Perfusion index as adjunct to SNAPPE-II for early risk stratification of sepsis-dominated adverse outcomes in neonates: Exploratory prospective cohort study

Background Illness-severity scores such as the Score for Neonatal Acute Physiology with Perinatal Extension-II (SNAPPE-II) discriminate mortality well but morbidity less well. Whether serial perfusion index (PI), a continuous bedside perfusion measure, adds prognostic information beyond SNAPPE-II is unknown, particularly in late-preterm and term populations in low- and middle-income countries. Methods In this exploratory prospective cohort, 50 neonates admitted within 12 hours of birth to a level-3 neonatal intensive care unit and requiring arterial blood-gas analysis were enrolled consecutively. SNAPPE-II used worst first-12-h values; PI was measured at 6, 12 and 24 hours. Outcomes were death, major morbidity (sepsis, necrotising enterocolitis, bronchopulmonary dysplasia) and their composite. Given eight deaths, multivariable models were exploratory, using Firth-penalised regression. Discrimination was compared by change in area under the curve (ΔAUC; DeLong and bootstrap confidence intervals) with optimism correction. Serial PI was analysed using a random-intercept linear mixed model with time × outcome interaction. Results Mortality was 16.0% (8/50); major morbidity, exclusively sepsis, was 44.0% (22/50). Non-survivors had higher SNAPPE-II scores (median 50.5 vs 5.0) and lower PI at all time points ( p < 0.001). SNAPPE-II discriminated mortality well (AUC 0.936); adding 6-h PI did not improve discrimination (ΔAUC 0.0015, 95% CI −0.029 to 0.032). For morbidity outcomes, 6-h PI improved model fit (likelihood-ratio p = 0.006 and 0.016), but ΔAUC confidence intervals included zero. Neonates with culture-proven sepsis ( n = 6) had lower 6-h PI (median 0.76 vs 1.49, p = 0.014). PI trajectories did not diverge by outcome (time × outcome p = 0.671); values remained lower in non-survivors. Conclusions In this selected cohort, low PI was associated with illness severity and sepsis-dominated adverse outcomes but did not demonstrably improve discrimination beyond SNAPPE-II. Unblinded ascertainment precludes excluding incorporation bias. These hypothesis-generating findings warrant testing, with prespecified thresholds, blinded ascertainment and internal validation, in adequately powered multicentre studies of culture-proven sepsis.