Acute cardiac care & resuscitationCross-sectional study
An Artificial Intelligence–Enabled Cardiopulmonary Resuscitation Instructor
- cross-sectional study
- simulated emergency scenarios
- early proof-of-concept study design
- standard test-retest methods
Authors' conclusion
These findings suggest that AI-enabled CPR instruction shows promise for supporting bystanders. Further validation in diverse, general population settings is warranted to define the role of AI-based CPR instruction as a scalable public health intervention for OHCAs.
Study context
- Population
- Simulated out-of-hospital cardiac arrest scenarios and 911 dispatcher-assisted calls where cardiopulmonary resuscitation (CPR) was indicated
- Exposure
- Adherence to guideline-derived minimally viable and maximally effective CPR instruction criteria: widely available artificial intelligence (AI) models and the purpose-built AI CPR instructor agent in simulated scenarios; the agent vs dispatchers in 911 calls
- Widely available AI modelsn = 6
Main finding
As reported in the abstract
Adherence to minimally viable criteria, simulated OHCA scenarios
| Group | Value | Gemini | Grok and Claude |
|---|---|---|---|
| Widely available AI models | 89.7% (95% CI 84.8%–93.2%) | 79.4% (95% CI, 63.2%-89.7%) | 97.1% (95% CI, 85.1%-99.5%) |
| ChatCPR (instructor agent) | 100% (95% CI 89.8%–100%) |
Observational data: associations, not effects.
Further findings
| Outcome | Widely available AI models | ChatCPR (instructor agent) | Dispatchers | absolute improvement (percentage points) (95% CI) | P value |
|---|---|---|---|---|---|
| Adherence to maximally effective criteria, simulated OHCA scenarios | mean 69.8% (95% CI 65.5%–73.7%) | 100% (95% CI 95.4%–100%) | |||
| Llama | 61.3% (95% CI, 50.3%-71.2%) | Not reported in the abstract | Not reported in the abstract | ||
| GPT-4o | 75.0% (95% CI, 64.5%-83.2%) | Not reported in the abstract | Not reported in the abstract | ||
| Adherence to minimally viable criteria, retest using 911 calls · improvement over dispatchers | Not reported in the abstract | 100% (95% CI 88.6%–100%) | 84.5% (95% CI 73.1%–91.6%) | 15.5 (2.1–27.4) | P = 0.02 |
| Adherence to maximally effective criteria, retest using 911 calls · improvement over dispatchers | Not reported in the abstract | 98.9% (95% CI 94.0%–99.8%) | 62.8% (95% CI 55.0%–70.0%) | 36.1 (27.2–44.3) | P < 0.001 |
Original publication
An Artificial Intelligence–Enabled Cardiopulmonary Resuscitation Instructor. JAMA Intern Med. Published 1 July 2026. doi:10.1001/jamainternmed.2026.1552
Summarised from the published abstract; every number is checked against it