Artificial intelligence in the prediction of intraoperative red blood cell transfusion in cardiac surgery: a systematic review and diagnostic test accuracy meta-analysis.

BACKGROUND: Red Blood Cell (RBC) transfusion in cardiac surgery is associated with risks. Conventional prediction scores lack accuracy, conflicting with Patient Blood Management (PBM) principles. Artificial Intelligence (AI) offers a potential avenue for developing precise, data-driven predictive models to enhance clinical decision-making and optimize patient outcomes. METHODS: A systematic review and meta-analysis of diagnostic test accuracy studies was conducted following PRISMA guidelines, searching PubMed, Embase, and Cochrane Library databases until April 2025. Studies developing AI models to predict intraoperative Red Blood Cell (RBC) transfusions in adult cardiac surgery were included. Pooled sensitivity, specificity, and Area Under the receiver operating Characteristic Curve (AUC) were calculated using a bivariate random-effects model. To summarize overall diagnostic performance, Summary Receiver Operating Characteristic curves were generated. RESULTS: Five studies encompassing 3,063 patients were analyzed in the meta-analysis. AI models demonstrated high pooled specificity, ranging from 86.3% (95% CI 82.8%‒89.1%) to 93.6% (95% CI 84.3%‒97.6%), whereas pooled sensitivity ranged from 50.3% (95% CI 20.9%‒79.6%) to 55.7% (95% CI 33.8%‒75.6%). AUC values ranged between 0.793 (95% CI 0.634‒0.899) and 0.892 (95% CI 0.740‒0.943). Preoperative hemoglobin levels and patient age were consistently identified as clinical predictors for intraoperative RBC transfusion. CONCLUSION: This systematic review and meta-analysis suggests that AI models may have potential for predicting intraoperative RBC transfusion in cardiac surgery, with consistently high specificity but only moderate sensitivity. However, given the low certainty of evidence, substantial heterogeneity, and reliance on non-standardized transfusion practices, these findings should be interpreted cautiously.