Arrhythmias & electrophysiology
Clinical Decision-Making of Artificial Intelligence vs Medical Professionals in Patients With Syncope
Authors' conclusion
GPT-4o proposed a diagnosis in all cases; however, with a low DPS and is not yet suitable for unsupervised clinical use interpreting referral letters.
Study context
- Population
- GP referral letters to a syncope unit1,045 assessments
- Participants
- 55 participants
- Comparison
- GPT-4o vs medical professionals
- Physiciansn = 12
- Allied professionalsn = 6
- Outcome
- Diagnostic performance (final diagnosis present in initial differential diagnosis) and Diagnostic Precision Score (DPS)
Methodological detailsthree-phase, adjudicated final diagnosis, 1.5-year follow-up
- Three-phase
- Adjudicated final diagnosis
- 1.5-year follow-up
Performance
As reported in the abstract
Diagnostic performance (final diagnosis in initial differential)
| Test or model | Value | Diagnostic Precision Score (DPS) |
|---|---|---|
| GPT-4o | 75.9% | −6.9% (−4.00/54) |
| Physicians | 48.6% | 22.9% (148.75/648) |
| Allied professionals | 36.7% | 12.6% (40.75/324) |
Value
GPT-4o75.9%
Physicians48.6%
Allied professionals36.7%
Diagnostic Precision Score (DPS)
GPT-4o−6.9% (−4.00/54)
Physicians22.9% (148.75/648)
Allied professionals12.6% (40.75/324)
Further findings
Diagnostic yield (any suggested diagnosis)
- Lifestyle measures suggested (GPT-4o, not medical professionals (MPs))
- 29/55; 52.7% · Counterpressure maneuvers · Increased fluid intake 28/55; 50.9%
Safety
- Cardiac diagnoses mislabelled as reflex syncope by GPT-4o
- 3 of 4 · 3 events
Original publication
Clinical Decision-Making of Artificial Intelligence vs Medical Professionals in Patients With Syncope. JACC Adv. Published 19 December 2025. doi:10.
Summarised from the published abstract; every number is checked against it