← Back to incidents

AI Healthcare Chatbot Misdiagnosed Heart Attack Symptoms as Anxiety, Delaying Critical Care

High

AI healthcare chatbot misdiagnosed heart attack symptoms as anxiety in a 45-year-old woman, causing her to delay emergency care for six hours and suffer preventable cardiac damage.

Category
Medical Error
Industry
Healthcare
Status
Litigation Pending
Date Occurred
Jan 15, 2025
Date Reported
Jan 28, 2025
Jurisdiction
US
AI Provider
Other/Unknown
Application Type
chatbot
Harm Type
physical
Estimated Cost
$150,000
People Affected
1
Human Review in Place
No
Litigation Filed
Yes
Litigation Status
pending
healthcaremisdiagnosisemergency_medicinecardiologysymptom_checkermedical_liabilitybias

Full Description

On January 15, 2025, Sarah Martinez, a 45-year-old teacher from San Jose, California, experienced chest pressure, shortness of breath, and nausea while at home. Having a history of anxiety disorders, she consulted her health insurance company's AI-powered symptom checker app before deciding whether to seek medical care. The chatbot, after asking about her symptoms, medical history, and current stress levels, concluded she was experiencing a panic attack and recommended breathing exercises and contacting her primary care physician during regular hours. Martinez followed the AI's guidance, using relaxation techniques and waiting until morning to call her doctor's office. However, her symptoms persisted and worsened overnight. When she finally reached her physician's nurse the next morning, she was immediately advised to go to the emergency room. Upon arrival at Santa Clara Valley Medical Center, cardiac enzyme tests and an ECG revealed she had suffered a myocardial infarction approximately 6-8 hours earlier. Cardiologist Dr. James Chen, who treated Martinez, stated that the delay in treatment likely contributed to additional heart muscle damage that could have been prevented with immediate intervention. Martinez required emergency cardiac catheterization and stent placement, followed by an extended hospital stay and cardiac rehabilitation. Her medical bills exceeded $150,000, and she faced several weeks of recovery time away from work. The AI symptom checker, developed by HealthAssist Technologies and licensed to major insurance providers, had been trained primarily on general symptom patterns but appeared to have insufficient training data on atypical heart attack presentations, particularly in women under 55. The system's algorithm heavily weighted Martinez's anxiety history and age demographics, leading it to dismiss cardiac symptoms that emergency physicians would have immediately recognized as concerning. HealthAssist Technologies has since stated they are reviewing their diagnostic protocols and considering additional safeguards for potentially life-threatening conditions.

Root Cause

AI symptom checker trained on insufficient cardiac emergency data failed to recognize atypical heart attack presentation in women, instead categorizing chest pressure and shortness of breath as panic attack symptoms based on patient's age and anxiety history.

Mitigation Analysis

A mandatory human clinical review layer for any symptoms flagged as potentially cardiac could have prevented misdiagnosis. Additionally, implementing strict escalation protocols for chest pain symptoms regardless of patient demographics, and training the model on diverse cardiac presentation data including atypical symptoms in women and younger patients would reduce similar errors.

Lessons Learned

This incident highlights the critical need for AI healthcare tools to incorporate comprehensive training on atypical disease presentations and implement mandatory escalation protocols for potentially life-threatening symptoms, regardless of patient demographics or medical history.

Sources