Late-Onset Sepsis in Very Low Birth Weight Neonates: The Experience of the NICHD Neonatal Research Network

Population Database

The NICHD Neonatal Research Network is a consortium of tertiary neonatal centers. The Network maintains a registry of all VLBW infants (401–1500 g) born and/or admitted to participating centers within 14 days of birth. This registry was developed to describe the populations at each participating center, to survey neonatal practice, to assess morbidity and mortality rates, and to provide information for the planning of randomized clinical trials.¹¹ Trained research nurses collect maternal demographic, pregnancy and delivery data soon after birth and infant data until 120 days, discharge, or death.

The registry includes data on rates of late-onset sepsis, infecting organisms, and antibiotic therapy. In September 1998, infection surveillance was expanded to include detailed data on maternal intrapartum antibiotic use, results of all blood and cerebrospinal fluid cultures, C-reactive protein (CRP) results, timing and rates of infection, antibiotic use, indwelling lines, and other risk factors for infection. Blood cultures were processed by the clinical microbiology laboratories at each academic center using either the Bactec (Becton Dickinson, Sparks, MD) or BacT/Alert (Organon Teknika, Durham, NC) systems. One-half to 1 cc of blood was inoculated per blood culture.

Late-onset sepsis was defined as a positive result on 1 or more blood cultures obtained after 72 hours of life. Blood cultures positive for organisms generally considered to be contaminants, including corynebacterium, propionibacterium, penicillium, and diphtheroids were excluded from analysis. Records of patients with blood cultures positive for coagulase-negative staphylococci (CONS) were reviewed. The following definitions were used to distinguish definite or possible CONS infections from contaminants: definite infection: 2 positive blood cultures drawn within 2 days of each other or 1 positive blood culture and elevated CRP (>1) within 2 days of blood culture; possible infection: 1 positive blood culture and patient treated with vancomycin, oxacillin, or a semisynthetic antistaphylococcal agent for 5 or more days; probable contaminant: 1 positive culture without an elevated CRP or antibiotic therapy as outlined above. Patients with definite or possible CONS sepsis were included in the analysis. It is acknowledged that this definition may overestimate true rates of CONS infection. For all other pathogens, sepsis was defined by the presence of the organism in the blood culture.

An episode of sepsis was defined if a patient had a positive blood culture treated with antibiotics for 5 or more days or treated for a shorter period if the patient died. If the same organism was cultured after 10 days of appropriate antibiotic therapy or if a different organism was cultured from a subsequent culture, this was considered an additional episode. The focus of this study is on the first episode of late-onset sepsis.

Gestational age (GA) was determined by best obstetric estimates. Intraventricular hemorrhage (IVH) was graded according to the method of Papile et al¹² Necrotizing enterocolitis (NEC) was classified according to the system of Bell et al¹³ For this analysis, respiratory distress syndrome (RDS) was defined by having each of the following: 1) need for oxygen at 6 to 24 hours of life, 2) clinical features of RDS within first 24 hours, 3) need for respiratory support (continuous positive airway pressure or ventilation) to age 24 hours, and 4) abnormal chest radiograph within first 24 hours. Chronic lung disease/bronchopulmonary dysplasia (BPD) was defined by an oxygen requirement at 36 weeks’ postmenstrual age. The clinician’s presumed cause of death was recorded for all infants who died. Coded causes of death related to infection include the following: proven sepsis/infection, suspected sepsis/infection, RDS with infection, BPD with infection, NEC with sepsis, and other (infection-related).

The participating centers and the numbers of patients each contributed to the study were: Yale University, 258; University of New Mexico, 309; Case Western Reserve University, 329; Stanford University, 336; University of Texas Southwestern Medical Center, 343; Wayne State University, 347; Emory University, 379; University of Tennessee at Memphis, 393; University of Miami, 412; Brown University, 426; University of Texas-Houston, 448; Harvard University, 462; University of Alabama, 504; Indiana University, 619; and University of Cincinnati, 650.

Statistical Methods

All analyses were based on the first episode of late-onset sepsis. Late-onset sepsis was treated as a binary variable. Associations between late-onset sepsis and maternal and neonatal variables, hospital course and death were explored. Statistical significance for unadjusted comparisons (eg, the variation in incidence of infection among clinical centers) was determined by the χ² or Fisher exact test. When characteristics or conditions were considered antecedent or concurrent with late-onset sepsis (eg, maternal variables) adjusted relationships were evaluated using logistic regression models fit to the binary sepsis outcome. When sepsis was considered antecedent, adjusted relationships were evaluated by logistic regression if the outcome was binary (eg, the relation of sepsis to death) and by linear regression if the outcome was continuous (eg, the relation of sepsis to number of days in the hospital). The relation of variables other than GA and birth weight to late-onset sepsis were evaluated after adjusting for GA and clinical center, with other covariates included as noted in the text and footnotes to the tables. All analyses were completed using SAS software (SAS Institute, Cary, NC).¹⁴

Incidence of Infection

Between September 1, 1998, and August 31, 2000, 6956 VLBW neonates were admitted to Network centers. Because this study reviews infections occurring after 72 hours of life, neonates who died within the first 3 days of life were excluded from the analysis. The results presented are based on 6215 neonates who survived at least 3 days.

Sixty-two percent of infants (3856/6215) had 1 or more blood cultures obtained after day 3. Infants had between 1 and 39 late blood cultures obtained (24% had a single blood culture; 18% had 2; 13% had 3; and 45% had 4 or more cultures). A total of 17 712 blood cultures were obtained after day 3, of which 3584 (20%) were positive.

Twenty-five percent of infants (1536/6215) had at least 1 positive blood culture after day 3. We excluded the culture results of 168 infants with a single blood culture positive for an organism considered to be a contaminant (see definitions above) and 55 infants with a single positive blood culture that grew >1 organism (dual infections). The analyses we report are based on 1313 infants with late-onset sepsis—ie, 34% of all infants evaluated with a blood culture (1313/3856) or 21% of all infants studied (1313/6215). The majority of infants had a single episode of late-onset sepsis (943/1313, 72%); 20% (262/1313) had 2 episodes; 6% (76/1313) 3 episodes, and 2% (32/1313) 4 or more episodes. Infants with the lowest birth weights were more likely to have multiple episodes (401–500 g, 38%; 501–750 g, 39%; 751-1000 g, 25%; 1001–1250 g, 21%; 1251–1500 g, 13%; P < .001). The focus of this study was on the first episode of late-onset sepsis. There was considerable center-to-center variability in the incidence of late-onset sepsis, with rates ranging from 10.6% to 31.7% of VLBW infants at study sites, (P < .001). This difference remains significant after adjusting for birth weight, GA, race, and sex.

Pathogen Distribution

The vast majority (70%) of first episode late-onset infections were caused by Gram-positive organisms (Table 1). CONS were the most common late-onset pathogens (48% of all infections, 68% of Gram-positive infections). Using the definitions outlined above, 276 (44%) CONS were definite infections and 353 (56%) were possible infections. If possible infections are excluded, CONS still remains the single most common agent isolated (276/1313, 21%). Other Gram-positive organisms included Staphylococcus aureus, Enterococcus spp, and group B Streptococcus. Gram-negative pathogens accounted for 18% of first episode infections. Escherichia coli, Klebsiella, Pseudomonas, Enterobacter, and Serratia were the most frequent Gram-negative pathogens. Fungal organisms were responsible for 12% of first episode late-onset sepsis. Of all pathogens, Candida albicans was the third most frequent organism isolated (76/1313, 6%).

The average age for the first episode of late-onset sepsis was 22 ± 0.5 days (median: 17 days; 75th percentile: 28 days; 95th percentile: 57 days). The age for first infection did not differ significantly by infecting pathogen (data not shown).

Maternal History

Maternal demographic data and clinical histories were reviewed. After adjustment for GA and clinical center, there were no statistically significant associations between infection and maternal age, marital status, prenatal care, multiple gestation, hypertension or preeclampsia, duration of rupture of membranes, presence of labor, mode of delivery, or antenatal antibiotic or steroid therapy (data not shown).

Neonatal History

Birth weight and GA were strongly associated with risk of late-onset sepsis. Forty-three percent of neonates with birth weights of 401 to 750 g had at least 1 episode of blood culture-proven late-onset sepsis. The rate of infection decreased with increasing birth weight to 28% for infants 751 to 1000 g, 15% for infants 1001 to 1250 g, and 7% for those 1251 to 1500 g (P <.001). Similarly, the infection rate was inversely related to GA. Almost half (46%) of the neonates born at <25 weeks’ gestation developed late-onset sepsis. This rate of infection declined to 29% at 25 to 28 weeks, 10% at 29 to 32 weeks, and only 2% for infants born after 32 weeks (P < .001). Rates of first episode infection per 1000 hospital days also decreased with increasing birth weight and GA (Table 2). After adjusting for GA, birth weight, and clinical center, there was no relationship between either race or sex and the risk of late-onset sepsis (percentage of neonates with infection by race and sex: black, 23%; Hispanic, 23%; white, 19%; other, 15%; P = not significant [NS]; male, 22%; female, 21%; P = NS).

Most VLBW neonates have complex medical problems and prolonged hospitalization. Infants with late-onset sepsis had a longer length of hospitalization (79 versus 60 days; P < .001). This difference remained significant after additional adjustment for IVH, BPD, and NEC in infants hospitalized >28 days (99 versus 89 days; P <.001). There were no associations between late-onset sepsis and RDS, surfactant use, or severe IVH/periventricular leukomalacia (data not shown). However, patients with PDA (34% vs 16%, odds ratio [OR]: 1.2 [1.0–1.4]; P = .04), BPD (41% vs 16%, OR: 1.3 [1.0–1.5]; P =.02), and NEC (50% vs 19%, OR: 2.7 [2.1–3.4]: P <.001) had a higher incidence of late sepsis.

Enteral feeding history and the need for prolonged intravenous access reflect severity of illness. Our data document the widespread use of central catheters among VLBW infants; 46% of all infants had a percutaneously inserted central catheter (PICC) at some time during their hospital course, 15% had a peripheral arterial line (PAL), and 8% had a surgically placed central line (CVL). Umbilical lines were also used frequently: 51% had an umbilical arterial (UAC) and 40% had an umbilical venous catheter (UVC). The risk of late-onset sepsis increased as the duration of central lines and parenteral hyperalimentation increased (Tables 3 and 4). The earlier an infant was started on enteral feedings, the earlier he or she reached full enteral feedings (ie, supplemental intravenous fluids <20 μL/kg per day), and the earlier birth weight was regained, the less likely he or she was to develop infection.

The risk of infection increased in patients with increasing duration of ventilator support (Table 3). Half of the infants who were ventilated for >28 days developed late-onset sepsis versus only 9% of those who were ventilated for 1 week or less. Moreover, infants who developed late-onset sepsis had a significantly longer duration of mechanical ventilation than those without late-onset sepsis (24 vs 12 days; P < .001, Table 4). This difference remained significant after additional adjustment for IVH, BPD, and NEC in infants hospitalized >28 days (32 vs 25 days; P < .001). In a multivariate model that evaluated various risk factors for infection (GA, clinical center, central line days-PICC, UVC, UAC, surgical CVL, PAL-hyperalimentation days and ventilator days) the following remained significant: GA, center, surgical CVL, PAL, hyperalimentation, and ventilator days (each P <.01).

Antibiotic/Antifungal Therapy

Fifty-six percent of all infants (3459/6215) received at least 1 course of antibiotics started after day 3. The rate of antibiotic use was inversely related to birth weight and GA. Infants who developed infection were more likely to receive antibiotics after day 3 (100% vs 44%; P <.001). The most frequently used antibiotics were gentamicin, vancomycin, cefotaxime, ampicillin, and tobramycin. Forty-four percent of all infants (2712/6215) were treated with vancomycin after day 3. Vancomycin use was inversely related to birth weight (401–500 g, 78%; 501–750 g, 75%; 751-1000 g, 60%; 1001–1250 g, 36%; 1251–1500 g, 18%). Although significantly more infants with late-onset sepsis were treated with vancomycin (93%), 30% of patients without proven infection also received this drug (P < .001).

Nine percent of all patients (578/6215) received antifungal therapy. Use of antifungal drugs was inversely related to birth weight (401–500 g, 34%; 501–750 g, 28%; 751-1000 g, 10%; 1001–1250 g, 4%; 1251–1500 g, 2%). Amphotericin B was the most frequently used drug (85% of all doses), followed by fluconazole (10% of all doses).

Mortality

Among this cohort of 6215 VLBW infants who survived >3 days, 597 infants (10%) died. Infants who developed late-onset sepsis were significantly more likely to die than those who did not (18% vs 7%; P <.001). GA, study center, and male sex were all highly significant predictors of death (each, P <.001). The increased risk of death for infants with late-onset sepsis remained after adjustment for GA, study center, and sex.

Overall 242 (18%) of the 1313 infants with late-onset sepsis died. We evaluated the last positive blood culture before death. Forty-five infants (19%) died on the same day the blood culture was obtained; 54 (22%) died on the following day; 19 (8%) died on day 3; 29 (12%) died between days 4 and 7; and 95 (39%) infants died >7 days after the last positive culture was obtained. Infants with Gram-negative organisms were more likely to die acutely—ie, within 3 days of the positive blood culture (Table 5). The clinician’s presumed cause of death was evaluated for each patient with late-onset sepsis who died. Infants who died within 3 days of a positive blood culture were more likely to have a death attributed to infection (87%) than were infants who died later (4–7 days, 71%; >7 days, 46%). Infants with Gram-negative sepsis were more likely to have deaths attributed to infection (81%) than were those infected with other agents (fungi, 73%; non-CONS Gram-positives, 61%; and CONS, 49%).

There were significant differences in death rates with different organisms isolated from the last positive culture before death (Table 6). VLBW infants with late-onset Gram-negative sepsis were at greatest risk of death (36% died; OR for death, Gram-negative vs other organisms, 3.5 [2.5–4.9]; P <.001). Within this high-risk group, infants with Pseudomonas infections were most likely to die (74% of infants with Pseudomonas infections died). Infants with late-onset fungal sepsis were also at increased risk of death compared with infants with other late-onset infections (32% died; OR, fungi vs other organisms, 2.0 [1.3–3.0]; P <.01). Overall, VLBW infants with late-onset Gram-positive sepsis were not more likely to die than infants who were uninfected (11% died; OR for death, Gram-positive vs uninfected 0.9 [0.7–1.2], P = NS). This is primarily attributable to the large number of infections with CONS (9% died; OR for death, CONS vs uninfected 0.8 [0.4–1.4]; P = NS). Although numbers were small and differences not statistically significant, patients infected with S aureus or with group B streptococci were more likely to die than patients who were uninfected (17% and 22%, respectively, vs 7%).