neurodevelopment* or impairment or long-term outcome. (3) Newborn: newborn or neonate or premature or preterm or very low *birth weight or VLBW or extremely low* birth weight or ELBW. (4) Four-study design: prevalence or cohort or follow-up or incidence or case control or prognosis. 1 Department of Community Health Sciences, Faculty of Medicine, University of Calgary, Calgary, AB, Canada and 2 Department of Pediatrics, University of Calgary, Calgary, AB, Canada. Correspondence: Dr B Alshaikh, Department of Pediatrics, Faculty of Medicine, University of Calgary, Room C211, 1403 29th Street NW Calgary, AB T2N 2T9, Canada. E-mail: belal.alshaikh@albertahealthservices.ca Received 15 August 2012; accepted 10 December 2012; published online 17 January 2013 Journal of Perinatology (2013) 33, 558–564 & 2013 Nature America, Inc. All rights reserved 0743-8346/13 www.nature.com/jp Study selection criteria Two authors (BA and KY) searched the literature independently and assessed the eligibility of identified articles for this systematic review and meta-analysis. All following preselected criteria were considered for including any study in the analysis: (1) studies on VLBW infants (o1500 g). (2) Studies involving infants with neonatal culture-proven sepsis, that is, sepsis accompanied by the presence of an organism in the blood during the admission period in the neonatal intensive care unit. (3) Long-term follow-up for a minimum of 12 months. (4) A priori definition of moderateto-severe neurodevelopmental impairment (NDI) that included at least one of the following: CP, cognitive delay (cognitive score 2 s.d. less than mean on standardized psychological testing), vision loss or deafness. In addition, only studies reported original data (that is, no review articles) were included. Data extraction Data extraction was performed independently by two reviewers (BA and KY) using a standardized data collection form. Any discrepancies in extracted data were solved by consensus. The following data were collected: journal name, year of publication, single-center or multicenter status, demographic data, year of birth, number of infants in both proven sepsis and comparison groups, definitions adopted, inclusion and exclusion criteria, and methods and duration of follow-up. The primary outcome was one or more of NDIs in VLBW infants with culture-proven sepsis in neonatal period as assessed by odds ratio (OR) in individual studies. We used the OR, lower and upper limit of the 95% confidence interval (CI) as reported in the studies. In the studies where ORs for the association between neonatal sepsis and long-term NDI were not reported, we calculated ORs using 2 2 tables. Data synthesis and analysis STATA version 11 (Stata, College Station, TX, USA) was used for all statistical analysis. Estimates for OR together with the corresponding 95% CIs and the percentage weight contributed to the overall meta-analysis from each study were calculated. For each outcome of interest, effect estimates were pooled assuming a random-effect modeling approach given that the data were retrieved from the literature and expected to have variable sizeeffect. Heterogeneity across the observational studies was evaluated using both the I 2 statistics; Po0.1 was defined to note statistical significance in the analysis of heterogeneity. In addition, we assessed for potential evidence of publication bias through visual inspection of the Begg’s funnel plot and the Egger test for funnel plot asymmetry. Finally, the following sensitivity and subgroup analyses were planned: (1) Studies in which at least 80% of infants had follow-up because followup rate is crucial in accurate interpretation of long-term neurodevelopmental outcome. (2) Studies involving only neonates born in the post-surfactant era given that long-term outcome of VLBW infants improved significantly after introducing surfactant. (3) Analysis of studies reporting long-term outcome for infants with coagulase negative staphylococcus (CoNS) infection given that CoNS is the most common cause of infection in VLBW infants. (4) Comparison between the components of neurodevelopmental outcome. RESULTS Our specified search strategy identified 1362 abstracts (Figure 1). In all, 33 studies reporting sepsis in neonates and long-term neurodevelopment impairment were retrieved for full detailed evaluation. Seventeen studies met the inclusion criteria. There was an excellent agreement for the inclusion of individual studies (kappa 0.94, 95% CI (0.92 to 0.96)). The characteristics of these 17 studies (14 cohort and 3 case–control studies) involving 15 331 VLBW survivors of neonatal sepsis are summarized in Tables 1 and 2. Analysis combining all studies VLBW infants with sepsis in the neonatal period had poor longterm neurodevelopment outcome compared with those without sepsis (OR 2.09; 95% CI 1.65 to 2.65; Figure 2). There was considerable heterogeneity within the studies (I 2 ¼ 36.9; P ¼ 0.06), representing variations related to a number of factors, including patient characteristics, specifics of the illness and long-term follow-up rates. There was some evidence to suggest probable publication bias in Egger’s test (coefficient 0.80, 95% CI 0.03 to 1.52, P ¼ 0.04); however, this was not evident in Begg’s test (P ¼ 0.53, with continuity correction) although it appeared highly likely in