<![CDATA[Prognostication in neurocritical care remains challenging due to patient heterogeneity and the limitations of static conventional scores. This systematic review and meta-analysis evaluated the diagnostic accuracy of artificial intelligence (AI) models as retrospective predictive and prognostic frameworks, assessing their potential utility as clinical early warning systems (EWS). Conducted via PubMed, ScienceDirect, and SCOPUS per PRISMA guidelines, the study evaluated deterioration of consciousness, mortality, and neurological outcomes. Out of 20 included studies, 18 were meta-analyzed. For mortality prediction, AI demonstrated a pooled sensitivity of 0.569, specificity of 0.810, diagnostic odds ratio (DOR) of 9.620, and area under the curve (AUC) of 0.698. In predicting deterioration in consciousness, AI achieved a pooled sensitivity of 0.648, specificity of 0.901, DOR of 38.346, and AUC of 0.796. For neurological outcomes, the pooled sensitivity was 0.864, specificity 0.865, DOR 45.566, and AUC 0.858. AI models demonstrate high accuracy in predicting long-term neurological outcomes and acceptable performance in predicting short-term deterioration in consciousness, but remain highly limited and offer no clear advantage over conventional scoring for ICU mortality. These retrospective findings provide a robust foundation for future prospective designs, though widespread clinical integration is not yet warranted.]]>
Copyrights © 2026
