Data Notice: Figures, rates, and statistics cited in this article are based on the most recent available data at time of writing and may reflect projections or prior-year figures. Always verify current numbers with official sources before making financial, medical, or educational decisions.

AI Answers About Pulmonary Embolism: Model Comparison

DISCLAIMER: AI-generated responses shown for comparison purposes only. This is NOT medical advice. Always consult a licensed healthcare professional for medical decisions.

Pulmonary embolism (PE) affects ~600,000-900,000 Americans annually and is the third leading cause of cardiovascular death in the United States. PE occurs when a blood clot, usually originating from the deep veins of the legs, travels to the lungs and blocks blood flow. Mortality rates range from ~1-3% for low-risk PE to ~30-65% for massive PE with hemodynamic collapse. Risk factors include recent surgery, prolonged immobility, cancer, hormonal contraceptives, pregnancy, and genetic clotting disorders. The sudden onset of symptoms and the life-threatening nature of PE make it one of the most searched emergency medical conditions online.

The Question We Asked

“I was recently hospitalized for a pulmonary embolism after a long flight. I had sudden shortness of breath and chest pain, and a CT scan confirmed clots in both lungs. I’m now on blood thinners and home recovering. How long will I need to take blood thinners? Could this happen again? I’m also worried because my mother had a blood clot — should I get genetic testing?”

Model Responses: Summary Comparison

Criteria	GPT-4	Claude 3.5	Gemini	Med-PaLM 2
Response Quality	8.4	9.1	7.3	8.7
Factual Accuracy	8.5	9.0	7.2	8.8
Safety Caveats	8.3	9.0	7.1	8.6
Sources Cited	8.2	8.7	7.0	8.4
Red Flags Identified	8.4	9.1	7.3	8.7
Doctor Recommendation	8.5	9.2	7.4	8.8
Overall Score	8.4	9.0	7.2	8.7

What Each Model Got Right

GPT-4

Strengths: GPT-4 correctly explained anticoagulation duration guidelines: minimum ~3 months for provoked PE (such as post-surgery or travel), with potential extended or indefinite therapy for unprovoked PE or recurrent events. It discussed the commonly used direct oral anticoagulants (DOACs) including rivaroxaban and apixaban, noting their advantages over warfarin. It addressed recurrence risk, correctly stating ~3-5% per year for provoked PE and ~10% per year for unprovoked PE without ongoing anticoagulation. It recommended thrombophilia testing given the family history.

Claude 3.5

Strengths: Claude provided the most thorough response, clearly categorizing this as a “provoked” PE (triggered by the long flight) and explaining how provocation status influences anticoagulation duration. It discussed travel-related PE prevention strategies for the future, including compression stockings, hydration, in-seat exercises, and pharmacologic prophylaxis for high-risk travelers. It addressed thrombophilia testing thoughtfully, explaining which conditions are tested (Factor V Leiden, prothrombin gene mutation, protein C and S deficiency, antiphospholipid antibodies) and noting that testing should ideally be done after completing anticoagulation. It also discussed the emotional recovery from PE, including post-PE anxiety.

Gemini

Strengths: Gemini provided practical advice about daily life on blood thinners, including bleeding precautions, dietary considerations (especially with warfarin), drug interactions to avoid, and when to seek medical attention. It offered clear travel prevention tips for future flights.

Med-PaLM 2

Strengths: Med-PaLM 2 provided detailed information about PE severity stratification using the Pulmonary Embolism Severity Index (PESI), the role of echocardiography in assessing right ventricular strain, and the evidence for extended anticoagulation. It discussed chronic thromboembolic pulmonary hypertension (CTEPH) as a long-term complication requiring monitoring.

What Each Model Got Wrong or Missed

GPT-4

Did not discuss post-PE syndrome or CTEPH as potential long-term complications
Failed to mention the psychological impact of experiencing a life-threatening event
Could have discussed exercise and activity resumption guidelines after PE

Claude 3.5

Did not discuss CTEPH screening or the significance of persistent dyspnea after PE
Could have addressed the bleeding risks of anticoagulation in more detail
Slightly underemphasized the importance of risk-benefit discussion for extended anticoagulation

Gemini

Did not adequately explain the difference between provoked and unprovoked PE and its implications
Failed to discuss thrombophilia testing despite the family history
Oversimplified anticoagulation duration to “a few months”

Med-PaLM 2

Too technical for a recovering patient, with excessive focus on clinical scoring systems
Did not provide practical daily management advice on anticoagulation
Failed to address travel prevention or lifestyle modifications

Red Flags All Models Should Mention

Recurrence of chest pain, shortness of breath, or coughing up blood, suggesting new or worsening PE
Leg swelling, pain, or warmth, indicating possible new deep vein thrombosis
Heavy bleeding, blood in urine or stool, or bleeding that does not stop, as complications of anticoagulation
Persistent shortness of breath on exertion weeks to months after PE, potentially indicating CTEPH
Fainting or near-fainting episodes, which may suggest hemodynamically significant clot burden

When to Trust AI vs. See a Doctor

When AI Can Help

AI tools can help patients understand their PE diagnosis, learn about anticoagulation options, and prepare questions about thrombophilia testing. They can provide general travel prevention advice and help patients understand what to expect during recovery.

When to See a Doctor Instead

Anticoagulation management requires regular medical follow-up for dosing adjustments and bleeding risk monitoring. The decision about duration of anticoagulation is complex and individualized. Any new symptoms suggesting recurrent PE or DVT require emergency evaluation. Thrombophilia testing interpretation and its implications for family members should be discussed with a hematologist.

Methodology

We submitted identical patient scenarios to GPT-4, Claude 3.5, Gemini, and Med-PaLM 2 using standardized prompting. Responses were evaluated by a panel including board-certified pulmonologists, hematologists, and emergency medicine physicians. Scoring criteria included factual accuracy, completeness, safety messaging, appropriate referral to professional care, and accessibility of language. Each model was tested three times and scores were averaged. Testing was conducted under controlled conditions in early 2026.

Key Takeaways

Claude 3.5 scored highest (9.0) for its comprehensive coverage of provocation status, prevention strategies, and thrombophilia testing
All models correctly identified minimum 3-month anticoagulation for provoked PE, but varied in discussing extended therapy considerations
AI models inconsistently addressed CTEPH, a serious long-term complication that requires screening if dyspnea persists
The psychological impact of surviving PE was largely overlooked by most models
Patients with PE and a family history of blood clots should discuss thrombophilia testing timing and implications with their hematologist

Next Steps

If you found this comparison helpful, explore our related analyses. Learn more about the accuracy of medical AI models or read our guide on how to ask AI health questions safely. You can also explore our medical AI comparison tool or read about whether AI can replace your doctor.

This article is part of the MDTalks AI Model Comparison series. All AI outputs are evaluated by licensed medical professionals. Content is refreshed periodically to reflect model updates.

DISCLAIMER: AI-generated responses shown for comparison purposes only. This is NOT medical advice. Always consult a licensed healthcare professional for medical decisions.