Objective To investigate the accuracy of procalcitonin (PCT) as a diagnostic marker of nosocomial sepsis (NS) and define the most accurate cut-off to distinguish infected from uninfected neonates.

Setting Six neonatal intensive care units (NICUs).

Patients 762 neonates admitted to six NICUs during a 28-month observational study for whom at least one serum sample was taken on admission.

Main outcome measures Positive and negative predictive values at different PCT cut-off levels.

Results The overall probability of an NS was doubled or more if PCT was >0.5 ng/ml. In very-low-birth-weight (VLBW) infants, a cut-off of >2.4 ng/ml gave a positive predictive value of NS near to 50% with a probability of a false-positive diagnosis of NS in about 10% of the patients.

Conclusions In VLBW neonates, a serum PCT value >2.4 ng/ml prompts early empirical antibiotic therapy, while in normal-birth-weight infants, a PCT value ≤2.4 ng/ml carries a low risk of missing an NS.