Answers are grounded in this paper and cite their sources.
Background: Continuous biosignals contain physiological information that artificial intelligence (AI) may use for adverse-event forecasting, concurrent detection, anesthesia-state classification, and nociception assessment. Objectives: This critical narrative review examined the technical and clinical evidence for AI-driven biosignal analysis in perioperative and critical care settings. Search Strategy: PubMed and Web of Science were searched through August 2026, supplemented by Google Scholar and citation searching. The documented initial selection retained 18 studies from 529 records. One additional eligible study was added, yielding 19 studies for detailed synthesis: 15 clinical patient cohorts, 3 healthy-volunteer experiments, and 1 EEG architectural benchmark. Key Findings: Performance was heterogeneous and not directly comparable across tasks, horizons, prevalences, and validation designs; reported AUROC values ranged from 0.731 to 0.990. A descriptive, PROBAST-informed appraisal of the 18 task-relevant studies identified recurrent methodological concerns, particularly in analysis and validation; applicability was considered separately. Principal limitations included inconsistent outcome definitions, class imbalance, incomplete calibration and clinical-utility assessment, leakage risks, sensor-related bias, and limited prospective evaluation. Conclusions: Translation requires explicit task definitions and prediction horizons, patient-level validation, representative data, appropriate discrimination and calibration assessment, prospective multicenter evaluation, and evidence of clinical utility.
No takes yet. Share an insight, caveat, or question.
Hassine et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: