Diagnostic markers of serious bacterial infections in febrile infants younger than 90 days old


  • Adi Nosrati,

    1. Department of Pediatrics, Dana-Dwek Children's Hospital, Tel Aviv Sourasky Medical Center, Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel
    Search for more papers by this author
    • These authors contributed equally.
  • Amir Ben Tov,

    Corresponding author
    1. Department of Pediatrics, Dana-Dwek Children's Hospital, Tel Aviv Sourasky Medical Center, Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel
    • Correspondence: Amir Ben Tov, MD, Department of Pediatrics, Dana-Dwek Children's Hospital, Tel Aviv Sourasky Medical Center, 6 Weizman Street, Tel Aviv 64239, Israel. Email: amir.bentov@gmail.com

    Search for more papers by this author
    • These authors contributed equally.
  • Shimon Reif

    1. Department of Pediatrics, Dana-Dwek Children's Hospital, Tel Aviv Sourasky Medical Center, Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel
    Search for more papers by this author

  • This work was performed in partial fulfillment of the MD thesis requirements (N.A.) for the Sackler Faculty of Medicine, Tel Aviv University, Israel.



The aim of this study was to assess correlations between demographic, clinical and laboratory characteristics and the risk of serious bacterial infection (SBI) in febrile <90-day-old infants.


Medical records of all infants younger than 90 days old hospitalized at Dana-Dwek Children's Hospital (2006–2008) for evaluation of fever were retrospectively reviewed. Data on clinical, laboratory and demographic characteristics were retrieved and evaluated.


Forty-eight of the 401 study infants (12%) had SBI: most of them had urinary tract infection (43 infants; 90% of all SBI), three infants had bacteremia, one had bacterial pneumonia and one had bacterial meningitis. Significant independent clinical predictors for the diagnosis of SBI included duration of fever, absence of rhinitis and the absence of lung and skin manifestations. Significant independent laboratory predictors were absolute neutrophil count (ANC), platelets, blood urea nitrogen and C-reactive protein (CRP) level. On receiver operating characteristic curve analysis, the CRP area under the curve (0.819) was significantly superior to ANC and leukocyte count.


Of the clinical and laboratory variables selected for evaluation, qualitative CRP was the strongest independent predictor for diagnosing SBI and a significantly better diagnostic marker than clinical characteristics, ANC and white blood cell count.

The management of febrile infants <90 days old has been a source of much controversy and debate during the past decades. While most of these infants will have no more than a mild viral infection, there is a substantial minority that will be diagnosed as having serious bacterial infection (SBI) at a reported prevalence of 10–14%.[1] Diagnosis in this age group is challenging due to the high prevalence of SBI and the lack of specific signs and symptoms to differentiate SBI from simple viral illness.[1-4] The most commonly accepted approach worldwide is to identify which infants are at high risk of SBI and need to be hospitalized for i.v. empirical antibiotic treatment and which ones are at low risk for SBI and can safely undergo outpatient care with or without antibiotics.[2-5] The efficacy and safety of outpatient management of febrile infants thought to be at low risk for SBI have been reported in three large studies.[2-4] The most widely accepted approach has been to use a combination of historical, physical, and laboratory findings known as the Rochester criteria, which evaluate clinical and laboratory data in order to identify low- and high-risk groups. These criteria include: lack of focal signs on examination; white blood cell count (WBC) between 5000 and 15 000/L; band forms <1500/L; and normal urinalysis.[2]

The idea of a simple, readily available, inexpensive diagnostic marker that yields results within 1 h and accurately identifies bacterial infections in febrile infants is very attractive. Testing for WBC is familiar to emergency physicians as being simple, readily available, yielding prompt results, and inexpensive. Its use has long been the cornerstone of many protocols for distinguishing between low- and high-risk febrile infants.[2-9] Recent studies, however, reported that the contribution of WBC to detection of SBI was only minor,[10-13] and many reports have supported the importance of C-reactive protein (CRP) level as being superior to many other laboratory values, including WBC.[12, 14-23] The objective of the present study was to evaluate clinical and laboratory characteristics of febrile infants <90 days of age and to determine the power of these features in diagnosing SBI.


This study was conducted at Dana-Dwek Children's Hospital, Tel Aviv, a tertiary care university-affiliated medical center, after being approved by the local institutional review board. The study design was a retrospective cohort analysis of data retrieved from hospital records of all febrile infants aged <3 months who met the study inclusion criteria and were hospitalized in the Department of Pediatrics from 1 January 2006 to 31 December 2008. The patients were considered febrile if their recorded rectal temperature had been ≥38°C. Exclusion criteria were preterm birth (<35 weeks of gestation); presence of a chronic disease (heart failure, lung disease or renal failure); congenital or acquired immune deficiency; current antibiotic use; and/or incomplete records.

The retrieved data included demographic information, medical history, physical examination findings, laboratory test results and final diagnosis. The selected demographics were age at admission; gender; country of birth; ethnicity; number of siblings; birthweight; month of admission; and immunization status. Age was calculated electronically from the date of birth and the date of presentation. All infants underwent a thorough full physical examination for localizing the source of the fever (appearance, fontanel, heart, lungs, pharynges, ears, skin and neurological examination). Appearance of illness was defined as a clinical picture on admission consistent with sepsis, including signs such as lethargy, poor perfusion, marked hypoventilation or hyperventilation, cyanosis, convulsion or apnea.[2, 24]

Laboratory investigations included complete blood count (CBC), CRP level, blood chemistry and culture, urine sample analysis and culture, and cerebrospinal fluid sample analysis and culture. Chest radiography was done if respiratory symptoms were present and a lumbar puncture was conducted according to the ward's policy, which was based upon the Rochester criteria (performing a lumbar puncture for every infant under the age of 28 days and for older infants who were at high risk according to these criteria). Urine cultures were obtained by suprapubic aspiration or midstream collection.

The diagnosis of SBI, namely, urinary tract infection (UTI), meningitis, bacteremia and bacterial enteritis, was based on the growth of a known pathogen in culture. Blood culture isolates were considered pathogens if the organism was known to cause disease in healthy infants. Organisms that were considered contaminants included coagulase-negative staphylococci, Bacillus spp., aerobic and anaerobic diphtheroids and viridans group Streptococcus. The diagnosis of pneumonia was based on the presence of typical clinical signs and symptoms and findings on chest radiograph, as defined by the World Health Organization Standardization of Interpretation of Chest Radiographs for the diagnosis of community-acquired pneumonia in children.[25] We analyzed all clinical outcomes, initial choice of antibiotics, any subsequent changes in the patient's clinical status as well as any complications that had developed during the infant's hospitalization.

Statistical analysis

Normally distributed data (e.g. weight, WBC) are expressed as mean ± SD, non-normally distributed data are expressed as median and interquartile range, and categorical variables are reported as percentages. Statistical comparisons were performed using a two-tailed t-test and chi-squared test for all numerical and categorical values, respectively. Significance was set at P < 0.05. Univariate and multivariate logistic regression models were used to identify variables independently associated with the outcome variable, namely, the presence of an SBI (duration of fever, blood urea nitrogen [BUN], absolute neutrophil count [ANC], CRP). The diagnostic performance of the parameters (WBC, ANC and CRP) was evaluated on receiver operating characteristic (ROC) analysis.[26] The data were managed and analyzed using SPSS version 17.0 (SPSS, Chicago, IL, USA).


Demographic variables

During the study period, 450 febrile infants <90 days old were evaluated in the Emergency Department and hospitalized due to fever without source. A total of 401 infants met the inclusion criteria, representing 57.2% of all admissions of infants ≤90 days old during that period of time (Fig. 1). A total of 55.8% of the patients in the present study were boys (n = 224). The mean age of the cohort at the time of presentation to the emergency room was 49.6 ± 18.6 days (9.9% were ≤29 days old; 69.5% were 30–60 days old; and 20.4% were 61–90 days old). Their mean temperature at admission was 38.15° ± 0.5°C.

Figure 1.

Subject enrollment.

Final diagnosis

Febrile illness without source was the most common final diagnosis of the study infants, accounting for 51.8% of all diagnoses. Aseptic meningitis and UTI were the next common diagnosis, accounting for 14.4% and 10.7%, respectively (Table 1).

Table 1. Final diagnosis in febrile infants <3 months old (n = 401)
Final diagnosisn (%)
Febrile illness without source208 (51.8)
Aseptic meningitis58 (14.4)
Urinary tract infection43 (10.7)
Bronchiolitis41 (10.2)
Upper respiratory tract infection16 (3.9)
Otitis media15 (3.7)
Viral diarrhea14 (3.4)
Bacteremia3 (0.7)
Pneumonia1 (0.2)
Bacterial meningitis1 (0.2)
Varicella chickenpox1 (0.2)

The overall rate of SBI during the study period was 12%. There were no cases of enteritis, osteomyelitis or septic arthritis, nor was there any mortality among the infants.

A total of 282 urine specimens were obtained using urinary catheter while 113 urine specimens were obtained by suprapubic aspiration. Of the 43 infants diagnosed with UTI, only 51% had pyuria (positive predictive value [PPV], 75.8%; specificity, 12.5%). The following pathogens were isolated from urine cultures: Escherichia coli (n = 27), Klebsiella pneumoniae (n = 5), Enterococcus (n = 3), Citrobacter (n = 3), Pseudomonas (n = 2) and Staphylococcus aureus (n = 2). The pathogens that were isolated from the blood among the infants with isolated bacteremia were: group B streptococci (n = 1) and Streptococcus pneumoniae (n = 1). S. pneumoniae was also isolated from the blood culture of the infant diagnosed with bacteremia and meningitis.

Table 2 lists the demographic characteristics, vital signs and laboratory findings of the infants with and without SBI. The significant independent predictors for SBI obtained from the medical history and physical examination were duration of fever, absence of rhinitis and absence of lung and skin manifestations. Maximal degree of fever, however, was not found to be a predictor of SBI. Significant independent laboratory predictors were ANC, platelets, BUN and CRP. WBC showed no superiority in identifying SBI (P = 0.086). These risk factors (i.e. duration of fever, ANC, BUN and CRP) were further assessed using logistic regression analysis: only CRP was found to be significantly associated with SBI (P < 0.001) in the univariate logistic regression model. In the multivariate model, again, CRP was the only parameter found to be significantly associated with SBI (odds ratio, 1.042; 95% confidence interval [CI]: 1.028–1.056; P < 0.001). The area under the ROC curve (AUC) was 0.819 for CRP (95%CI: 0.731–0.906; Fig. 2a), 0.588 for ANC (95%CI: 0.489–0.686; Fig. 2b) and 0.574 for leukocyte count (95%CI: 0.477–0.671; Fig. 2c). ROC data were used to select optimal CRP cut-offs for detecting SBI. According to the present data, a CRP cut-off of 20 mg/L was optimal for identifying possible SBI. At that threshold, febrile infants were 4.9-fold more likely to have a bacterial infection, with a sensitivity of 79%, a specificity of 84%, a negative predictive value (NPV) of 97% and a negative likelihood ratio of 0.25 (Table 3).

Figure 2.

Receiver operating characteristic curves for (a) C-reactive protein level (P < 0.001); (b) absolute neutrophil count (P = 0.048); and (c) white blood cell count (P = 0.095) for predicting severe bacterial infection.

Table 2. Subject characteristics vs presence of SBI (mean ± SD)
ParameterInfants without SBI (n = 353)Infants with SBI (n = 48)P
  1. SBI, serious bacterial infection.
M : F57.7:42.543.7:56.20.088
Age (days)49.07 ± 18.851.1 ± 16.70.479
Birthweight (g)3165.5 ± 5353154.1 ± 4670.881
Gestational age (weeks)39.01 ± 5.738.5 ± 1.480.616
Body temperature at admission (°C)38.1 ± 0.638.2 ± 0.790.417
Maximum body temperature (°C)38.3 ± 0.938.3 ± 0.80.746
White blood cell count (cells/μL)12 200 ± 609614 070 ± 69440.086
Absolute neutrophil count (cells/μL)5066.2 ± 4000.56551.6 ± 4523.060.035
Platelets (109/L)446.5 ± 161.9499.7 ± 770.032
Blood urea nitrogen (mg/dL)8.1 ± 3.59.3 ± 2.70.01
C-reactive protein (mg/L)12.6 ± 19.848.5 ± 36.08<0.001
Table 3. C-reactive protein decision thresholds (n = 48)
Threshold (mg/L)Sensitivity (%)Specificity (%)PPV (%)NPV (%)LR+LR−
  1. The prevalence of bacterial infection in the present study was 12%. LR+, likelihood ratio for a positive test; LR−, likelihood ratio for a negative test; NPV, negative predictive value; PPV, positive predictive value.


The management of febrile illnesses in infants <90 days of age varies considerably among physicians.[24] Part of this variability is probably associated with the wide range of management protocols tested and reported throughout the past decades.[2-6] Following the more conservative guidelines, most physicians hospitalize the majority of febrile infants <3 months old and initiate empirical antibiotic therapy. Baker et al. considered that these admission policies increased injudicious exposure of infants to antibiotics, medical costs and iatrogenic complications.[3] The inability to accurately predict the presence of SBI on the basis of clinical impression and routine laboratory tests has led to the investigation of various biomarkers of SBI.[7-12, 14-23] To date, no marker, either alone or in combination, has demonstrated adequate sensitivity or specificity for the diagnosis of SBI in well-appearing febrile infants.

Although the vast majority of the SBI in the present study were UTI, some authors agree that the dipstick as a method of urine analysis is a less reliable screening tool for infants and should not be used to exclude UTI.[27] In the present study the sensitivity, specificity, and PPV of the urine analysis were low. This decreases the significance of pyuria as a screening tool for UTI in this age group.

As of 2005, the routine evaluation of febrile infants at Dana-Dwek Children's Hospital Emergency Department included the measurement of CRP level, thus providing us with the ability to compare CRP to other inflammatory markers. Many algorithms have used peripheral WBC in the evaluation of febrile infants, mostly because of the widespread availability of testing it in both office- and hospital-based settings and its proven value in the evaluation of older infants.[2-7, 9, 28] Surprisingly, however, recent studies have shown only a modest discriminatory power of WBC to identify bacterial infections.[10, 11, 13] In a retrospective evaluation of infants ≤90 days old, Bonsu and Harper reported the usefulness of WBC in identifying young infants most at risk for bacteremia.[23] Those authors found that the use of the traditional WBC cut-off of 15 000/L would have yielded a sensitivity of only 45% and a specificity of 78%. Andreola et al. prospectively tested 408 febrile infants and reported a positive likelihood ratio of 2.08 (95%CI: 1.58–2.75) and a negative likelihood ratio of 0.65 (95%CI: 0.52–0.80) for WBC >15 × 109/L.[20] The present findings were similar: not only did the ability of WBC to predict SBI in febrile infants <90 days of age fail to reach a level of significance, but we demonstrated the absence of predictive power of WBC in identifying febrile infants with SBI, with an AUC of 0.57 (P = 0.095).

The search for a marker whose value increases quickly with a developing infection has led to numerous reports of the utility of CRP and procalcitonin as potential markers of SBI in febrile infants <90 days of age.[9, 12, 14-24] Van den Bruel et al. conducted a systematic review using 14 studies in order to collect all available evidence on the diagnostic value of laboratory tests for the diagnosis of serious infections in febrile children.[1] They found that WBC were less useful for diagnosing an existing SBI and not useful for ruling out SBI. Furthermore, they reported that testing CRP provided the greatest diagnostic value, with a positive likelihood ratio of 3.15 (95%CI: 2.67–3.71) and a negative likelihood ratio of 0.33 (95%CI: 0.22–0.49). Galetto-Lacour et al. also claimed that CRP level was superior to WBC in detecting SBI, with a sensitivity and a specificity of 79%.[16] Riordan and McWilliam reported the importance of serial CRP levels in infants being treated for suspected bacterial infection.[14] Those authors were able to exclude that diagnosis with an NPV of 99% if two CRP measurements that were 24 h apart were <10 mg/L.

In the present study, the test found to have the highest diagnostic value for assessing the risk of SBI in febrile children, compared to all other clinical and laboratory findings, was CRP. Univariate and multivariate logistic regression models found CRP to have the only significant value associated with SBI (P < 0.001). On ROC analysis, the AUC for CRP was significantly superior to both ANC and leukocyte count (AUC, 0.89, 0.588, and 0.57, respectively). In addition, the present data suggest CRP cut-off of 20 mg/L as the optimal figure for predicting the presence of SBI, with a positive likelihood of 4.9, a negative likelihood of 0.25, a sensitivity of 79% and a specificity of 84%.

Recent studies have suggested that procalcitonin is a superior promising screening marker for SBI, with a better sensitivity and specificity than WBC and CRP.[12, 15, 16, 18-20, 22, 23] Maniaci et al. conducted a prospective observational study on 234 infants ≤90 days old and reported that procalcitonin had favorable test characteristics for detecting SBI in young febrile infants (AUC: procalcitonin, 0.92; CRP, 0.75; WBC, 0.66).[11] We believe that although CRP and procalcitonin are both valuable markers for predicting SBI, CRP may be a better screening test in emergency settings because of its wider availability and lower cost.

One limitation of this study was the inclusion of only hospitalized infants who were referred from the emergency department, a factor that may have caused a selection bias. In the Dana-Dwek Children's Hospital Emergency Department, infants ≥6 weeks old who are evaluated according to the Rochester criteria and identified as low risk are discharged and managed as outpatients, and they were not included in this study. Although this was a retrospective study, we had all the information that we needed from the medical records.

Another limitation of this study was the inability to evaluate the Gram staining method as a predictor of SBI. The Gram stain classifies bacteria phenotypically and it is a simple, readily available and inexpensive method that gives immediate information regarding the presence or absence of bacterial disease and can guide initial antibiotic treatment. According to the authors' lab policy Gram stain results are reported only if there is initial bacterial growth in the culture. Because this was a retrospective study, this information was not available.


The present data on hospitalized infants who were evaluated due to fever indicate that qualitative CRP was the strongest independent predictor for existing SBI and that it was highly superior to clinical characteristics, as well as to ANC and WBC. Signs or symptoms suggestive of viral infection, such as rhinitis and rash, were associated with a lower risk of SBI. We suggest the use of CRP as part of the workup for every infant <90 days old who presents to the emergency room with fever.


We thank Esther Eshkol for assistance with the manuscript.