting to extract latent clinical information that conventional reading does not routinely provide.
Pulmonary hypertension is especially illustrative because the ultimate disease measurement, pulmonary artery pressure, comes from an entirely different physiological domain. If an ECG algorithm can reliably enrich the population referred for more definitive testing, the humble ECG becomes something closer to a reusable digital biomarker platform.
Are ordinary diagnostic tests becoming AI platforms?
That possibility may be the most important implication of the clearance. Hospitals already generate enormous volumes of electrocardiograms, chest X-rays, CT scans, retinal images and pathology slides. Historically, each test had a relatively fixed clinical purpose, but artificial intelligence can potentially extract additional information from the same underlying data.
An ECG collected because a patient reports chest discomfort could therefore support rhythm interpretation, atrial-fibrillation risk analysis, low-ejection-fraction detection and pulmonary-hypertension screening without requiring four separate acquisition procedures. The challenge becomes determining which algorithms should run on which patients and how multiple outputs should be presented without overwhelming clinicians.
Tempus AI’s expanding cardiovascular portfolio is an early example of this transformation. The eventual competitive advantage may not belong solely to the company with the best single algorithm, but to platforms capable of integrating many validated algorithms into clinical workflows while directing each output toward an actionable next step.
