Machine learning as a decision-support tool for spontaneous bacterial empyema and spontaneous bacterial peritonitis in cirrhosis

BACKGROUND Bacterial infections are common in patients with cirrhosis and are associated with a high mortality rate.Although the polymorphonuclear leukocyte (PMN) count has been widely used as a key diagnostic indicator for spontaneous bacterial empyema (SBE) and spontaneous bacterial peritonitis (SBP), the sensitivity and accuracy of current diagnostic methods remain limited.AIM To develop an auxiliary diagnostic model based on laboratory test results via a multicenter, large-scale, retrospective study.bacterial peritonitis (SBP).For SBE, the traditional polymorphonuclear leukocyte (PMN) cutoff showed limited sensitivity, whereas the NAM-TABR model increased the Youden index from 0.480 to 0.618, achieving area under the curves (AUCs) of 0.865 (internal) and 0.925 (external), respectively.For SBP, PMN > 50 × 10 6 /L improved the Youden index compared to PMN > 250 × 10 6 /L (0.534 vs 0.429), and the NAM-TABR model further increased it to 0.635, with AUCs of 0.885 (internal) and 0.823 (external), respectively.