The difficulty in identifying sources of infection in febrile neonates ⩽28 days of age has been recognised for decades.¹ Equally well recognised is that most febrile neonates do not harbour serious bacterial infections and probably would do well without any diagnostic testing or treatment.^{2,3,4,5,6,7,8,9,10,11} For the past decade, however, well publicised recommendations have suggested that all neonates with a temperature ⩾38°C undergo a diagnostic evaluation including a full blood count, urinalysis, lumbar puncture with cerebrospinal fluid (CSF) analysis and bacterial culture, receive parenteral antibiotics (typically ampicillin and cefotaxime), and be admitted to the hospital.^12,13 Although the risk posed to a neonate with an untreated bacterial infection is substantial, the process of the “septic workup” is not undertaken without exposing the infant to other risks such as nosocomial infections and medication errors.¹⁴ Unfortunately, identifying bacterial infections typically requires awaiting culture results and occurs long after the neonate has left the emergency department (ED).

The idea of a simple, readily available, inexpensive diagnostic test that yields results within an hour and accurately identifies bacterial infections in febrile neonates is alluring. If such a test was available, these febrile neonates would likely benefit from immediate parenteral antibiotic therapy whereas those with a negative test could forego antibiotics and, if other discharge criteria were met, perhaps even be discharged. A potential candidate for such a diagnostic test is the peripheral white blood cell (WBC) count. The WBC count is familiar to emergency physicians, simple, readily available, yields prompt results, and is inexpensive. The use of WBC count thresholds has been widely advocated to risk stratify febrile children.^{2,3,5,7–9,12,13} The use of the WBC count to identify bacterial infections in febrile neonates, however, remains unexplored.¹⁵

The objective of this study was to evaluate the discriminatory power of the WBC count to identify bacterial infections in a cohort of febrile neonates who presented to our ED.

METHODS

We retrospectively reviewed the medical records of all neonates ⩽28 days of age who presented to our tertiary care paediatric ED between 1 January 1999 and 22 August 2002 and underwent lumbar puncture during the index visit. An initial screening of all medical records was undertaken to exclude those neonates for whom the medical record was unavailable, no ED triage temperature was recorded, or the ED triage temperature was <38°C. A trained data abstractor then used a standardised data collection form and recorded the following study variables for the remaining neonates: date of birth, date of visit, ED triage temperature, initial WBC count, blood culture results, CSF white and red blood cell counts, CSF culture and bacterial antigen study results, viral study results (including but not limited to respiratory syncytial virus (RSV), rotavirus, enterovirus, and herpes simplex virus), urinalysis and urine culture results, stool culture results, and the radiologist’s final interpretation of a chest x ray if taken. Age was calculated electronically from the date of birth and date of visit. Subjects were to be subsequently excluded if no WBC count had been done in the ED.

We divided the neonates on the basis of the microbiological and radiographic data into four groups:

• those with bacterial infections—that is, positive cultures of blood, urine, CSF, or stool or a clinical diagnosis of cellulitis, fasciitis, omphalitis, osteomyelitis, or mastitis

• those with viral infections—that is, positive viral culture, polymerase chain reaction, or immunofluorescence study

• those with radiographic evidence of pneumonia—that is, infiltrate on chest x ray without meeting the criteria for the bacterial or viral group

• those with no positive findings (“negative sepsis evaluation”).

Statistical analyses included descriptive statistics and the generation of a receiver operating characteristic (ROC) curve. Our data were not normally distributed so we have presented these as medians and interquartile ranges (IQRs). Statistical analyses were performed using Stata 8 (STATA Corporation, College Station, TX). Our institutional review board responsible for monitoring human subject research approved the study.

RESULTS

We identified 206 neonates ⩽28 days of age who had been seen in our ED and had undergone lumbar puncture during the study period. Medical records were available for all neonates and they all had an ED triage temperature recorded. A total of 135 neonates had ED triage temperatures <38°C. Of the remaining 71, two did not have a WBC count done while in the ED. Our main study group consisted of the remaining 69 (31 girls (45%)) neonates. The age, temperature, and WBC counts of these 69 neonates were tabulated. The median age was 17 days (IQR 10–22; range 2–28). The median temperature was 38.4°C (IQR 38.2–38.8°C; range 38.0–39.9°C). The median WBC count was 11.9×10⁹ cells/l (IQR 9.8–15.0×10⁹ cells/l; range 4.3–27.0×10⁹ cells/l).

Eight neonates (12%; two girls) had bacterial infections. Four neonates, aged 15, 16, 17, and 17 days, had urinary tract infections (three had Escherichia coli and one had enterococcus) and their temperatures ranged from 38.4°C to 38.6°C. Two infants aged 4 and 12 days had bacteraemia (one E. coli and one group B streptococcus) with temperatures of 38.2°C and 38.8°C, respectively. A 20 day old neonate with an ED temperature of 38.2°C was diagnosed as having Salmonella meningitis and a brain abscess. It appeared from the medical record that this infant did not have stool studies performed. One 11 day old neonate with a temperature of 38.7°C had a relatively confusing picture. The family initially refused lumbar puncture and antibiotics were administered. On the second day of hospitalisation, the family consented to a lumbar puncture which revealed 12 618 white blood cells and 109 red blood cells/mm³. Subsequent CSF culture, viral polymerase chain reaction testing for herpes simplex virus, and bacterial antigen tests were all negative. This infant’s urine culture revealed 50 000–100 000 colony forming units/ml of mixed E. coli and enterococcus.

We identified viral infections in 10 infants (three girls). RSV was identified in the nasopharyngeal secretions of six infants aged 13, 17, 21, 25, 27, and 27 days whose temperatures ranged from 38.0°C to 38.9°C. One 19 day old infant with a temperature of 38.2°C had rotavirus in stool and three infants aged 9, 14, and 26 days with temperatures of 39.9°C, 38.8°C, and 38.6°C, respectively, had enterovirus (two cases identified in the CSF and one in the nares and conjunctiva). There was no case of herpes simplex virus.

Three neonates were diagnosed as having pneumonia. One 4 day old boy demonstrated a temperature of 38.2°C and a non-specific hazy opacity of the right lung. A 12 day old boy had a temperature of 38.0°C and a right lower lobe patchy infiltrate. A 3 day old girl with a temperature of 38.8°C had bilateral opacities at the lung bases. None of these neonates with radiographic pneumonia had documented bacteraemia or RSV.

The remaining 48 (26 girls (54%)) neonates were categorised in the negative sepsis evaluation group. This group had a median age of 17.5 days (IQR 7.5–22; range 2–28) and a median temperature of 38.4°C (IQR 38.2–38.8°C; range 38.0–39.8°C). All 69 neonates were admitted to the hospital and 68 (99%) received parenteral antibiotics. All were subsequently discharged in good condition.

The WBC counts for neonates with bacterial infections, viral infections, pneumonia, and negative sepsis evaluations had substantial overlap among the groups (fig 1). Given the difficulties in differentiating bacterial from viral pneumonia by radiographic appearance,¹⁶ the three neonates with pneumonia were excluded from the remaining analyses. Of the remaining 66 neonates, the eight neonates with bacterial infections had WBC counts ranging from 11.0×10⁹ cells/l to 19.9×10⁹ cells/l (table 1). The highest positive likelihood ratio occured when a threshold of 17×10⁹ cells/l was chosen (table 2). Febrile neonates in our study with WBC counts ⩾17×10⁹ cells/l were 3.8 times more likely to have a bacterial infection than those with lower WBC counts. At this threshold the WBC was 38% sensitive (95% confidence interval (CI) 9% to 76%) in detecting bacterial infections. The highest threshold at which the WBC count retained 100% sensitivity was 10×10⁹ cells/l. At this threshold, the specificity was 31% (95% CI 20% to 45%). The area under the ROC curve was 0.7231 (95% CI 0.5665 to 0.8797) representing moderate discriminatory power (fig 2).¹⁷

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

 Distribution of white blood cell (WBC) counts by microbiological/radiographic group.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

 Receiver operating characteristic (ROC) curve for white blood cells (WBCs) predicting bacterial infections.

DISCUSSION

The WBC count demonstrated moderate discriminatory power for identifying febrile neonates with bacterial infections in our study population. We were unable to identify an appropriate, clinically useful threshold value that differentiated febrile neonates with bacterial infections from other febrile neonates in the ED. Although a WBC count threshold of 17×10⁹ cells/l yields a positive likelihood ratio of 3.8, nearly two thirds (62%) of febrile neonates with bacterial infections would not be identified using this threshold because the sensitivity is only 38% at this threshold.

Prior studies have examined the utility of the WBC count for evaluating young febrile infants. The applicability of these studies to the clinical practice of evaluating febrile neonates in the ED, however, is problematic. These studies are occasionally from decades ago and included only neonates still in the hospital following delivery^18–20 or infants older than one month of age,^2,10,21–25 or used a specific WBC count range as a component of the inclusion criteria for the study.^3,5,8

One interesting feature of the test characteristics of the WBC count in febrile neonates in our study is the presence of 100% negative predictive value for WBC counts <10×10⁹ cells/l. No febrile neonate with a bacterial infection had a WBC count <10×10⁹ cells/l; 18/69 (26%) febrile neonates had a WBC count <10×10⁹ cells/l. Only one infant in our study demonstrated any degree of neutropenia with a WBC count of 4.3×10⁹ cells/l. Rather than focusing on identifying bacterial infections, future studies could be directed towards examining the potential use of the negative predictive value of the WBC count to identify a subset of neonates for whom it is appropriate to withhold antibiotics or discharge. This may be especially helpful in conjunction with rapid viral testing that is becoming increasingly available in the ED.²⁶

Our study has limitations. The retrospective study design did not allow for the study of the presence of ill appearance, a feature commonly used to risk stratify febrile children.^2,12 We used the performance of a lumbar puncture as an inclusion criterion. We selected this procedure because we wanted to include only those febrile neonates for whom the most complete set of microbiological data was available. This may have led to a selection bias towards sicker neonates. Although we have no practice guidelines regarding the care of febrile neonates in our ED, our experience suggests that a “septic workup” is always or nearly always performed on febrile neonates presenting to our ED. Although the presence of immature neutrophils has been suggested to be associated with bacterial infections in children,^2,3,5,6,18 we elected not to study these—that is, “bands”. The reproducibility of the “band count” or “left shift” has been sufficiently challenged to suggest against the use of immature neutrophils in clinical decision making.^27,28 Our modest sample size resulted in a wide 95% confidence interval for the area under the ROC curve. A larger study would yield a more accurate estimation of the true area under the ROC curve.

SUMMARY AND CONCLUSION

In a sample of febrile neonates evaluated in our ED, the WBC count had only modest discriminatory power in identifying neonates with bacterial infections and demonstrated substantial overlap among groups. Although it is an intuitively appealing possibility that a test as simple as a WBC count would identify bacterial infections in febrile neonates, our data suggest against the use of any WBC count threshold to identify bacterial illnesses in febrile neonates presenting to the ED.