Tempus AI, Inc. (NASDAQ: TEM) has received FDA 510(k) clearance for ECG-PH, artificial-intelligence software designed to analyze a standard resting 12-lead electrocardiogram for signs associated with pulmonary hypertension. The Class II device is intended for ECGs recorded in healthcare facilities from patients aged 40 years or older who have cardiovascular symptoms such as dyspnea, fatigue, chest pain or edema but no known history of pulmonary hypertension. The algorithm looks for patterns associated with mean pulmonary artery pressure above 20 mmHg and returns a binary output intended to support referral or further diagnostic evaluation rather than diagnose pulmonary hypertension by itself.
The August 24 clearance makes ECG-PH Tempus’ third FDA-cleared cardiovascular AI product after ECG-AF and ECG-Low EF. Its practical proposition is particularly interesting because the required input is not a new imaging examination, laboratory assay or wearable device. Hospitals already perform enormous numbers of 12-lead ECGs, creating the possibility that the same recording obtained for another clinical reason can carry an additional layer of information about a disease that often presents through nonspecific symptoms.
Why is pulmonary hypertension difficult enough to justify an AI screening layer?
Pulmonary hypertension is an umbrella term covering several disease mechanisms that produce abnormally elevated pressure in the pulmonary circulation. Symptoms including breathlessness, fatigue, chest discomfort and peripheral edema overlap with far more common cardiopulmonary disorders, meaning the disease can remain unrecognized until substantial right-heart strain develops.
Echocardiography can raise suspicion, but definitive hemodynamic characterization generally requires right-heart catheterization. That diagnostic pathway is appropriate when clinical suspicion is sufficiently strong, but it is not practical to send every patient with unexplained dyspnea directly to an invasive procedure.
FDA’s classification for pulmonary-hypertension machine-learning notification software explicitly describes the category as using noninvasive physiological information to suggest the likelihood of PH for referral or diagnostic follow-up. That framing is important because the regulatory purpose is triage, not replacing invasive hemodynamic confirmation.
A useful ECG algorithm could therefore intervene earlier in the pathway. It does not need to tell a physician exactly what type of PH a patient has. It needs to surface enough previously unrecognized high-risk patients that subsequent echocardiography, specialist evaluation or right-heart catheterization occurs sooner.
What exactly can ECG-PH tell a clinician?
The cleared device analyzes resting, non-ambulatory 12-lead ECG recordings and produces a binary result associated with elevated mean pulmonary artery pressure above 20 mmHg. Tempus specifies that the software is intended for symptomatic patients aged 40 and older without a known PH history. It should not be used on paced ECG rhythms, should not be used for serial monitoring and is not intended as a stand-alone diagnostic.
That is a deliberately narrow claim. ECG-PH does not report pulmonary artery pressure in millimeters of mercury, identify the WHO pulmonary hypertension group or decide whether a patient should receive a particular therapy. The output must be interpreted with the underlying ECG, symptoms, clinical history and other diagnostic information.
The distinction protects against one of the common ways medical AI can be overinterpreted. A machine-learning model can identify patterns correlated with disease that are too subtle or distributed for conventional ECG interpretation, but the statistical signal is not equivalent to directly measuring pulmonary hemodynamics.
How can pulmonary vascular disease leave a signature on an ECG?
As pulmonary vascular resistance and pressure rise, the right ventricle has to work harder to pump blood through the lungs. Over time this can produce changes in right-heart structure, conduction and electrical depolarization or repolarization that influence the ECG.
Traditional clinicians already recognize findings such as right-axis deviation, right ventricular hypertrophy and right-heart strain patterns. Machine learning can potentially exploit a much larger set of waveform relationships simultaneously, including combinations that do not correspond neatly to one familiar human-readable ECG sign.
The advantage is not necessarily discovering an entirely new electrical phenomenon. It can be combining many weak signals into a probability estimate that becomes meaningful when processed across a large training dataset.
That type of model is particularly suited to opportunistic screening because the ECG has already been obtained. The incremental diagnostic layer is software rather than another physical procedure.
Why does Tempus still need the MOMENTOUS trial after receiving FDA clearance?
Regulatory clearance and clinical utility answer different questions. FDA clearance allows ECG-PH to be marketed according to its authorized use. It does not prove that showing the AI result to clinicians increases PH diagnosis, accelerates appropriate treatment or improves outcomes.
Tempus is therefore conducting MOMENTOUS, a prospective, multisite randomized study in people with interstitial lung disease. Participants undergo an ECG analyzed by the PH algorithm, and the study evaluates whether returning the AI risk information to clinicians increases appropriate diagnostic evaluation and eventual pulmonary-hypertension detection compared with standard care. The study remains recruiting.
ILD is a particularly relevant test environment because pulmonary hypertension can complicate interstitial lung disease and substantially change prognosis and treatment planning, yet symptoms of worsening PH can resemble progression of the underlying lung disease. The AI therefore has an opportunity to answer a meaningful clinical question rather than simply demonstrate another high area-under-the-curve number.
A positive MOMENTOUS result could strengthen the case for embedding ECG-PH automatically into health-system workflows. A neutral result would show that an accurate prediction model does not necessarily change physician behavior or diagnostic yield once introduced into real practice.
Why could integration matter more commercially than the algorithm itself?
Diagnostic AI creates little value when physicians have to remember to open a separate application, upload a file and manually retrieve a result. Its commercial value rises when analysis can occur automatically after a standard ECG enters the clinical system and the result reaches the right clinician without materially increasing workflow burden.
Tempus has been building a cardiovascular AI portfolio around exactly that approach. ECG-AF assesses atrial-fibrillation risk, ECG-Low EF identifies patterns associated with reduced ejection fraction, and ECG-PH extends the concept into pulmonary vascular disease.
The portfolio model could allow one routine ECG to support several independent screening functions in the future, although every algorithm requires its own intended-use constraints and evidence. That creates potentially powerful health-system economics because the marginal cost of extracting another signal from an already digitized waveform is fundamentally different from purchasing another imaging system.
Could an AI-positive ECG lead to unnecessary invasive testing?
It could if clinicians interpret the output without sufficient clinical context. No screening or triage algorithm has perfect specificity, meaning some patients flagged as higher risk will not ultimately have pulmonary hypertension.
The cleared indication therefore places the result inside a broader clinical assessment. Physicians may use echocardiography, pulmonary testing, imaging, laboratory measurements and specialist evaluation before deciding that right-heart catheterization is appropriate.
The value equation consequently depends on both sensitivity and downstream efficiency. Missing fewer PH patients is desirable, but an algorithm that triggers very large numbers of unnecessary workups can add cost and anxiety while overwhelming specialist services.
Prospective clinical-utility studies can help quantify that trade-off more meaningfully than retrospective model validation alone.
What does the third cardiovascular clearance say about Tempus’ broader AI strategy?
Tempus is moving beyond precision oncology into a model where routinely collected clinical data can be reinterpreted algorithmically for diseases not obvious from the original test. ECG-PH is a strong example because the electrical recording remains exactly the same while the software changes what information clinicians can potentially extract from it.
The technology is also a reminder that AI does not need to replace a physician or autonomous diagnostic system to become commercially important. A binary risk flag inserted at the right point in a diagnostic pathway can be valuable if it moves the correct patient toward definitive testing earlier.
FDA clearance settles whether Tempus can market ECG-PH for that notification function. MOMENTOUS and subsequent real-world deployment will answer the more economically important question: whether one additional line generated from a routine ECG changes enough clinical decisions to close a meaningful part of the pulmonary-hypertension diagnostic gap.
