Prediction of nosocomial sepsis in neonates by means of a computer-weighted bedside scoring system (NOSEP score)

Objective: To develop an easy-to-use bedside scoring system, composed of clinical variables, hematologic variables, and risk factors of infection, to predict nosocomial sepsis in neonatal intensive care unit patients. Setting: A neonatal intensive care unit in a university hospital, Antwerp, Belgium. Patients: Over 2 yrs, we analyzed two groups of patients. First, we prospectively studied 104 episodes of presumed nosocomial sepsis in 80 neonates (derivation cohort), and then we retrospectively studied 50 episodes in 39 neonates (validation cohort). Interventions: None. Measurements and Main Results: We developed two versions of a scoring system to predict nosocomial sepsis in sick neonates. The first scoring system (NOSEP-1 score) was based on 15 clinical, 12 laboratory, and 17 historical variables potentially connected with infection; the second one (NOSEP-2 score) also included the culture results of central vascular catheters. Based on the odds ratios of all independent variables, an additive and weighted score was developed and validated in a cohort of 39 patients screened for nosocomial sepsis in the same center. The NOSEP-1 score consisted of three laboratory variables (C-reactive protein greater than or equal to 14 mg/L, thrombocytopenia <150 x 10(9)/L, and neutrophil fraction >50%), one clinical factor (fever >38.2 degrees C [100.8 degrees F]), and one risk factor (parenteral nutrition for greater than or equal to 14 days). The NOSEP-2 score consisted of the same variables plus catheter-hub and catheter insertion site colonization data. Receiver operating characteristic curve analysis demonstrated good predictor performance of the NOSEP-1 score (area under the curve [Az] = 0.82 +/- 0.04 [SEMI) and NOSEP-2 score (Az = 0.84 +/- 0.04, p < .05). We checked whether a complex computer-generated scoring system (CD-1 and CD-2 scores) based on the original numerical values of the items used in NOSEP-1 and NOSEP-2 would improve the prediction of nosocomial sepsis. The analysis showed the accuracy of bedside NOSEP-1 and NOSEP-2 scores to be comparable with the more cumbersome computer-generated CD-1 and CD-2 scores (receiver operating characteristic curve, Az: CD-1 score = 0.81 +/- 0.04, p = .69, and CD-2 score = 0.86 +/- 0.04, p = .96). Finally, in the validation cohort, we showed that the developed scoring system has a good prediction potential for nosocomial sepsis (Hosmer-Lemeshow goodness-of-fit test, chi(2) [19] = 16.34, p > .75). Conclusions: The simple bedside scoring system NOSEP-1 composed of C-reactive protein, neutrophil fraction, thrombocytopenia, fever, and prolonged parenteral nutrition exposure provides a valuable tool for early identification of nosocomial sepsis. Its predictive power can be improved by adding central Vascular catheter insertion site and hub colonization to the score.