Article Text

Download PDFPDF

Risk factors for late onset gram-negative infections: a case–control study
  1. Srabani Samanta1,
  2. Kate Farrer2,
  3. Aodhan Breathnach3,
  4. Paul T Heath4,5
  1. 1Neonatal Unit, St Mary's Hospital, Manchester, UK
  2. 2Neonatal Unit, Rosie Hospital, Addenbrooke's NHS Trust, Cambridge, UK
  3. 3Department of Medical Microbiology, St George's Hospital, London, UK
  4. 4Child Health, St George's, University of London, London, UK
  5. 5St George's Vaccine Institute, St George's, University of London, London, UK
  1. Correspondence to Dr Paul T Heath, Division of Child Health, St George's, University of London, Cranmer Terrace, Tooting, London SW17 0RE, UK; pheath{at}


Objectives To determine the incidence, mortality and risk factors for neonatal late onset gram-negative sepsis and meningitis (LOGNS).

Design Retrospective case–control study.

Setting Tertiary neonatal unit in London.

Patients Consecutive inborn infants with late onset (>48 h of life) invasive gram-negative infections diagnosed between 1999 and 2005. Controls were healthy infants matched for gestation and time of admission to the neonatal unit.

Main outcome measures Clinical and risk factor data.

Results 73 cases of LOGNS were identified of which 48 were inborn and included in the study (incidence 1.85/1000 live births). Enterobacter spp. (28%), Escherichia coli (27%) and Klebsiella spp. (21%) were the most common pathogens. The majority of infants were of very low birthweight (VLBW; 79%), and cases and controls were well matched (median gestation 26 weeks). Overall case death was 27% in cases versus 13.5% in controls (p=0.08). There was no significant difference between cases and controls regarding maternal risk factors. Mechanical ventilation, total parenteral nutrition (TPN) and its duration, presence of a central venous line and its duration, use of specific antibiotics and the occurrence of necrotising enterocolitis at or before the first positive culture were all significantly associated with case status in univariate analysis. In multivariate logistic regression analysis, only duration of TPN at or before first positive blood culture remained independently associated with LOGNS (p<0.001).

Conclusions LOGNS occurs predominantly in VLBW infants. When the influence of gestational age is accounted for, the only independent risk factor found for late onset gram-negative neonatal infections is the duration of TPN.

Statistics from


Late onset infections pose a major problem in neonatal units, affecting approximately 25% of very low birthweight (VLBW) infants.1 Late onset gram-negative sepsis and meningitis (LOGNS) in particular, is recognised as a significant cause of morbidity and mortality in the newborn population.

Surveillance data over the last two decades reveal the increasing contribution of gram-negative organisms to late onset sepsis, one study demonstrating a more than twofold increase from 18% (1986–1994) to 43% (1996–2000).2 The mortality of LOGNS is also high and an inverse relationship with gestational age and birth weight has been well described.1,,4

Premature infants are at particular risk of late onset infections as a result of their immature immune system, further compromised by an ineffective skin barrier and the increased need for invasive devices to sustain life-supporting care. In contrast to early onset infection, where vertical transmission of organisms from mother to baby is responsible, late onset infections are generally believed to result from nosocomial or community sources. This is reflected in the different spectrum of organisms responsible.3 Most late onset infections are due to coagulase negative Staphylococci,2 5 followed by Staphylococcus aureus and gram-negative bacilli.2 6

What is already known on this topic

  • Neonatal late onset gram-negative sepsis and meningitis (LOGNS) occurs predominantly in premature infants.

  • Studies have shown risk of infection is associated with a number of factors, many of which are also commonly found in premature infants.

What this study adds

  • When the influence of gestational age is accounted for, the only independent risk factor found for late onset gram-negative neonatal infections is the duration of total parenteral nutrition.

  • This finding suggests that there may be further opportunities for the prevention of gram-negative infections in neonates.

A range of studies has defined possible risk factors for late onset neonatal infections, including aspects of both maternal7 and neonatal care.1 8,,10 Many of these factors, however, are common among premature babies, making it difficult to draw conclusions regarding causality.

The aim of this study was to identify potential risk factors for cases of LOGNS using a case–control design, and to review the incidence, epidemiology and mortality of LOGNS in a tertiary neonatal unit.


We conducted a retrospective case–control study over a period of 6 years (1999–2005) at St George's Hospital in London. LOGNS was defined as a positive culture of a gram-negative organism from a normally sterile site in a baby of >48 h of age. Positive cultures were identified from the hospital microbiology database.

Cases of LOGNS were eligible for inclusion if they remained in the neonatal unit for more than 72 h. If a case had more than one gram-negative infection, only the first episode was used for analysis of risk factors. Controls were matched according to gestation (within 2-week age bands, eg, 24–26 weeks, etc) and time of admission to the neonatal unit (first baby in the appropriate age band born before and first baby born after the relevant case). Control infants were eligible if they remained on the neonatal unit for more than 72 h and did not develop a gram-negative infection during their neonatal stay. Only babies born in our hospital were included.

One investigator (SS) collected data from the hospital records of cases and controls using a standardised proforma. This sought basic demographic, clinical and outcome details as well as information on possible risk factors for infection, identified from published studies of LOGNS.1 7,,10 Maternal risk factors included chorioamnionitis, intrapartum antibiotic prophylaxis (IAP) use, duration of membrane rupture before delivery and use of antenatal steroids. Neonatal risk factor data collected included ventilation type and duration, total parenteral nutrition (TPN) use and duration, the age at onset of enteral feeds, central line use and duration, antibiotic use by type and duration, and the presence of necrotising enterocolitis (NEC). For cases, exposure to these risk factors was determined from their day of admission to the neonatal unit until the day of their first positive culture with a gram-negative organism. For controls, exposure was determined from their day of admission to the neonatal unit until the day of positive culture for their matched case.

Statistical methods

Data were analysed using SPSS v 16 for Windows statistical package. Categorical variables were analysed using a χ2 test and non-parametric continuous variables using Mann–Whitney's test. A backward stepwise (conditional) multivariate logistic regression analysis was performed to identify the relationship between variables. For both univariate and multivariate analysis, p<0.05 was considered statistically significant.


Initially, 78 cases of gram-negative infection were identified, of which five had early onset infection and 25 had been born in another hospital. Thus, 48 cases (who had 53 episodes of LOGNS) were eligible for inclusion. The overall incidence of LOGNS was 1.85 per 1000 live births. LOGNS occurred in 2.2% of total admissions to the neonatal unit during this period and in 4.4% of inborn VLBW admissions. The median age at presentation was 18 days (range 3–192 days) and 13 (27%) cases died during their neonatal stay; in 10 cases gram-negative infection was considered to be the primary cause of death. The overall mortality in the control group was 13.5% (n=13; p=0.08 vs overall case death rate for cases).

Table 1 shows the demographic characteristics of cases and controls. There was no significant difference between cases and controls regarding these factors.

Table 1

Demographic characteristics of case and control subjects

A wide variety of gram-negative organisms were isolated with Enterobacter spp. (15, 28%), Escherichia coli (14, 26%) and Klebsiella spp. (11, 21%) the predominant organisms followed by Acinetobacter spp. (5, 9.5%), Serratia spp. (4, 7.5%), Pseudomonas aeruginosa (2, 4%), Stenotrophomonas maltophilia and Citrobacter freundii (one of each). One baby had a simultaneous growth of an Enterobacter spp. and a Pseudomonas spp.

Blood was the only source of the gram-negative isolate in 90% (n=43) of infections. Two babies also had confirmed meningitis with positive bacterial growth in cerebrospinal fluid (CSF). Another three babies whose blood cultures were positive with a gram-negative isolate, had evidence of meningitis with increased CSF cellularity or evidence of the pathogen on autopsy (positive brain culture with coliforms). All five of these babies were treated for meningitis (either empirically or following bacteriological confirmation).

There was no significant difference between cases and controls in terms of maternal risk factors (table 2).

Table 2

Maternal factors in cases and controls

Table 3 shows the significant associations identified on univariate analysis.

Table 3

Potential risk factors for LOGNS on univariate analysis

Cases were more likely to be ventilated and for a significantly longer duration (14 vs 7 days). There was a significant delay in starting enteral feeds in cases compared to controls. Cases were also more likely to be on TPN and for a longer duration. Of those on TPN, 44% of cases (19 of 43) and 23% of controls (16 of 71) had been on TPN for more than 10 days before the first episode of LOGNS (p=0.02). More cases had a central line at some point before first positive culture compared to controls and for a significantly longer duration (17 vs 8 days). A diagnosis of NEC at or before the first positive blood culture was also identified more commonly in cases than in controls. Although total usage of any antibiotics prior to first positive culture was no different between cases and controls, specific use of vancomycin and piperacillin-tazobactam was more common among cases.

After multivariate logistic regression analysis, only duration of TPN at or before first positive culture remained independently associated with cases of LOGNS (p<0.001).


Our study has described the burden of LOGNS in a tertiary neonatal unit over a 6-year period with a specific focus on potential risk factors. Using a case–control approach in which we matched controls to cases on gestational age, we have identified duration of TPN as an independent risk factor for infection. The importance of this approach is evident from the high proportion of control infants with risk factors potentially relevant to susceptibility to gram-negative infections. The relevance of this finding is therefore strengthened.

Several studies have examined longitudinal data and estimated the incidence of LOGNS in neonatal units.2 6 11 Other studies have examined potential risk factors for infection but have mainly focused on specific gram-negative bacteria in the context of outbreaks or as part of another trial. Graham et al,12 for example, performed a case–control study addressing risk factors for LOGNS in VLBW infants, but as the study was part of a larger clinical trial looking at hand hygiene techniques, it was not adequately powered for this analysis. Some studies have considered potential maternal and neonatal risk factors.

Maternal risk factors

Several studies have suggested that although IAP for group B streptococcal (GBS) infections has significantly reduced the incidence of early onset GBS sepsis, it may have done so at the expense of an increase in early onset gram-negative sepsis, particularly in preterm and VLBW infants.13,,18 More recently, Glasgow et al7 have reported an association between IAP and late onset bacterial infections in term infants with mainly ampicillin-resistant organisms. In this study the majority of cases with serious bacterial infection had urinary tract infections without bacteraemia; only 21% had any evidence of blood stream infection. In contrast, longitudinal prospective surveillance by the Australasian Study Group for Neonatal Infections has showed that IAP use has been associated with a steady decline in early onset GBS disease coupled with an overall trend towards decreasing early onset E coli sepsis; the rate in VLBW infants has remained stable.19 With this background we were interested to assess the relationship between maternal IAP in our population and LOGNS. Although IAP use was relatively high among case mothers, it was also high among control mothers, with no significant difference found. IAP use is generally higher among preterm pregnancies, stressing the importance of matching controls to cases on the basis of gestation. Maternal antenatal steroid exposure has also been associated with early onset neonatal sepsis.20 In our population there was no association with antenatal steroids or other maternal factors including chorioamnionitis, prolonged rupture of membranes and maternal fever. These findings are similar to those from the National Institute of Child Health and Human Development Neonatal Research Network.1

Neonatal risk factors

Several studies of late onset sepsis have identified associations with mechanical ventilation, use and duration of TPN, use and duration of central lines, and the presence of NEC.1 18 Others have identified mechanical ventilation,8 21 TPN,9 22 NEC,9 exposure to antibiotics,23 24 duration12 and use of central lines,10 24 and duration of nasal continuous positive airway pressure12 as independent risk factors for LOGNS. We found all of these to be associated with LOGNS in univariate analysis, but only duration of TPN was found to be an independent risk factor in the multivariate model. There is good evidence for the protective effect of early enteral feeding,25 especially with human breast milk, in preventing neonatal sepsis.1 26 Our study also suggests that later commencement of enteral feeding is associated with the occurrence of LOGNS (in univariate analysis) but provides more powerful, albeit indirect, evidence that reaching full enteral feeds sooner (ie, no longer requiring TPN) might confer a protective effect.

Apart from absence of the benefits of enteral feeding, the mechanism by which TPN may predispose to infection is uncertain. It is widely believed that the gut is an important source of the bacteria that cause bacteraemia in neonates and this may be especially so in those receiving TPN. In a prospective study of neonates receiving TPN, surveillance cultures of the oropharynx and gut were obtained twice weekly and cultured organisms were compared with those isolated in blood cultures taken for clinical indications. In six of 10 infants with septicaemia, the blood isolate was the same as the enteric isolate, supporting this hypothesis.27

An association has also been described between TPN related hyperglycaemia, prolonged hospital stay and mechanical ventilation in infants with sepsis.28 TPN related hyperglycaemia is well recognised in the newborn preterm population because of their diminished tolerance for parenteral glucose delivered at high rates.29 Conversely, infection itself can cause impaired hepatic glucose uptake and increase insulin resistance, thus further exacerbating hyperglycaemia.30 One mechanism proposed for immune compromise is through mitochondrial dysfunction secondary to hyperglycaemia.31 Unfortunately, glycaemia was not systematically documented in our patient population. This hypothesis therefore requires further study.

Limitations of our study

There are several limitations to our study and study design. Because it was conducted retrospectively, it was not possible to accurately determine the types and amounts of enteral feeds and so we were not able to draw conclusions regarding the influences of breast milk, artificial formula or mixed feeding. We were also not able to draw conclusions on risk factors for specific gram-negative organisms because of relatively small numbers of individual pathogens. As with any assessment of risk factors, it may be that other risk factors exist which we did not consider. It is therefore conceivable that any differences found between cases and controls might in fact reflect other unmeasured factors. Finally, because we matched cases and controls on gestational age, we are not able to draw conclusions about the importance of prematurity as a risk factor. However, this association is evident from a range of studies.


LOGNS predominately occurs in premature and VLBW babies. Maternal factors do not appear to be associated with LOGNS and although a number of significant associations with neonatal factors were identified on univariate analysis, many of these likely reflect their prevalence in the low birthweight population. The duration of TPN appears to be an important risk factor for LOGNS and although identified in previous studies, we have further clarified its independent role. The mechanisms for this require further research.



  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.