Early risk stratification of late-onset sepsis in very preterm infants by intestinal microbiota profiling: a multicenter case-control validation study.

Intestinal bacterial translocation to the bloodstream is a route of infection for late-onset sepsis (LOS) in preterm infants, highlighting the potential of fecal microbiota profiling for early risk stratification. We aimed to identify and validate LOS-specific gut microbiota signatures. Fifty-eight preterm infants (gestational age < 30 weeks) with blood culture-proven LOS (excluding coagulase-negative staphylococci) were matched to controls (1:1) across three cohorts (Discovery (DC) n = 18; Validation 1 and 2; VC1 n = 12, VC2 n = 28). Fecal samples collected up to 10 days before LOS onset underwent 16S rRNA gene sequencing. Microbial composition, diversity, and discriminatory taxa were compared across LOS subgroups. Random Forest (RF) models were trained in DC and validated in VC1/VC2. Microbiota variation was largely explained by LOS pathogen (R2 = 17%, P < 0.001). Infants with non-staphylococcal and E. coli-LOS showed a temporal increase in relative abundance of Escherichia/Shigella. The RF model distinguishing E. coli-LOS from controls displayed the highest discriminatory performance (AUC = 0.99/0.78/0.61 for DC/VC1/VC2) compared to non-staphylococcal LOS (AUC = 0.96/0.46/0.41). Our findings demonstrate profound microbiota shifts preceding E. coli-LOS, with higher discriminatory ability compared to non-staphylococcal-LOS. While pathogen-specific microbiota-based risk stratification may offer added clinical value, reduced validation performance highlights the limited generalizability and underscores the need for future research before clinical translation.