Variations in rates of nosocomial infection among Canadian neonatal intensive care units may be practice-related

The study cohort included all infants admitted to 17 Canadian NICUs prior to 4 days of age. Neonates without a recorded birthweight were excluded. The study cohort was derived from a larger study of 9506 babies admitted to 17 tertiary NICUs in the Canadian Neonatal Network from January 1996 through to October 1997 [10]. These units represent approximately three-quarters of all the tertiary NICU beds in Canada, a nation with >350 000 births annually and almost 30 million people [11, 12]. Information about these neonates was prospectively collected by trained abstractors with standardized definitions and operations as part of the larger study of Canadian NICUs [10].

For the purpose of this study, nosocomial infection (or NI) was defined as one or more positive single organism blood or CSF culture sampled after 48 h of admission in an infant with clinical suspicion of infection. To differentiate between nosocomial and primary (maternal origin) infections, the infant blood culture isolates were required to be different from maternal isolates or to occur at least 7 days after a treated positive blood culture obtained during the first 48 h of life. CSF samples were obtained when indicated according to local practices and policies. The time period of 2 days was chosen to exclude any neonates born with primary infection. Absence of NI included those infants who were never cultured and those with negative blood and/or CSF cultures. An infection episode was defined as a positive culture occurring at least 7 days after a previous treated culture or if the culture isolates were different from the previous culture [13]. We defined nosocomial infections (NI) in this way because blood and CSF are normally sterile, thus a positive culture from one of these sites in an unwell baby is likely to be of serious clinical significance. Additionally, from a practical perspective, these data are well defined and readily abstracted from patient charts.

Other study definitions were in accordance with the Canadian Neonatal Network Data Abstractor's Manual [14]: gestational age was defined as the attending obstetrician's best estimate based on obstetric history and examination, and prenatal ultrasound, except where this was not available or the postnatal physical assessment disagreed with obstetrical estimate by greater than 2 weeks. In that case, the pediatric estimate of gestational age was used instead. Small for gestational age meant being born below the 3^rd percentile for weight (corrected for gestation using growth charts developed by Whitfield et al [15]). Major congenital and chromosomal anomalies were specified in the Abstractor's Manual. Outborn infants were those born at a different hospital from the admitting NICU. The Score for Neonatal Acute Physiology, Version II (SNAP-II) was collected on day 1 as a measure of patient acuity on the day of admission. The SNAP-II is a validated score [16] using 6 empirically weighted physiological measurements made during the first 12 hours after admission to the NICU. To be consistent with previous reports, SNAP-II score was categorized as 0 to 9, 10 to 19, 20 to 29 and greater or equal to 30. Therapy related risk factors were extracted from day-1 and day-3 evaluations, using a subset of relevant variables from the Neonatal Therapeutic Intensity Scoring System (NTISS) [17]. The subset of variables included the use of assisted ventilation, peripheral intravenous (IV) access, central venous access, arterial line access, use of antibiotics (day 1 or 3), gavage feeding, intravenous amino acids, and intravenous fat emulsion.

Univariate and bivariate analyses using SPSS (Chicago, Illinois) were used to describe the study population and to explore the relationships between baseline population risks (gestational age, birth weight, gender, Apgar score at 5 minutes, small for gestational age, use of antenatal corticosteroids, multiple births, outborn status, caesarean section, maternal hypertension, congenital anomalies, admission illness severity) and NI. Chi-square test was used to compare mortality and morbidity among infants with and without NI, for infants with birth weight less than 1500 g and 1500 g or more.

We used a multivariable logistic regression model to develop a risk adjustment model for NI. The outcome variable was NI and independent variables were baseline population risks significantly correlated with NI on bivariate analysis. Correlation matrix was used to examine correlation between variables. The full model was created by adding population risk factors, then SNAP-II score, and finally therapy-related factors – variables were considered to be included if their p-value was <0.10. Backward and forward selections from the full model were applied to help determine the parsimonious model. Finally, centres were compared using adjusted odds ratios and 95% confidence intervals derived from the multivariable logistic regression analysis, using a reference centre (the hospital with the median incidence of nosocomial infection). Sensitivity analyses were conducted by excluding outlier sites with large confidence intervals.

This study was part of a larger study surveying morbidities and mortality in Canada, for which ethics approval for anonymous collation of data was obtained at each individual centre.

16538 infants in the dataset met entry criteria (age less than 4 days on admission). 765 (23.5%) of 3253 VLBW infants, and 329 (2.5%) of 13244 HBW infants (41 were missing birth weight and were excluded), developed at least one episode of NI. Among VLBW infants who developed NI, 78.7% had only 1 episode, 16.2% had 2 episodes, and 5.1% had 3 or more episodes of NI, whereas among HBW infants, 87.9% had 1 episode, 9.3% had 2 episodes, and 2.8% had 3 or more episodes of NI. The median onset of the first infection from day of admission was at 19 days for VLBW infants and 15 days for HBW infants.

Among infants with the first episode of NI, the incidence of CSF culture positive infections was 3.7% in VLBW infants, and 2.8% in HBW infants. The most common organism obtained from blood culture was coagulase negative Staphylococcus (72% of isolates in VLBW, and 57% in HBW). The same was true of CSF isolates (68% and 41% respectively). Gram-negative organisms were the next most common pathogen in VLBW, whereas Gram positives, particularly group B Streptococcus, was the next most prevalent in HBW.

Ninety percent of VLBW infants and 73.3% of HBW infants received at least one course of antibiotics in the NICU on either day 1 and / or day 3, as recorded by NTISS

Among VLBW infants, bivariate analysis showed that infants with NI were more likely to be outborn, delivered by vaginal birth, and have 5 minute Apgar score <7, but less likely to be SGA (see Table 1). Infants with NI also had significantly lower birth weight and gestational age, and higher mean day 1 SNAP-II score, than those without NI. A number of therapy related risk factors were significantly associated with NI: these included use of assisted ventilation, peripheral IV access, central venous access, arterial line access, use of antibiotics (on day 1 and/or day 3), gavage feeding, IV amino acid, and IV fat. For HBW infants, NI was associated with similar infant characteristics, but also with congenital anomalies, 5 minute Apgar score <7, and higher SNAP-II score. Correlation between variables: NI has significant correlation with all variables identified except for antibiotic use. However, when stratified by birth weight, antibiotic use has significant correlation with NI for HBW infants. The Spearman correlation coefficient between CVL and parenteral nutrition is 0.26, which is significant. Furthermore, a cross-tabulation of NI by CVL controlling for IV parenteral nutrition (TPN) indicates significant correlation beween NI and CVL and a contingency coefficient of 0.1 for either TPN present or absent. Therefore, there is no evidence that the presence of TPN or CVL in the models renders the other redundant.

Mortality was more strongly associated with the NI group compared to the group without NI for HBW infants, but not for VLBW infants. Major morbidities (necrotizing enterocolitis, chronic lung disease at 36 weeks post-menstrual age, severe intraventricular hemorrhage/periventricular lesions, and severe stages (3 and 4) of retinopathy of prematurity) were more common in the NI group (Table 2).

In VLBW infants, factors associated with NI in the multivariable logistic model were: gestational age, use of assisted ventilation and IV fat and, and outborn status (see table 3). Ninety-nine percent of infants receiving IV fat also received IV amino acid, so these interventions were combined. SNAP-II scores were combined into ≥10 and 0–9 (the reference) groups because the parameters for the higher score groups [10–19, 20–29, > = 30) appeared to be similar. Compared to infants with a gestational age of greater than 28 weeks, lower gestation was increasingly more likely to be associated with NI. SNAP-II score ≥10 was of borderline significance.

For VLBW infants, cross-tabulation of site and NI showed significant variation between sites (chi-square = 248.7, df = 16, P < 0.001). Crude rates of NI ranged from 6.7% to 74.5% of infants having at least one episode of NI. The crude incidences of NI for 8 hospitals were significantly (p < 0.05) different from the reference hospital (site O) (Figure 1). Site variations persisted after adjustment for baseline patient risk factors (gestational age, admission day SNAP-II score, outborn status) using multivariate logistic regression analysis (Figure 1), with the highest odds ratio being 8.6 (95% confidence interval (CI) 4.1 to 18.2), and the lowest 0.2 (95% CI 0.1 to 0.4) when compared to the reference site (Figure 1).

For HBW infants, significant variation was also present between sites (chi-square = 360.0, df = 16, P < 0.001). Crude rates of NI ranged from 0.1% to 17.0 %. When only site variables were in the regression model, 6 hospitals had significantly different (p < 0.05) NI rates from the reference hospital (F). Significant site differences largely corrected after adjustment for baseline population risks (gestational age, admission day SNAP-II score, outborn status) (Figure 2).

Comparison of Figures 1 and 2 shows that hospitals with low and high incidences of NI among VLBW infants also tended to have lower and higher incidences of NI among larger infants. When sites P and N were excluded from the analysis, site A was also found to have significantly higher incidence of NI than the median hospital for both VLBW and HBW infants. (Figures 3 and 4)

Holly Bavinton; Stella Karuri; Zhenguo Qiu; Sarka Lisonkova, Shawn Stewart.