Robotic Chest-Compression Systems with Real-time Physiologic Feedback: A Systematic Review of Closed-Loop and Machine-Learning Approaches

Abstract

Background:
Mechanical chest-compression devices standardize cardiopulmonary resuscitation (CPR), but fixed compression parameters may not match patient physiology. Robotic and algorithm-driven systems that use real-time physiologic signals could enable closed-loop CPR optimization.
Methods:
PubMed was searched from database for clinical, preclinical, and modeling studies evaluating robotic/mechanical chest-compression systems incorporating real-time physiologic feedback (e.g., end-tidal carbon dioxide [ETCO₂], arterial pressure, coronary perfusion pressure [CPP], carotid flow) consistent with closed-loop or machine-learning approaches. Eligible studies were synthesized narratively without meta-analysis.
Results:
Five studies met inclusion: two randomized porcine trials, one porcine machine-learning modeling study (n=7), and two simulation/model studies. A closed-loop machine-controlled CPR system sustained higher CPP at 30 minutes versus guideline CPR (22±3 vs 8±3 mmHg) and preserved carotid blood flow during prolonged resuscitation. An AI-driven CPR robot achieved similar hemodynamics to a standard piston device, with no difference in carotid flow (−23.2±20.2 mL/min; P=0.250) and comparable ROSC (83.3% vs 66.7%; P=1.00). Simulation studies suggested that CPP-targeted controllers improved modeled flow/ETCO₂, and an ML model predicted carotid flow per compression with high accuracy (R²=0.96).
Conclusions:
Evidence for physiologic-feedback robotic CPR is limited to preclinical and simulation studies, but supports technical feasibility and potential hemodynamic advantages over fixed-parameter CPR. Human feasibility trials with standardized ventilation, safety outcomes, and neurologic endpoints are required before clinical deployment.