AI models show promise for personalized critical care, but clinical adoption lags
Researchers reviewed 183 computational models designed to guide treatment decisions for hemodynamically unstable patients in intensive care. While half had direct clinical applications, most remain in early development stages—creating a gap between what's scientifically possible and what hospitals can actually deploy to save lives.
Originaltitel: Computational physiological models for hemodynamic management in critical care: a systematic literature review focusing on model design, credibility and clinical readiness
BACKGROUND: Hemodynamic instability is a highly prevalent, complex and life-threatening condition in critically ill patients. Its multifactorial nature and patient-specific variability challenge standardised treatment approaches. Computational physiological models (CPMs) offer a promising solution by simulating cardiovascular dynamics to guide individualised hemodynamic management. This systematic review evaluates the current landscape of cardiovascular CPMs, focusing on their design, credibility, and clinical readiness. METHODS: A systematic search was conducted in MEDLINE ALL, Embase, Scopus, and Web of Science. Original research articles describing zero-dimensional, closed-loop cardiovascular models with (potential) applications in critical care were included. Data were extracted on context of use, model design, and validation. Model credibility was assessed using a risk-based framework and clinical readiness using a nine-level technology maturity scale. RESULTS: Out of 10,704 screened articles, 183 were included. Direct clinical applications were described in 50% of these studies, including diagnosis, decision support, and closed-loop control. Fluid management was the most common application domain (30%). Personalisation of model parameters was reported in 25% of the articles. While 66% of the articles presented model validation, only 21% achieved moderate credibility scores. Reporting of model characteristics was consistently (100%) insufficient. Most models (75%) were at clinical readiness level 3-4 (model prototyping and development), with four studies reaching clinical testing (level 6-8). CONCLUSION: A substantial body of cardiovascular CPMs exists with promising prospects for relevant applications in critical care, while a large part is currently confined to pre-clinical research settings. Advancing clinical integration requires leveraging existing models, improving transparency in verification and validation, and establishing robust personalisation strategies. TRIAL REGISTRATION: PROSPERO - CRD42022300137, registered on February 11, 2022.