Early Detection of Sepsis Using Artificial Intelligence in Intensive Care Units: A Systematic Review and Meta-Analysis

Purpose This meta-analysis aimed to assess the diagnostic performance of artificial intelligence for detecting sepsis in intensive care unit. Methods A thorough literature search was performed using PubMed, Embase, and Web of Science to locate relevant studies published through November 2024. The selected studies specifically examined the diagnostic accuracy of artificial intelligence in identifying septicemia. To estimate pooled sensitivity and specificity values, a bivariate random-effects model was employed, with results reported alongside 95% confidence intervals. Heterogeneity across studies was evaluated using the I 2 statistic. Results Of the 1495 studies initially identified, 16 studies encompassing a total of 159,947 patients, met the inclusion criteria for this meta-analysis. For the internal validation set, the pooled results for sepsis detection showed a sensitivity of 0.76 (95% CI: 0.71-0.80), a specificity was of 0.85 (95% CI: 0.81-0.89), and an area under the curve (AUC) of 0.87 (95% CI: 0.84-0.90). In comparison, the external validation set yielded a sensitivity of 0.78 (95% CI: 0.65-0.87), a specificity of 0.82 (95% CI: 0.76-0.86), and an AUC of 0.87 (95% CI: 0.83-0.89). Deeks’ funnel plot and Egger's test indicated no significant publication bias in both the internal and external validation sets( P = .63,.89). Conclusions The findings of this meta-analysis indicate that artificial intelligence demonstrates a high diagnostic performance in identifying sepsis and septic shock. However, substantial heterogeneity across studies may impact the robustness of this evidence. Further research using external validation datasets is required to confirm these results and evaluate their applicability in clinical settings.