BACKGROUND: A complete blood count (CBC) with white blood cell differential is commonly ordered to evaluate newborns at risk for sepsis.
OBJECTIVES: To quantify how well components of the CBC predict sepsis in the first 72 hours after birth.
METHODS: For this retrospective cross-sectional study we identified 67 623 term and late-preterm (≥34 weeks gestation) newborns from 12 northern California Kaiser hospitals and 1 Boston, Massachusetts hospital who had a CBC and blood culture within 1 hour of each other at <72 hours of age. We compared CBC results among newborns whose blood cultures were and were not positive and quantified discrimination by using receiver operating characteristic curves and likelihood ratios.
RESULTS: Blood cultures of 245 infants (3.6 of 1000 tested newborns) were positive. Mean white blood cell (WBC) counts and mean absolute neutrophil counts (ANCs) were lower, and mean proportions of immature neutrophils were higher in newborns with infection; platelet counts did not differ. Discrimination improved with age in the first few hours, especially for WBC counts and ANCs (eg, the area under the receiver operating characteristic curve for WBC counts was 0.52 at <1 hour and 0.87 at ≥4 hours). Both WBC counts and ANCs were most informative when very low (eg, the likelihood ratio for ANC < 1000 was 115 at ≥4 hours). No test was very sensitive; the lowest likelihood ratio (for WBC count ≥ 20 000 at ≥4 hours) was 0.16.
CONCLUSION: Optimal interpretation of the CBC requires using interval likelihood ratios for the newborn's age in hours.
WHAT'S KNOWN ON THIS SUBJECT:
Components of the complete blood count (CBC) provide information about the likelihood of sepsis in newborns, but previous studies have used varying definitions of abnormal and yielded inconsistent results.
WHAT THIS STUDY ADDS:
White blood cell counts and absolute neutrophil counts increase the probability of sepsis only when they are low. The informativeness of the CBC increases with age and when interval likelihood ratios are used rather than a “normal” range.
The evaluation of newborns for possible early-onset sepsis is difficult because risk factors for infection are common and early signs and symptoms are nonspecific. When newborns are symptomatic or have significant risk factors, a complete blood count (CBC) is commonly used to help assess the likelihood of infection and the need for antibiotics. In fact, Centers for Disease Control and Prevention guidelines for the prevention of early-onset group B Streptococcal infection,1 endorsed by the American Academy of Pediatrics,2 recommend a CBC for certain high-risk infants, such as those whose mothers were positive for group B Streptococcus (GBS) but were not adequately treated with antibiotics.
The Centers for Disease Control and Prevention recommendations do not provide guidance on how to use the results of the CBC to estimate the likelihood of infection. Published reference ranges vary widely for components of the CBC, including the total white blood cell (WBC) count, the absolute neutrophil count (ANC), and the proportion of total neutrophils that are immature (I/T).3,–,7 Total and differential white blood cell counts are affected by many factors besides infection, including infant age in hours,3,–,6 the method of blood sampling,8,–,10 the method of delivery,11,–,13 maternal hypertension,4,14,15 and the infant's gender.12 It is not surprising that many different values for the sensitivity and specificity of different components of the CBC as predictors of infection have been published, depending on the population studied and what levels of these tests were considered “abnormal.”
A significant limitation of previous studies is that they have generally dichotomized each of the CBC results (ie, classified results as normal or abnormal). In addition to the problem of inconsistency of such classification systems across studies and populations, this wastes information by failing to quantify the difference between borderline and profoundly abnormal results.16 Most studies have had small numbers of subjects with culture-confirmed infections, leading to imprecise estimates of test characteristics. Finally, whether or how to take into account factors other than infection that affect white blood cell counts has not been evaluated. For the current study we took advantage of information systems available in the Northern California Kaiser Permanente Medical Care Program (KPMCP) and Brigham and Women's Hospital (BWH) to investigate (1) whether accounting for other predictors of WBC count, ANC, and I/T, such as type of delivery, maternal hypertension, and infant gender, would improve performance of these tests for predicting infection, (2) interval likelihood ratios (LRs)16 for various ranges of normal and abnormal results, and (3) how these LRs vary with age at the time of the CBC.
PATIENTS AND METHODS
The study was approved by the KPMCP, BWH, and the University of California, San Francisco, institutional review boards for the protection of human subjects.
We obtained data for this cross-sectional study from KPMCP and BWH demographic, laboratory, and hospitalization databases. We queried microbiology databases to identify all infants for whom a blood culture was obtained at <72 hours of age. We kept the first positive blood culture for infants with positive cultures, and the first blood culture for other infants, and then matched all blood cultures by date and time to the (single) CBC obtained closest in time to the blood culture for each infant.
Newborn infants were eligible for the study if (1) they were born between January 1, 1995, and September 30, 2007, at a KPMCP hospital that had at least 100 total births in that time period, or at the BWH from January 1, 1993, through December 31, 2007, (2) their estimated gestational age was ≥34 weeks, and (3) they had a CBC and blood culture drawn within 1 hour of one another at <72 hours of age. CBCs and blood cultures were drawn according to the protocols and clinical judgment of clinicians at each site; the timing was often based on convenience (eg, when the infant was in the nursery) as well as medical indications. In many cases, the infants were asymptomatic and the tests were obtained because of maternal risk factors.
We obtained maternal data from the electronic record, including method of delivery (vaginal versus cesarean) and diagnoses of preeclampsia (International Classification of Diseases, Ninth Revision17 codes 642.3–642.7) and chorioamnionitis (code 658.4X). Except for the infants with positive blood cultures whose paper medical charts we reviewed, we did not have data on other maternal risk factors (eg, fever or length of rupture of membranes) or infant symptoms for this study because these were not yet in an electronic medical chart. CBCs were completed with Beckman-Coulter (Brea, CA) or Sysmex (Munderlein, IL) hematology analyzers at KPMCP hospitals and an Advia 120 automated hematology analyzer (Slemens USA, Deerfield, IL) at the BWH. The differential WBC count, which allowed estimation of the ANC and I/T, was estimated manually with a mean of 100.0 cells (SD: 0.9) to allow for identification of bands. The ANC was calculated as the automated estimate of the WBC count × (% segmented neutrophils + % bands)/100. I/T was calculated as the total number of immature neutrophils (promyelocytes, myelocytes, metamyelocytes, and bands) divided by the total number of cells in the neutrophilic cell line (immature plus segmented neutrophils).
We classified all blood cultures as positive or not without regard for CBC results by using an algorithm based primarily on the organism risk category and the time to culture positivity.18 Three infants whose blood cultures only grew coagulase-negative Staphylococcus were included because they had significant symptoms. Three blood cultures were positive for more than 1 organism.
We compared demographic and clinical characteristics of infants with and without infection by using t tests or χ2 tests as appropriate. To determine if adjusting the WBC count, ANC, and I/T for other variables would improve discrimination, we estimated predicted values of these tests using multiple linear regression models on the infants who did not have infection. Models included age (using linear splines with knots at 3, 6, 9, 12, and 24 hours), age squared, birth weight (in 5 categories), gender, birth facility, year of birth, maternal preeclampsia, cesarean delivery, and 5-minute Apgar score (dichotomized at <7). We subtracted these predicted values from observed values and visually inspected receiver operating characteristic (ROC) curves for these residuals and raw test results to ensure that their slopes were monotonically decreasing. We then compared the ability of the residuals with the raw test results to discriminate between newborns who did and did not have infections by comparing areas under ROC curves in 3 time periods.
We estimated percentiles according to age for the WBC count, ANC, and I/T using quantile regression (the qreg procedure in Stata [Stata Corp, College Station, TX]) for tests performed at <24 hours after birth, with terms for age, age squared, and linear splines with knots at 3, 6, 9, and 12 hours.
To estimate LRs for the WBC count, ANC, and I/T, we selected cutoffs that were round numbers and that generated 5 strata with approximately equal numbers of cases. We chose this method rather than identifying inflection points in observed ROC curves to avoid overfitting. We similarly stratified according to age at the time of sampling. We estimated LRs as the proportion of bacteremia cases, with each result divided by the proportion of noncases with that result.16 We performed all analyses by using Stata/SE 11, supplemented by a Stata algorithm that we developed to calculate interval LRs (available from the authors).
Of the 550 367 infants eligible for the study on the basis of their hospital, year of birth, and gestational age, we identified 311 (0.57 of 1000 live births) with positive blood cultures as defined above. We included in this study the subset of 67 623 infants (12.3% of the 550 367 eligible newborns) who had a CBC performed within 1 hour of a blood culture, including 245 of the 311 whose blood culture was positive (3.6 of 1000 infants who received CBCs).
Characteristics of the newborns included in the study and the proportions with documented infections are shown in Table 1. The majority of infants had white mothers, were born at Kaiser, were ≥39 weeks' gestation, and had their CBCs measured in the first 8 hours after birth. Infants were at higher risk of infection if they were born by cesarean delivery, born in the earlier years of the study, had 5-minute Apgar scores of <7, or had their CBCs measured at <1 hour of age.
The organism most commonly identified in the positive cultures (Table 2) was GBS (56%), followed by Escherichia coli (22%) and Enterococcus faecalis (4%). The proportion with GBS infections was lower at BWH (49% vs 59%) and declined over the years of the study, from 67% of infections to 37%. As a result, the proportion of infections caused by E coli increased slightly (from 19% to 28%), although the rate of E coli infections in each time period was similar.
Results for components of the CBC are shown in Fig 1. Because 95% of the infants with infections had their CBCs measured at <24 hours, we restricted the figures to that time period. The figures show a bimodal distribution in timing of CBCs, with the first peak shortly after birth, and a second peak at ∼4 hours. The median WBC count and ANC values rise after birth, peak at 6 to 8 hours, and then decline slightly during the next 18 hours. In contrast, the median I/T declines slightly and approximately linearly during the first day. Among newborns with infection, mean WBC counts were 29% lower, mean ANCs were 39% lower, and I/Ts were 133% higher than in newborns without infection; platelet counts did not differ significantly.
Because the WBC count, ANC, and I/T may be influenced by many factors other than sepsis, we created new variables for each test that corresponded to the difference between that test result and what would be predicted on the basis of other factors available in this data set that were known or observed to influence test results (see “Methods”). Resulting areas under ROC curves, stratified according to age, are shown in Table 3. Overall and for each age stratum, the differences in predictive ability with and without the additional factors were generally small in practical and statistical terms; we therefore elected to continue with raw test results for the remaining analyses.
There was a dramatic improvement in the discrimination of WBC count, ANC, and I/T during the first 4 hours after birth (Fig 2, which includes all of the data, not just results from the first 24 hours). This occurred similarly for GBS and non-GBS infections; sample sizes were not sufficient to consider other organisms separately. The platelet count did not predict infection, regardless of age. Both the WBC counts and ANCs provided little information about infection risk in the first hour, with only very low counts being informative. The WBC counts and ANCs improved significantly at 1 to 4 hours, and improved more at ≥4 hours. In contrast, I/T did provide some information in the first hour, although it also discriminated better after 4 hours. The slopes of all ROC curves monotonically decrease until close to the upper right corner, where the slope increases a little, but remains <1. Because the slope of the ROC curve is equal to the LR, this shows that both WBC counts and ANCs were associated with a higher risk of infection only when they were low. For both WBC counts and ANCs, however, very high values, although not worrisome, are also not reassuring. The LR for WBC count > 30 000 for all ages was 0.844 and the LR for ANC > 25 000 for all ages was 0.935).
LRs for WBC counts, ANCs, and I/T at <1 hour, 1 to <4 hours, and ≥4 hours are shown in Table 4 (we present point estimates of LRs; 95% confidence intervals are available on request). LRs for platelet counts are not shown; whether dichotomized at 150 or divided into 5 categories, no level of the platelet count had an LR significantly different from 1. The lowest LRs for the other tests, corresponding to the most normal results at ≥4 hours of age, are in the range of 0.15 to 0.3, which reflects the limited sensitivity of these tests for diagnosing sepsis, and hence their limited ability to be reassuring. In contrast, the LRs for the most abnormal tests are high, in the range of 50 to 120 for the lowest values of the WBC count and ANC.
This large study of components of the CBC as predictors of bacterial infection in newborns provides several results that should be clinically useful in the early evaluation of newborns for possible infection. First, although many factors other than infection affect WBC values in the first several hours after birth, adjusting a normal range to account for these factors led to little improvement in the discrimination of the components of the CBC. Second, the usefulness of the CBC in distinguishing between infants who do and do not have infections increases dramatically during the first 4 hours after birth. Third, rather than attempting to dichotomize CBC results into those in and outside of a “normal range,” it is more informative to use the age-specific LRs associated with intervals of specific test results. Finally, related to this last point, in this age group using a normal range with upper limits for the WBC count and ANC will lead to newborns with high WBC count and ANC being labeled as having “abnormal” results, when in fact only low WBC count and ANC are associated with increased likelihood of infection.
Our results confirm and extend those of previous studies. Multiple studies have shown that the WBC count and ANC increase rapidly during the first 6 hours, leveling off thereafter.3,–,5,12 Our fifth and 50th percentiles closely match those for term infants reported by Schmutz et al,12 who used a similar design by using data systems at Intermountain Health care. Our finding that the test characteristics of the CBC improve with age is in agreement with studies that reveal poor performance of the CBC when it is used primarily in young (<4 hours old) infants,5,7,19,–,21 including some who specifically show improvement with later CBCs.19,20
Our study has important limitations. First, this is a study of early-onset sepsis in term and late-preterm infants. Most (75%) of the CBCs were obtained in the first 8 hours after birth, including 54% obtained in the first 4 hours. In older infants, not only low but also high WBC counts and ANCs (and low platelet counts) may be suggestive of infection. Thus, these results should not be generalized to infants older than ∼12 hours of age. Similarly, WBC counts and ANCs are normally lower in more premature infants12; this would presumably worsen discrimination in that group.
Second, this study included only infants who had CBCs and blood cultures measured by their treating clinicians, either because of risk factors for infection or symptoms. As a result, odds ratios reported in Table 1 should not be used to infer risk relationships in the general population of infants. For example, although preterm delivery is a risk factor for infection in newborns,22,–,24 the fact that preterm infants were more likely to be selected for CBCs and blood cultures masked the relationship between gestational age and infection in this study, as evidenced an the odds ratio of 0.61 for infants with a gestational age of 34 to 36 weeks compared with infants with a gestational age of ≥39 weeks.
Third, although this was a large study, the need to stratify on both age at the time of the CBC and intervals of test results meant that each LR we reported was based on a smaller sample size, leading in some cases to wide confidence intervals around the LRs. Because of this need to stratify according to age, we were unable to estimate joint LRs for specific combinations of CBC results (eg, an LR for an ANC of 2000 to 4999 in an infant with an I/T of 0.3–0.44). Although it may be tempting to use the product of the interval LRs for each test in Table 4 as a joint LR, this approach may not be valid because (particularly in the case of the WBC count and ANC) the tests cannot be assumed to be independent.
Finally, it is possible that the LRs for these tests vary not only with the age of the infant but also with the distribution of organisms,25 gestational age, and other risk factors, and level of symptoms in those with infection. Thus, optimal prediction of sepsis will require additional studies to better define these relationships and at least a moderately complicated algorithm that would best be implemented within an electronic medical chart. In future analyses we plan to explore the best way to combine results of clinical and laboratory data to estimate the probability of sepsis, and how the process might be automated.
CONCLUSIONS AND CLINICAL IMPLICATIONS
Although additional refinements are needed, this study has important clinical implications. Perhaps most importantly, results of the CBC can provide more information about the risk of sepsis after the first few hours after birth. If the intent of drawing a CBC is to use that information to make clinical decisions regarding the likelihood of infection, for example, whether to initiate empiric antibiotic therapy, then delaying the CBC for a few hours may be advisable. On the other hand, if an infant's risk factors or symptoms are sufficiently worrisome to draw a CBC and blood culture before ∼4 hours of age, it may be prudent to start antibiotics at the same time, rather than waiting for the results of the initial CBC.
Among infants for whom a decision to start antibiotics can be deferred until after 4 hours of age, the CBC is more likely to be helpful. However, most of these infants will have CBC results with modest LRs, in the range of 0.2 to 5. Thus, even when the CBC is optimally interpreted, decisions about antibiotic treatment should remain highly dependent on maternal risk factors and newborn symptoms of infection.5 Future approaches to the sepsis evaluation will require models that are explicit about how multiple information sources (maternal risk factors, newborn clinical examinations, and laboratory test results) are integrated.
This work was funded by National Institute of General Medical Sciences grant R01-GM-80180-3.
We thank Juan Carlos LaGuardia for creating a Stata ado file to calculate LRs, Sherian Xu Li for SAS programming, Amy Zolit for chart review and data abstraction, Manuel Chinchilla and Issa Alaweel for database construction and management, Paul Hughes, MBA, and Gregory Tomilonus for providing hospital demographic information, and Stella Kourembanas, MD, Steven A. Ringer, MD, PhD, and Dennis L. Kasper, MD, for their ongoing support of our work.
- Accepted July 20, 2010.
- Address correspondence to Thomas B. Newman, MD, MPH, Department of Epidemiology and Biostatistics, University of California, San Francisco, Box 0560, San Francisco, CA 94143. E-mail:
This work was presented at the annual meeting of the Pediatric Academic Societies; May 2, 2009; Baltimore, MD.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
- CBC =
- complete blood count •
- GBS =
- group B Streptococcus •
- WBC =
- white blood cell •
- ANC =
- absolute neutrophil count •
- I/T =
- proportion of neutrophils that are immature •
- KPMCP =
- Northern California Kaiser Permanente Medical Care Program •
- BWH =
- Brigham and Women's Hospital •
- LR =
- likelihood ratio •
- ROC =
- receiver operating characteristic
- 2.↵American Academy of Pediatrics. Practice guideline endorsement. Prevention of perinatal group B streptococcal disease: revised guidelines from CDC2002. Available at: http://aappolicy.aappublications.org/misc/Prevention_of_Perinatal_Group_B.dtl. Accessed December 7, 2009
- Escobar GJ,
- Li DK,
- Armstrong MA,
- et al
- Jackson GL,
- Engle WD,
- Sendelbach DM,
- et al
- Newman T,
- Kohn M
- Adair CE,
- Kowalsky L,
- Quon H,
- et al
- Benitz WE,
- Gould JB,
- Druzin ML
- Oddie S,
- Embleton ND
- Schuchat A,
- Zywicki SS,
- Dinsmoor MJ,
- et al
- Copyright © 2010 by the American Academy of Pediatrics