Risk Factors for Late-Onset Sepsis in Preterm Infants: A Multicenter Case-Control Study

Late-onset sepsis (LOS), defined as sepsis onset after 72 h of life, is a leading cause of mortality in the neonatal intensive care unit (NICU) [1]. The incidence rates for LOS in preterm infants vary between 20 and 38% in the first 120 days of life, and mortality rates range from 13 to 19% [1-4]. Survivors are at risk for prolonged hospitalization, development of necrotizing enterocolitis (NEC), bronchopulmonary dysplasia, and neurodevelopmental impairment [1, 2, 5]. The diagnosis of LOS in daily clinical practice may be challenging, especially in preterm infants, as clinical symptoms have limited sensitivity and specificity. The gold standard for diagnosis is confirmation of a pathogen in the blood culture, which is limited by suboptimal sensitivity and delay of a definite diagnosis because of the turnaround time to become positive [6]. In addition, screening of bodily fluids (e.g., blood and urine) may also require an invasive procedure, increasing the risk for LOS independently [6]. Several studies have identified risk factors for LOS, including a lower birth weight, gestational age (GA), and the presence of central venous catheters [1, 3, 7]. In addition, breastmilk feeding within the first month of life has been shown to be protective against LOS development [8]. However, most of these studies are characterized by a small number of cases, retrospective and single-center study designs, and the absence of detailed (daily) clinical data, limiting the possibility of adequate matching with controls and thus the ability to draw firm conclusions. Identification of independent risk factors for LOS in preterm infants may allow selection of infants at an increased risk and the development of novel, personalized therapeutic strategies aimed at reducing the LOS incidence. Therefore, we aimed to identify independent risk factors contributing to the development of LOS in preterm born infants in a multicenter case-control study with an overview of the clinical characteristics of patients with LOS within the first month of life.

This case-control study was conducted between October 2014 and January 2017 at 2 level II and 7 level III NICU situated in The Netherlands and Belgium (online suppl. Table 1; for all online suppl. material, see www.karger.com/doi/10.1159/000497781). None of the participating centers administered probiotics routinely. The current study was nested in an ongoing study on the identification of early diagnostic biomarkers for NEC and LOS [9]. In that study, fecal samples and clinical data were collected (if applicable on a daily base) from infants born at a GA ≤30 weeks, up to 28 days’ postnatal age (Table 1). In order to identify factors associated with LOS development the variables were assessed prior to clinical onset in the matched case-control cohort. In case of transfer from the NICU to a referral hospital or death prior to 28 days’ postnatal age, data collection was ceased.

From this original cohort, infants diagnosed with LOS were strictly matched to 1 healthy control infant based on GA (±3 days), birthweight (±10 g), and postnatal age at LOS onset (±0 days). Infants who developed LOS as defined below were included as cases, and infants who did not develop LOS were included as controls. Both cases and controls were excluded in case of early-onset sepsis (positive blood cultures < 72 h postnatally) or in case of NEC (≥Bell’s stage 2A) or SIP during the follow-up period and an incomplete or missing medical file.

LOS cases were defined as infants with a pathogen isolated from the blood culture drawn ≥72 h postnatally and pathogen-based antibiotic treatment was continued for ≥5 days, according to Vermont Oxford criteria [10]. Only the first LOS episode was included in the analysis. Isolated pathogens from blood cultures were classified into 4 categories: gram-positive bacteria, gram-negative bacteria, fungi, and coagulase-negative staphylococci (CoNS). A CoNS-positive culture was considered a CoNS-LOS when a CRP level ≥10 mg/L was measured within 72 h of LOS onset. When ≥2 pathogens were isolated from the blood culture (one being CoNS), the presence of CoNS was considered as contamination. Remaining definitions of collected data are described in Table 1.

Statistical Package for the Social Science (SPSS) version 22.0 (IBM, Armonk, NY, USA) was used for the statistical analysis. First, during the entire inclusion period of 28 days, collected demographic and clinical variables from all infants with LOS were compared with non-LOS cases. Potential associations between assessed variables and the development of LOS were identified via univariate logistic regression analysis.

Secondly, in the strictly 1: 1 matched case-control cohort univariate logistic regression analysis was performed on clinical and demographical variables in the period preceding the day of LOS onset. Subsequently, independent risk factors were identified via multivariable regression analysis. This model was constructed using the backward likelihood ratio method, ultimately including statistically significant variables (p value < 0.05). Variables included in this model were selected based on their two-tailed p value calculated from the univariate regression analyses. Only variables with a p value ≤0.30 were included. For every 10 cases one variable was included in the multivariable regression analysis. Results were considered statistically significant for p values ≤0.05.

Thirdly, potential predictive factors were assessed for the 3 subgroups via univariate logistic regression. In addition, independent risk factors were identified using multivariable logistic regression models. This model was constructed using the forward likelihood ratio method, considering the smaller sample size, ultimately including statistically significant variables (p value < 0.10). Other statistical settings remained the same as described for the total matched case-control cohort.

All results are displayed as (unadjusted) OR with corresponding 95% CI and p values.

In total, 755 infants were included in the analysis, constituting 194 LOS cases (23%). The demographic and clinical characteristics of the LOS cases and controls in the overall cohort are depicted in online supplementary Table 2. Further clarification of the LOS incidence, the causative pathogen distribution, and the median age of onset are provided in online supplementary Tables 3 and 4.

Table 2 provides an overview of the demographics and characteristics of the LOS cases versus the matched controls (1: 1) irrespectively of the causative pathogen. Duration of parenteral feeding was identified as an independent risk factor for LOS development irrespectively of the causative pathogen (OR = 1.125; 95% 1.041–1.216; p = 0.003). Third-generation cephalosporins administration was identified as an independent factor inversely associated with LOS development (OR = 0.562; 95% CI 0.320–0.987; p = 0.045). Remaining variables showed no significant differences.

No differences in clinical and demographic characteristics were found between gram-negative LOS cases and controls prior to LOS onset. However, a higher mortality rate was observed in LOS cases (unadjusted OR = 11.400; 95% CI 1.367–95.043; p = 0.025) (Table 3).

Duration of parenteral feeding (days) was identified as an independent risk factor for Gram-positive LOS (OR = 1.289; 95% CI 1.074–1.547; p = 0.006). Antibiotics exposure prior to clinical onset was inversely related to LOS development (OR = 0.078; 95% CI 0.007–0.879; p = 0.039). Remaining variables showed no significant differences (Table 4).

The total number of days of peripheral line exposure (OR = 1.238; 95% CI 1.086–1.411; p = 0.001) and formula feeding (OR = 3.779; 95% CI 1.257–11.363, p = 0.018) preceding clinical onset were identified as independent risk factors for CoNS-LOS. Administration of third-generation cephalosporins was found to be an independent factor inversely associated with CoNS-LOS (OR = 0.229; 95% CI 0.086–0.612; p = 0.003). There were no significant differences regarding the remaining variables between the 2 subgroups (Table 5).

This case-control study aimed at identifying demographic and clinical risk factors associated with the development of LOS in preterm infants in a multicenter setting. We demonstrated that every additional day of parenteral feeding was associated with an increased risk of LOS development. Third-generation cephalosporins administration was identified as an independent factor inversely associated with the development of CONS-LOS, whereas formula feeding was associated with an increased risk.

In a previous study, formula-fed infants showed increased odds for CoNS-LOS development compared to breast-fed infants [11], and this was confirmed in the current study. Breast milk might be protective due to its anti-infective, microbiome-modulating, and immune-stimulatory properties [12]. Several studies have demonstrated that infants who receive breastmilk are more likely to achieve full enteral feeding at an earlier stage compared to formula-fed infants, resulting in earlier cessation of parenteral feeding [1, 4, 13]. We demonstrated that exposure to parenteral feeding for more than 10 consecutive days was associated with an increased risk of LOS development. It could be debated whether clinicians should aim to limit the exposure to parenteral feeding to no longer than 10 days by a more rapid advancement of enteral feeding with preferably breastmilk to reduce the risk of LOS development. On the contrary, a rapid advancement of enteral feeding might increase the risk for NEC development. However, studies have shown that rapid advancement of the enteral feeding volume within the first week of life is not significantly associated with NEC in preterm and very low birthweight infants [14, 15].

In this study, exposure to antibiotics was associated with decreased odds for the development of gram-positive LOS, irrespectively of the type and duration of antibiotics. Cephalosporin exposure was associated with a decreased risk for CoNS-LOS, possibly due to the sensitivity of CoNS species to cephalosporins. Therefore, exposure to this agent could reduce the risk of invasion of CoNS from either the skin or the gut into the bloodstream [16, 17]. However, implementation of routine administration of cephalosporins in preterm infants remains a topic of debate mainly because of the increased risk for colonization with extended-spectrum β-lactamase producing bacteria [18]. The observed protective effects of early exposure to specific antibiotics against the development of LOS indicate that the microbiota may be involved in the pathophysiology of at least a selection of LOS cases. The influence of early microbiota colonization and alterations in microbiota composition in LOS pathophysiology has been considered in several studies [19, 20]. This phenomenon might lead to the development of strategies aimed at early manipulation of the microbiota to prevent LOS development, for example by administration of probiotics instead of antibiotic prophylaxis, reducing the risk for colonization with multi-resistant pathogens [21]. It has been demonstrated that probiotic supplementation significantly reduced the risk of LOS in preterm infants (n = 9,416) [22]. However, additional studies are needed to evaluate the optimal dosage, duration, and identification of the best suitable bacterial strains for supplementation.

Previous studies have demonstrated an association between (the duration of) central line exposure and the development of LOS in preterm infants [1, 7, 23]. Line exposure significantly increased the risk of gram-positive bacteria-related LOS in preterm infants, especially CoNS-LOS. This increased risk may be caused by contaminated intravenous fluids or catheter hubs (intraluminal contamination) or by skin-colonizing organisms invading the bloodstream via the catheter track (extraluminal contamination) [24]. We observed that every additional day that a peripheral line was present the risk of CoNS-LOS increased, while central line exposure (presence/absence and duration) was not an independent risk factor for CoNS-LOS. The apparent discrepancy in the study results might be explained by differences in study design. The majority of preterm infants have either a central or a peripheral line during the first month of life, and a younger GA is associated with an increased risk for LOS [1, 7]. In the current study we matched study participants on GA to prevent bias by this age-related catheter exposure. A positive association between the dwell time of peripheral catheters and central venous lines and LOS development has been described in several studies, although the results are contradictory [23, 25, 26]. We found no association between the dwell time of both central and peripheral catheters and LOS. We hypothesized that an increased risk of LOS development is not merely influenced by the dwell time of either central or peripheral lines but predominantly by frequent replacement of central and/or peripheral catheters. This may increase the risk of insertion of potential causative pathogens by contaminated catheter hubs or by creation of new entrance sites [25, 26]. However, this variable was not taken into account in the present study.

This study has several strengths; detailed data collection in a multicenter design allowed for a strictly matched case-control comparison and the relatively large sample size allowed to determination of predictive factors per subgroup of causative pathogens. This study has also several limitations that need to be addressed. First, data collection was limited to the first 28 days postnatally, which might have resulted in a lower LOS incidence and mortality rate. Hypothetically, limiting data collection until a postnatal age of 28 days might also result in allocation of infants into the control group while they might have developed sepsis after the defined follow-up period, therefore possibly resulting in an underestimation of the potential risk factors. Secondly, this study contained limited obstetric data. This could hypothetically have influenced the outcome, since maternal factors may also include risk factors for LOS as they have been described to influence the neonatal immune system. Thirdly, center-specific effects could not be excluded from the analyses due to variating LOS incidences, limiting center-based matching. However, this could allow for identification of factors leading to an increased risk for LOS development as a result of local protocols used. Lastly, prolonged parenteral nutrition could also be seen as an early sign of LOS, particularly in less-fulminant CoNS-LOS, rather than a preonset risk factor. However, the relatively large number of LOS cases allowed us to focus on risk factors per pathogen. So, this possible limitation may only account for CoNS-LOS cases. In the case of other pathogens, the course of sepsis is considered to be more fulminant.

In conclusion, since in the current study parenteral feeding was strongly associated with LOS development, it could be hypothesized that reduction of the number of parenteral feeding days might reduce the risk of LOS, which may be achieved by advancement of enteral feeding, preferably with breastmilk. Protective effects of early exposure to specific antibiotics underline the increasing notion that a disturbed microbial colonization may be involved in the pathophysiology of at least a selection of LOS cases.

We thank Dr. Lissenberg-Witte for her excellent help with the statistical analysis and interpretation of the results.

The local institutional review boards of all 9 participating centers granted approval (amendment A2016.363). The parents of all of the included infants gave written informed consent.

The authors have no conflicts of interests to declare.

Dr. el Manouni el Hassani conceptualized and designed this study, coordinated and supervised data collection, carried out the initial analyses, drafted the initial version of this paper, and reviewed and revised this paper.

Dr. Berkhout, Dr. de Boer, and Dr. de Meij conceptualized and designed this study, coordinated and supervised data collection, and critically reviewed this paper for important intellectual content.

Dr. Mann designed the data collection instruments, collected data, and carried out the initial analyses.

Dr. Niemarkt, Dr. de Boode, Prof. Dr. Cossey, Dr. Hulzebos, Prof. Dr. van Kaam, Prof. Dr. Kramer, Dr. van Lingen, Prof. Dr. van Goudoever, Dr. Vijlbrief, Prof. Dr. van Weissenbruch, and Prof. Dr. Benninga critically reviewed this paper for important intellectual content.