Palivizumab Prophylaxis during Nosocomial Outbreaks of Respiratory Syncytial Virus in a Neonatal Intensive Care Unit: Predicting Effectiveness with an Artificial Neural Network Model

Study Objective To identify subgroups of premature infants who may benefit from palivizumab prophylaxis during nosocomial outbreaks of respiratory syncytial virus (RSV) infection. Design Retrospective analysis using an artificial intelligence model. Setting Level IIIB, 35-bed, neonatal intensive care unit (NICU) at a tertiary care hospital in the United Arab Emirates. Patients One hundred seventy six premature infants, born at a gestational age of 22-34weeks, and hospitalized during four RSV outbreaks that occurred between April 2005 and July 2007. Measurements and Main Results We collected demographic and clinical data for each patient by using a standardized form. Input data consisted of seven categoric and continuous variables each. We trained, tested, and validated artificial neural networks for three outcomes of interest: mortality, days of supplemental oxygen, and length of NICU stay after the index case was identified. We compared variable impacts and performed reassignments with live predictions to evaluate the effect of palivizumab. Of the 176 infants, 31 (17.6%) received palivizumab during the outbreaks. All neural network configurations converged within 4seconds in less than400 training cycles. Infants who received palivizumab required supplemental oxygen for a shorter duration compared with controls (105.2 +/- 7.2days vs 113.2 +/- 10.4days, p=0.003). This benefit was statistically significant in male infants whose birth weight was less than 0.7kg and who had hemodynamically significant congenital heart disease. Length of NICU stay after identification of the index case and mortality were independent of palivizumab use. Conclusion Palivizumab may be an effective intervention during nosocomial outbreaks of RSV in a subgroup of extremely low-birth-weight male infants with hemodynamically significant congenital heart disease.