1f - Are AI-based surveillance systems for Healthcare-Associated Infections really ready for clinical practice? A systematic review and meta-analysis
Fireside Abstracts
Information
Background : Healthcare-associated infections (HAIs) are a global public health concern, causing significant clinical and financial burdens. Despite advances, surveillance methods are often manual and resource-intensive, leading to underreporting. Automation, particularly through AI, holds promise, yet adoption challenges persist. This review aims to evaluate the performance and impact of AI in HAI surveillance, considering technical, clinical, and implementation aspects.
Methods : We conducted a systematic review of Scopus and Embase databases in accordance with PRISMA guidelines. Two independent reviewers performed study selection, data extraction, and quality evaluation. We synthesized information on AI-based models for HAI detection, including their development, performances, and impact. Accuracy, AUC, sensitivity, and specificity metrics were pooled using a random-effect model, stratifying by HAI type. Our study protocol was registered in PROSPERO (CRD42024524497).
Results : Of 2,834 identified citations, 249 studies were reviewed. The performances of AI models were generally high, but with significant heterogeneity between HAI type. Overall pooled sensitivity, specificity, AUC, and accuracy were respectively 0.835, 0.899, 0.864 and 0.880. About 36.7% studies compared AI system performances with others, with most achieving better or comparable results than clinical scores or manual surveillance. Less than 7.6% measured AI real impact, with the majority finding benefits. Only 30 studies deployed the model in a user-friendly tool, 9 tested it in real clinical practice.
Conclusions : Although AI shows promising performance in HAI surveillance, its adoption in clinical practice remains uncommon. Despite spanning over a decade, retrieved studies offer scant evidence on reducing burden, costs, and resource use. This prevents their potential superiority over traditional or simpler automated surveillance systems from being fully evaluated. Further research is necessary to assess impact, enhance interpretability, and ensure reproducibility.
