<![CDATA[Background: Hemodynamic instability is a major predictor of organ failure and mortality in ICU patients. Conventional monitoring often fails to detect early deterioration, which has encouraged the use of artificial intelligence (AI) to improve the detection and prediction of hemodynamic instability. Methods: This systematic review followed the PRISMA 2020 guidelines and analyzed studies using machine learning or deep learning to predict hypotension, vasopressor requirements, or hemodynamic instability in adult ICU patients. Six major databases were screened, and 16 studies met the inclusion criteria. Due to heterogeneity in model design and outcomes, the findings were synthesized narratively. Results: The included studies comprised retrospective model development, multicenter validation, prospective evaluation, and two randomized clinical trials. Multivariable models such as the hemodynamic stability index (HSI) demonstrated strong predictive performance (AUROC 0.76–0.90). Dynamic models such as TvHEWS consistently provided stable predictions with reduced false alarms. Waveform-based predictors, including the hypotension prediction index (HPI), were able to anticipate hypotension 5–15 minutes before onset, even in patients with sepsis. Personalized approaches, such as DynaCEL and HM-TARGET, generated patient-specific hemodynamic targets. Prospective studies showed a reduction in the duration of hypotension, although evidence regarding effects on mortality and organ failure remains limited. Conclusion: Artificial intelligence has the potential to improve the accuracy of hemodynamic monitoring and enable earlier intervention in the ICU. However, large-scale clinical trials are still needed to confirm its benefits on meaningful clinical outcomes.]]>
