AI tools for sepsis prediction: a systematic review

A peer-reviewed review in npj Digital Medicine, published August 31 by Aadith Ashok and co-authors at Alfred Health, Monash University and collaborating institutions, pooled 53 studies covering more than seven million admissions.

The authors report a pooled AUROC of 0.88, sensitivity of 77.2 per cent, specificity of 84.7 per cent and a positive predictive value of 34.2 per cent.

The discrimination is promising, but the low PPV could create an alert burden.

Limitations include heterogeneity above 95 per cent (I-squared), bias, and no proof of improved patient outcomes; there is no independent deployment validation.