International Journal of Science, Engineering and TechnologyAuthors: Oyeleke Stevens O, Adebayo Adekunle A, Oladokun David O, Ikotun Olufunmilayo, Dr Oyeleke S.O
Abstract: Background: Intensive care units generate high-frequency, multimodal data suited to artificial intelligence (AI)-enabled clinical decision support systems (CDSS), yet clinical translation remains inconsistent. Objectives: To synthesise evidence on AI-CDSS in adult critical care, appraise methodological maturity against AI reporting standards, and identify implementation barriers. Methods: Narrative synthesis following the SANRA framework. Literature was appraised against TRIPOD+AI, PROBAST+AI, DECIDE-AI, and CONSORT-AI standards across six dimensions: validation tier, data generalisability, metric completeness, human-AI workflow integration, equity reporting, and deployment status. Results: Diagnostic AI models report high internal discrimination but inconsistent calibration and limited prospective validation. Triage systems outperform static scoring retrospectively, yet alert fatigue and clinician override rates remain underreported. Treatment decision support models often conflate observational prediction with causal intervention effects. Cross-cutting gaps include single-centre training data, heterogeneous equity reporting, and absence of standardised post-deployment monitoring. Conclusions: AI-CDSS in critical care exhibits strong algorithmic promise but fragmented clinical validation. Priorities include prospective multi-centre evaluation, human-centred workflow integration, causal treatment framing, and standards-compliant reporting.
DOI: http://doi.org/10.5281/zenodo.20693087
