the funnel plot. Figure 3 displays this bias graphically, as there is asymmetry in the funnel plot with a predominance of studies with large standard errors (small studies) showing positive effect studies and a paucity of small negative studies. Sensitivity analysis Results of sensitivity analyses based on the year of birth before 1990 showed similar association between neonatal sepsis and NDI to studies included patients after 1990. However, the magnitude of association was higher in infants born before 1990; OR 2.74 (1.66 to 4.53) vs 1.82 (1.48 to 2.24). Results of using follow-up rate of 80% as a cutoff were not significantly different from the combined analysis of sepsis vs no sepsis (OR 2.06 (1.53 to 2.77) for follow-up 480% vs 2.18 (1.52 to 3.12) for o80%). In regard to study design, cohort studies (OR 2.01; 95% CI 1.69 to 2.39) yield similar result to the combined studies, but the three case–control studies showed higher association between sepsis and NDI (OR 2.89; 95% CI 1.75 to 4.77). Analysis of components of NDI Analysis of individual components of NDI indicated a significantly increased risk of CP (11 studies: OR 2.09; 95% CI 1.78 to 2.45; Figure 4), low psychomotor developmental index (PDI) (3 studies (14,18,22): OR 1.55; 95% CI 1.25 to 1.92) and deafness (2 studies (22,42): OR 2.07; 95% CI 1.09 to 2.94). Analysis for the mental developmental index component revealed no significant difference between sepsis and no sepsis groups (4 studies14,18,22,24: OR 1.16; 95% CI 0.84 to 1.61). Data on visual impairment were limited to one study22 OR 2.7 (2.14 to 3.41). Analysis of CoNS sepsis and candida infection Analysis of the relationship between neonatal sepsis caused by CoNS and neurodevelopmental outcome showed a similar association to that seen in any culture-proven sepsis and NDI (three studies14,20,22: OR 1.31; 95% CI 1.09 to 1.57) and with CP (three studies14,20,22: OR 1.7; 95% CI 1.02 to 2.87). Data on mental 1362 Citation screened 33 Potentially relevant citations identified for further review 16 Articles excluded: 4 All patients were included in other studies 8 No data for VLBW infants or neonatal sepsis 2 Sepsis was diagnosed clinically 1 Follow up was at < 12 months post menstrual age 1 Some data on NDI were collected by phone 17 Studies included in the analysis 1329 Articles excluded: Abstracts not including neurodevelopmental outcome of neonates with sepsis Figure 1. Flow diagram of study selection process. NDI, neurodevelopmental impairment; VLBW, very low birth weight. Neurodevelopmental outcomes of VLBW infants with neonatal sepsis B Alshaikh et al 559 & 2013 Nature America, Inc. Journal of Perinatology (2013), 558 – 564 developmental index and PDI were reported in only one study22. Analysis of the impact of candida infection on NDI showed similar association (three studies22,25,28: OR 2.62; 95% CI 1.37 to 5.0). DISCUSSION Our meta-analysis of 17 studies indicates that the diagnosis of neonatal sepsis in VLBW infants is associated with an increased risk of one or more long-term NDIs. Analyzing the components of NDI revealed a higher association with CP, low PDI and deafness. With the exception of CP, which was included in 11 studies in this analysis, results on PDI and deafness are based on small number of studies. Analyzing data from CoNS and candida infection yielded the same association. Neurodevelopmental outcome of VLBW infants with candida infection was reported separately in three studies.22,25,28 Two of them restricted their studies to fungal infection only. Analysis for infection resulting from bacterial or fungal infection elucidates the same kind of association seen in the combined analysis. Careful interpretation of these results is necessary given the previously mentioned variability in characteristics of the studies, patients and causative pathogens. In addition, the difference in definition of NDI is particularly an important issue. The two largest studies14,22 have similar definitions, whereas the other studies have one or more components from the NDI. Saw and Chen,21 Murphy et al.29 and Grether et al.30 limited their testing to hearing loss or CP. Our systematic review examined the association between neonatal sepsis and one or more of the NDI taking into account the variety and number of neurodevelopmental disabilities included in each study. Most studies defined psychomotor and cognitive impairment as a score of o70 or 42 s.d. below the mean of a validated scale of neurodevelopmental evaluation. Chen et al.19 used a cutoff point of a Bayley score o84, whereas Collados et al.26 used a developmental coefficient of o80 on Denver and Amiel Tison test. The definition of moderate-to-severe impairment based on cognitive function o2 s.d. below the mean can overestimate the rates of disability. However, excluding these two studies did not have any impact on the results of the metaanalysis. The results of our meta-analysis are based on cohort and case– control studies because randomized controlled trials do not exist given that a randomized controlled trial is ethically not possible. Pooling of data from case–control studies requires the use of ORs rather than relative risks, this may lead to overestimation of the effect size, particularly if the risk of NDI is high among infants with sepsis. Nonetheless, using relative risks instead of ORs in cohort