- Email: [email protected]
Predictors of Occult Pneumococcal Bacteremia in Young Febrile Children
PEDIATRICS/ORIGINAL CONTRIBUTION
Predictors of Occult Pneumococcal Bacteremia in Young Febrile Children From the *Division of Emergency Medicine, Department of Internal Medicine and the Department of Pediatrics, University of California, Davis School of Medicine, Davis, CA; the ‡Division of Emergency Medicine, Department of Pediatrics, Harvard Medical School, Boston, MA; and the § Division of Emergency Medicine, Department of Pediatrics, Washington University School of Medicine, St Louis, MO.
Nathan Kuppermann, MD, MPH* Gary R Fleisher, MD‡ David M Jaffe, MD§
Received for publication May 9, 1997. Revision received October 27, 1997. Accepted for publication November 24, 1997. Presented in part at the meeting of the Society for Pediatric Research, Washington DC, May 8, 1996. Copyright © 1998 by the American College of Emergency Physicians.
Study objective: Occult pneumococcal bacteremia (OPB) occurs in 2.5% to 3% of highly febrile children 3 to 36 months of age, and 10% to 25% of untreated patients with OPB experience complications, including 3% to 6% in whom meningitis develops. The purpose of this study was to identify predictors of OPB among a large cohort of young, febrile children treated as outpatients using multivariable statistical methods. Methods: We derived and validated a logistic regression model for the prediction of OPB. We evaluated 6,579 outpatients 3 to 36 months of age with temperatures of 39° C or higher who previously had been enrolled in a study of young febrile patients at risk of OPB in the emergency departments of 10 hospitals in the United States between 1987 and 1991; 164 patients (2.5%) had OPB. We randomly selected two thirds of this population for the derivation of the model and one third for validation. In the derivation set, we analyzed the univariate relationships of six variables with OPB: age, temperature, clinical score, WBC count, absolute neutrophil count (ANC), and absolute band count (ABC). All six variables were then entered into a logistic regression equation and those retaining statistical significance were considered to have an independent association with OPB. Results: Patients with OPB were younger, more frequently illappearing, and had higher temperatures, WBC, ANC, and ABC than patients without bacteremia. Only three variables, however, retained statistically significant associations with OPB in the multivariate analysis: ANC (adjusted odds ratio [OR] 1.15 for each 1,000 cells/mm3 increase, 95% confidence interval [CI] 1.06, 1.25), temperature (adjusted OR 1.77 for each 1° C increase, 95% CI 1.21, 2.58), and age younger than 2 years (adjusted OR 2.43 versus patients 2 to 3 years old, 95% CI interval 1.11, 5.34). In the derivation set, 8.1% of patients with ANCs greater than or equal to 10,000 cells/mm3 had OPB (95% CI 6.3, 10.1%) versus .8% of patients with ANCs less than 10,000 cells/mm3 (95% CI .5, 1.2%). When tested on the validation set, the model performed similarly.
JUNE 1998
31:6
ANNALS OF EMERGENCY MEDICINE
6 7 9
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
Conclusion: Independent predictors of OPB in children 3 to 36 months of age with temperatures of 39° C or higher treated as outpatients include ANC, temperature, and age younger than 2 years. These predictors may be used to develop clinical strategies to limit laboratory testing and antibiotic administration to those children at greatest risk of OPB. [Kuppermann N, Fleisher GR, Jaffe DM: Predictors of occult pneumococcal bacteremia in young febrile children. Ann Emerg Med June 1998;31:679-687.]
INTRODUCTION
Fever in children accounts for 10% to 20% of pediatric visits to emergency departments.1-4 In 2.5% to 3% of highly febrile children 3 to 36 months of age without physical signs of invasive bacterial disease, fever is due to occult bacteremia, caused primarily by Streptococcus pneumoniae.1,2,5-8 Among children with occult pneumococcal bacteremia (OPB) who are not treated with antibiotics at the time of initial evaluation, complications including cellulitis, pneumonia, and sepsis develop in 10% to 25%, and meningitis develops in 3% to 6%.6,9-17 Previous studies demonstrate that timely diagnosis and treatment of OPB with antibiotics reduces the risk of many of these complications. 8,12-18 Although controversial, several studies show that the risk of meningitis may also be reduced by antibiotic treatment of OPB.8,12,19,20 Laboratory evaluation and expectant antibiotic treatment of young febrile children at risk of OPB, however, requires time and transient patient discomfort, adds cost, and may play a role in the emerging resistance of S pneumoniae to selected antibiotics. Therefore it is important for clinicians choosing to administer antibiotics expectantly to limit their use to patients at greatest risk of OPB. During the past two decades, several investigators have explored the utility of the clinical examination and laboratory tests for predicting which young febrile patients have bacterial illnesses, including bacteremia, at the time of the evaluation.1,2,5,13,21-34 However, these studies have been limited by small numbers of patients with bacteremia, and many have included hospitalized patients with focal bacterial diseases. Furthermore, most of the analyses have been univariate in nature, and thus have not been adjusted for the presence of other important predictive and confounding variables. The objective of this study was to identify clinical and laboratory predictors of OPB in young highly febrile children treated as outpatients by deriving and validating a
6 8 0
multivariable model. These predictors are then used to stratify patients according to risk of OPB and to make suggestions for utilization of laboratory studies and antibiotics on an expectant basis. METHODS Patient Selection
The study sample was drawn from 6,680 pediatric patients participating in a multicenter, randomized trial of antibiotic use in young febrile children at risk of occult bacteremia. Patients were prospectively enrolled from 1987 to 1991 in the EDs of 10 hospitals in the United States.8 Inclusion criteria were as follows: age between 3 and 36 months, temperature greater than or equal to 39° C, and no apparent focal infection (other than uncomplicated otitis media). Exclusion criteria were the following: patients with a toxic clinical appearance requiring hospitalization, the presence of a specific viral infection (eg, croup, varicella) or focal bacterial infection other than otitis media (eg, cellulitis, urinary tract, meningitis), a known immunodeficiency or chronic illness that would affect the approach to a febrile illness, or immunization or antibiotic therapy within the preceding 48 hours. The study was approved by the institutional review board at each participating institution, and informed consent was obtained from the legal guardian of each participating child. Clinical Assessment
Before review of laboratory results, an attending pediatrician examined each patient and assigned a clinical score according to the Yale Observation Scale (YOS).29 Laboratory Assessment
Blood samples were obtained from each patient for culture. A CBC was strongly encouraged but not required. Other laboratory tests were performed at the discretion of the treating physicians. Of the 6,680 eligible patients in the original study, 101 were excluded from the current analysis. These included: (1) 43 patients whose specimens for blood cultures were lost, (2) 28 patients with non-pneumococcal bacteremia, and (3) 30 patients who were inadvertently enrolled despite age younger than 3 months or older than 36 months.8 The clinical and laboratory data of the remaining 6,579 patients were the subject of the current analysis; 164 of these patients had pneumococcal bacteremia and 6,415 had sterile blood cultures or growth of contaminant bacteria.
ANNALS OF EMERGENCY MEDICINE
31:6 JUNE 1998
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
A CBC was performed for 5,695 (89%) of the 6,415 patients without and 141 (86%) of the 164 patients with pneumococcal bacteremia (P=.26). Manual differential counts of the peripheral blood smears were performed for 5,340 (83%) of the 6,415 patients without and on 137 (84%) of the 164 patients with pneumococcal bacteremia (P=.92). Statistical Analysis
Generation of derivation and validation sets. To derive a model for the prediction of OPB, we randomly selected two thirds of the patients in our study sample to represent the derivation set, using a computer-generated randomization procedure.35 The remaining one third of the study sample represented the validation set, on which we tested the model’s performance. Univariate analysis of the derivation set. Six variables were evaluated for their univariate association with OPB in the derivation set: three clinical variables (age, YOS, temperature) and three laboratory variables (WBC count, absolute neutrophil count [ANC], and absolute band count [ABC]). The ANC and ABC were calculated on only those patients who had manual differential counts performed on their peripheral blood smears (5,477 [83%] of 6,579 patients in the complete database). The frequency distributions for the continuous variables were examined to determine whether to include the variables as dichotomous or continuous predictors in the multivariable analysis. Student’s t test was used to compare the variable age between patients with and without pneumococcal bacteremia. Because the risk of pneumococcal bacteremia increased from 2.4% in the first to 2.9% in the second year of life, but decreased to 1.6% in the third year of life, analysis of the age variable in a dichotomous manner (<24 months and ≥24 months) was considered more appropriate. The risk of OPB in these categories was compared using the χ2 test. The multiple-category variable YOS (values ranging from 6 to 30) was compared between patients with and without bacteremia using the Wilcoxon rank sum test. Because there were few patients with high scores (score >12, n=57, 1 with bacteremia; score >10, n=145, 6 with bacteremia), this variable was further analyzed by dividing the patients into two groups: patients with a YOS equal to 6 (the best possible score) or greater than 6. This dichotomous variable was analyzed using the χ2 test. The remaining continuous variables (temperature, WBC count, ANC, and ABC) were compared between patients with and without pneumococcal bacteremia using Student’s t test. These variables were not divided into dichotomous groups because of the continuous increase in the risk of
JUNE 1998
31:6
ANNALS OF EMERGENCY MEDICINE
bacteremia with increasing values. All statistical tests were conducted based on two-tailed alternatives. Finally, we compared the receiver operating characteristic (ROC) curves between WBC count, ANC, and ABC to determine which of these tests best distinguished patients with OPB from those without OPB.36 Multivariable analysis of the derivation set. A multiple logistic regression analysis was performed on the patients from the derivation set who had complete information on the six variables. All six variables from the univariate analysis were entered into the logistic regression model as each of these variables has been reported to be associated with bacteremia in previous reports.1,2,5,21,23-26,30,32,33,37,38 Variables that retained statistical significance (P<.05) were considered to have an independent association with OPB. An ROC curve was constructed for the final model identified in the regression analysis and compared with the ROC curve for the model applied to the validation set. In addition, model pseudo-R2 (the proportion of variation explained by the model) was compared. Clinical prediction of pneumococcal bacteremia. Risk profiles for pneumococcal bacteremia were constructed based on the significant predictive variables identified in the regression analysis of the derivation set. These risk profiles were then applied to the validation set. All statistical analyses were performed using Stata statistical software, version 5.0.35 R E S U LT S Study Sample
A total of 6,579 patients qualified for the current analysis, of whom 164 (2.5%) had OPB; 351 (5.3%) of these patients had otitis media.8 Of the entire study sample, 5,451 (82.9%) Table 1.
Comparison of patients in the derivation and validation sets. Characteristic* No. (%) of subjects No. (%) with OPB Age (mo) Median YOS (range) Temperature (° C) WBC (×103/mm3)† ANC (×103/mm3)† ABC (×103/mm3)†
Derivation Set
Validation Set
P Value
4,384 (67%) 109 (2.5%) 14.2±8.0 6 (6, 24) 39.8±0.6 13.1±6.7 7.4±5.2 .99±1.3
2,195 (33%) 55 (2.5%) 14.3±8.2 6 (6, 18) 39.8±0.6 13.1±6.6 7.5±5.1 .95±1.1
.96 .73 .39 .30 .91 .75 .26
*Values shown are means±SD unless noted otherwise. †WBC obtained on 89% of patients; ANC and ABC obtained on 83% of patients.
6 8 1
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
had complete information on all six variables. Compared with patients without complete information on all six variables, patients with complete information had a similar rate of bacteremia (2.5% versus 2.4%, odds ratio [OR] 1.05, 95% confidence interval [CI] .69, 1.59, P=.81, patients with complete information versus those without, respectively). The two groups also had similar mean age (14.22 versus 14.47 months, difference 0.25 months, 95% CI –.26, .77 months, P=.34) and mean temperature (39.78 versus 39.77° C, difference .01° C, 95% CI –.03, .04° C, P=.61). Patients with complete information, had slightly lower YOS scores than those patients without complete information (mean 6.6, median 6, range 6 to 24 versus mean 6.7, median 6, range 6 to 18, P<.01); however, there was substantial overlap between groups. Generation of Derivation and Validation Sets
From the study sample, 4,384 (66.6%) patients were randomly selected to represent the derivation set, of whom 109 (2.5%) had bacteremia. The remaining 2,195 (33.4%) patients, of whom 55 (2.5%) had bacteremia, represented the validation set. The patients in the derivation and validation sets were also similar in mean age, median YOS, mean temperature, and mean WBC count, ANC, and ABC (Table 1). Univariate Analysis of the Derivation Set
Patients with pneumococcal bacteremia were more frequently younger than 24 months, more frequently had elevated YOS scores, and had higher mean temperatures, WBC count, ANC, and ABC than patients without bacteremia (Table 2). Patients with YOS greater than 6 had a higher risk of pneumococcal bacteremia than patients with
scores of 6 (4.3 versus 2.1%, OR 2.1, 95% CI 1.4, 3.2, P<.001); however, the median YOS was 6 for patients with and without pneumococcal bacteremia, and 69% of the patients with bacteremia had the lowest possible score of 6. Examination of the frequency distributions of the continuous variables revealed a steady increase in the risk of pneumococcal bacteremia with increasing temperature, from a risk of 1.2% at a temperature less than 39.5° C to a risk of 4.4% at a temperature greater than or equal to 40.5° C (Figure 1, A). Similarly, there were steady increases in risk of pneumococcal bacteremia with increasing WBC count and ANC (Figure 1, B and C). Of note, there were no cases of pneumococcal bacteremia in the 175 patients with WBC counts less than or equal to 5,000 cells/mm3 (one-sided 95% CI 0, 1.7%) and only three cases in the 1,419 patients with ANCs less than or equal to 5,000 cells/mm3 (.2%, 95% CI 0, .6%). The risk of pneumococcal bacteremia also increased with higher ABC (Figure 1, D). Examination of the ROC curves for WBC count, ANC, and ABC revealed that the ANC had the greatest ability to distinguish patients with pneumococcal bacteremia from those without bacteremia, although the area under the curve for ANC (.835) was only slightly greater than the area under the curve for WBC (.812) (Figure 2). An ANC of 9,000 to 10,000 cells/mm3 appeared to have the best sensitivity/ specificity profile (the point at which each incremental increase in sensitivity produces a large loss of specificity). An ANC cutoff of 10,000 cells/mm3 had a sensitivity of 76% for detecting pneumococcal bacteremia (95% CI 66%, 84%) and a specificity of 78% (95% CI 76%, 79%); a WBC count cutoff of 15,000 cells/mm3 had a sensitivity of 80% (95% CI 70%, 87%) and a specificity of 69% (95% CI 68%, 71%).
Table 2.
Univariate analysis of the derivation set.
Characteristic* Age (mo) Age <2 yr (No., %) Median YOS (range) YOS >6 (No., %) Temperature (° C) WBC (×103/mm3) ANC (×103/mm3) ABC (×103/mm3)
OPB (n=109)
Non-OPB (n=4,275)
Difference Between Means or Odds Ratios for %† (95% CI)
P Value
14.17±6.94 99 (91%) 6 (6, 14) 34 (31%) 40.04±.58 21.49±8.21 14.70±7.06 2.13±2.32
14.23±8.05 3,670 (86%) 6 (6, 24) 751 (18%) 39.78±.55 12.90±6.54 7.25±4.97 .96±1.26
–.06 (–1.40, 1.28) 1.63 (0.86, 3.11)† — 2.12 (1.41, 3.20)† .26 (.16, .37) 8.59 (6.89, 10.29) 7.45 (5.99, 8.93) 1.17 (.68, 1.64)
.93 .14 <.001 <.001 <.001 <.001 <.001 <.001
*Values
shown are means±SD unless noted otherwise. ratios denoting the increased odds of OPB are given for categorical variables (age <2 years versus 2 to 3 years, YOS>6 versus YOS=6); differences in mean values between patients with and without OPB are given for continuous variables. †Odds
6 8 2
ANNALS OF EMERGENCY MEDICINE
31:6 JUNE 1998
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
Multivariable Analysis of the Derivation Set
For purposes of the multivariable analysis, only the 3,626 (83%) patients in the derivation set who had complete information on all 6 variables were included. Of these patients, 92 (2.5%) had pneumococcal bacteremia. All six variables were entered into the regression analysis. Only ANC, temperature, and age retained significant independent associations with pneumococcal bacteremia (P<.05, Table 3). The area under the curve for this model was .843, and the pseudo-R2 value was .158. When the three nonsignificant variables (YOS, WBC count, and ABC) were removed from the model, the area under the curve remained almost identical (.842), as did the pseudo-R2 (.156). Furthermore, both forward and backward stepwise modelbuilding techniques selected the same three-variable model with age, temperature, and ANC.39 The three-variable model was then tested on the 1,825 (83%) patients in the validation set who had complete in-
formation on all 6 variables, 45 (2.5%) of whom had pneumococcal bacteremia. When applied to this validation set, the model with age, temperature, and ANC had an area under the curve of .846, and a pseudo-R2 of .155, nearly identical to that of the same model applied to the derivation set (Figure 3). This suggests little if any degradation in the model’s performance between data sets. Clinical Prediction of Pneumococcal Bacteremia
Using the three significant predictors identified in the regression analysis (ANC, age, temperature), we developed and tested a simple clinical algorithm for the prediction of OPB (Figure 4). The patients in the derivation set could be divided into three groups with increasing risks of bacteremia: (1) patients 2 to 3 years of age with temperatures less than 39.5° C, with a 1.1% (2/185) risk of bacteremia; (2) patients 2 to 3 years of age with temperatures greater than or equal to 39.5° C, and patients 3 to 24 months of
Figure 1.
Univariate associations of continuous variables with OPB in the derivation set. A, Temperature; B, WBC count; C, ANC; D, ABC. The error bars represent the 95% CIs of the point estimates. % OPB
% OPB 10 8
10
4.4%
8
6 2.5%
4
3.2%
4.0%
6 4
1.2%
2 0
8.2%
12
1.3%
0%
.3%
2 39.0–39.4
A
39.5–39.9
40.0–40.4
0
≥40.5
10–14.9
15–19.9
≥20
WBC (×103/mm3)
% OPB
% OPB 10 12.2%
5.2%
6.3%
1.5–1.9
≥2
8 6
5.8% 4 .2% <5
C
JUNE 1998
5–9.9
B
Temperature (˚ C)
20 18 14 12 10 8 6 4 2 0
<5
1.4% 5–9.9
1.7%
1.7%
.5–.99
1–1.49
2 10–14.9
ANC (×103/mm3)
31:6
1.5%
ANNALS OF EMERGENCY MEDICINE
≥15
0
0–.49
D
ABC (×103/mm3)
6 8 3
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
age with temperatures greater than or equal to 39° C, with a 2.6% (107/4,185) risk of bacteremia ; (3) patients from group 2 who had ANCs greater than or equal to 10,000 cells/mm3, with an 8.3% (69/827) risk of pneumococcal bacteremia. When evaluated in the validation set, the risks of bacteremia in each of these three patient risk groups were very similar to those in the derivation set (Figure 4). Overall, the CBC allowed for effective stratification of patient risk. For the total study sample (combined derivation and validation sets), the risk of pneumococcal bacteremia for patients with ANCs greater than or equal to 10,000 cells/ mm3 was 7.8% (102/1,301, 95% CI 6.4%, 9.4%). For patients with ANCs less than 10,000 cells/mm3, however, the risk was only 0.8% (35/4,176, 95% CI .6%, 1.2%). Patients with WBC counts greater than or equal to 15,000 cells/mm3 had a 6% risk of pneumococcal bacteremia (112/ 1,869, 95% CI 5.0%, 7.2%) compared with a risk of .7% for patients with WBC counts less than 15,000 cells/mm3 (29/3,967, 95% CI 0.5%, 1.0%). DISCUSSION
The benefits of screening children at risk of OPB and administering antibiotics on an expectant basis must be balanced against the costs and inconvenience of excessive testing and the risks of indiscriminate antibiotic usage, including the potential for development of resistant bacterial strains. The identification of reliable predictors of OPB will increase
Figure 2.
Comparison of WBC count, ANC, and ABC ROC curves in the derivation set. AUC, Area under the curve.
diagnostic accuracy and decrease unnecessary antibiotic usage. In this large study of young, highly febrile children, we have identified ANC as the major predictor of OPB of the six variables we evaluated, and this predictor can be further refined on the basis of two other independent predictors, age and temperature. This study adds several new pieces of information to what is currently known about occult bacteremia. Because we used a very large cohort of febrile children at risk of OPB and performed a multivariable analysis, we were able to quantify the independent contribution of various risk factors accurately. By means of ROC analysis, we found that no factor was more important than the ANC. Furthermore, we found that the band count in the peripheral blood smear does not contribute predictive information after adjusting for the ANC, thus obviating the need for a manual differential count on these patients. Our results agree with previous research that identified age younger than 2 years1,2,26,37 and high fever1,2,5,26,33 as risk factors for pneumococcal bacteremia and further characterize these associations. In our study sample, the overall risk of pneumococcal bacteremia in the 2- to 3-year age group in children with a temperature of 39.0 to 39.4° C was only .7%. The risk of bacteremia in this older age group at temperatures greater than or equal to 39.5° C, however, was somewhat higher (2.1%), and similar to the overall risk of bacteremia of children younger than 2 years with temperatures greater than or equal to 39° C (2.6%, P=0.4). Furthermore, in patients younger than 2 years, the risk of pneumococcal bacteremia increased with increasing temperature after adjusting for the ANC. This study demonstrates, by means of ROC curve analysis, that the ANC is a somewhat better screening tool for OPB than is the WBC count, and both the ANC and WBC
Sensitivity (True Positive) 1 ANC≥10K
Table 3.
ANC≥9K
Logistic regression analysis of the derivation set.
.75 WBC≥15K
.50 ABC≥1.5K
.25
ROC curve for ANC: AUC=.835 ROC curve for WBC: AUC=.812 ROC curve for ABC: AUC=.668
0 0
.25
.50
1–Specificity (False Positive)
6 8 4
.75
1
Predictor
OR*
95% CI
P Value
ANC Temperature Age <2 years YOS >6 WBC count ABC
1.15 1.77 2.43 1.23 1.01 1.02
1.06, 1.25 1.21, 2.58 1.11, 5.34 .74, 2.04 .95, 1.08 .91, 1.14
.001 .003 .03 .42 .77 .71
*The odds ratios describe the increased odds of OPB for the following increments in the predictive variables: (1) an increase in ANC, WBC count, and ABC of 1,000 cells/mm3, (2) a 1° C increase in temperature, (3) patients younger than 2 years versus patients 2 to 3 years of age, and (4) patients with YOS scores greater than 6 versus patients with YOS scores of 6.
ANNALS OF EMERGENCY MEDICINE
31:6 JUNE 1998
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
count are better than the ABC. In addition, in the multivariable analysis we found that the ABC did not contribute significant predictive information after adjusting for the ANC (or WBC count). Several previous studies have compared the WBC count with the ANC, ABC, or both as predictors of bacterial illness, including bacteremia, with discrepant results.21,23-25,30,32,38 Our study differs from these previous studies in several ways. Most importantly, we included only young, highly febrile children without a focus of infection (other than otitis media) who were treated as outpatients, and identified a large number of patients with OPB. Many of the previous studies included patients hospitalized with focal bacterial infections, varying widely in age and temperature, and with bacteremias caused by many different pathogens (including H influenzae type b, which is now rarely seen). In addition, previous studies generally used univariable rather than multivariable statistical methods, and thus failed to adjust for the presence of other important predictive and confounding variables. This study also demonstrates that the YOS does not reliably detect patients with OPB and does not contribute significant predictive information after adjusting for the ANC, age, and temperature. These results have been previously reported, in a study including patients with bac-
teremias caused by a variety of pathogens40 and are not unexpected given that the majority of young, febrile children on whom the YOS was derived and validated, had overt (meningitis, pneumonia, cellulitis) rather than occult disease.29,41 Our study has several potential limitations. The patients were enrolled in EDs only. The similarity of OPB in this setting to the clinic and office settings, however, is well documented.1,5,13,42 In addition, we did not evaluate bacteremia caused by pathogens other than S pneumoniae. Therefore the model derived in this study does not apply to these other uncommon causes of occult bacteremia. In the current analysis, we excluded the 28 patients with nonpneumococcal bacteremia from the study sample. Of these, nine were infected with H influenzae, a cause of bacteremia now rarely observed in the fully immunized child.43 Of the remaining 19, 7 had Salmonella bacteremia, 2 each had Neisseria meningitidis, Staphylococcus aureus, and Escherichia coli bacteremias, and the remaining 6 had other bacteremias.8 The number of patients with each of these types of bacteremias was too small to permit us to reach meaningful conclusions about the relationship of fever and ANC to these bacteremias. Previous investigators, however, have noted that the WBC count may not be elevated in children with gram-negative infections.21,22,25 In the current era of
Figure 3.
ROC curves for the three-variable predictive model (ANC, temperature, and age younger than 2 years) in the derivation and validation sets. The 45-degree line through the origin represents the ROC curve of a test whose decision ability is no greater than chance. AUC, Area under the curve. Sensitivity (True Positive)
Figure 4.
Clinical preduction of OPB by risk group in the derivation and validation sets. Risk Group 1: Patients 24 to 36 months of age with temperatures <39.5° C. Risk Group 2: Patients 24 to 36 months of age with temperatures ≥39.5° C, and patients 3 to 24 months of age with temperatures ≥39.0° C. Risk Group 3: Patients from Risk Group 2 with ANCs ≥10,000/mm3. The error bars represent the 95% CIs of the point estimates.
1 % OPB .75
12 10
.50 8 ROC curve for the model applied to the Derivation set: AUC=.842 ROC curve for the model applied to the Validation set: AUC=.846
.25
0
Validation Set
6 4
2/185
0/101
107/4185 55/2086
2 0
.25
.50
.75
1–Specificity (False Positive)
JUNE 1998
69/827 32/411 Derivation Set
31:6
ANNALS OF EMERGENCY MEDICINE
1 0 Risk Group 1
Risk Group 2
Risk Group 3
6 8 5
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
widespread immunization against H influenzae type b, however, S pneumoniae is responsible for at least 90% of the cases of occult bacteremia in patients 3 to 36 months of age, and for the majority of complications of occult bacteremia.7 Applying the three independent predictors of OPB identified among our patients allows for the development of more efficient clinical strategies for the evaluation of young, nontoxic-appearing, highly febrile children. Such children could be divided into three groups with varying risks of pneumococcal bacteremia (based on the data from the combined derivation and validation sets), suggesting different clinical approaches: (1) patients 2 to 3 years of age with temperatures less than 39.5° C, who have a very low overall risk of bacteremia (.7%) and therefore do not merit routine laboratory evaluation or expectant antibiotics for OPB; (2) patients 2 to 3 years of age with temperatures greater than or equal to 39.5° C, and patients 3 to 24 months of age with temperatures greater than or equal to 39° C, with an overall risk of bacteremia of 2.6%, for whom an ANC (or WBC count) should be considered for stratification into high- and low-risk groups, and from whom blood cultures should be considered, depending on the ANC; (3) patients from group 2 with an ANC greater than or equal to 10,000 cells/mm3, and an 8.2% risk of pneumococcal bacteremia, for whom antibiotics should be considered on an expectant basis after samples for blood cultures have been obtained. Use of this clinical strategy for testing and treating young, highly febrile children at risk of OPB obviates the need for a manual differential of the peripheral blood smear and results in a decrease in blood culture collection and antibiotic administration compared with previous algorithms.15 This strategy would result in the administration of antibiotics on an expectant basis to less than 24% of patients 3 to 36 months of age with temperatures greater than or equal to 39° C without source (representing a small fraction of the total population of febrile pediatric patients evaluated as outpatients). This figure is substantially smaller than that for a strategy based on age and fever alone in which 100% of patients are treated with antibiotics or a strategy based on a WBC cutoff of greater than or equal to 15,000 cells/mm3 in which 32% are treated.15 We conclude that predictors independently associated with OPB in children 3 to 36 months of age with temperatures greater than or equal to 39° C without source (or with otitis media) treated as outpatients include the absolute neutrophil count, temperature, and age younger than 2 years. These predictors may be used to develop clinical strategies to limit laboratory testing and antibiotic administration to those children at greatest risk of OPB.
6 8 6
We gratefully acknowledge the helpful guidance of Wesley O Johnson, PhD, in the statistical analysis of this study. Contributing investigators from the Occult Bacteremia Study Group are Norman Rosenberg, DO, Department of Pediatrics, Wayne State University School of Medicine, Detroit, MI; Robert Vinci, MD, Department of Pediatrics, Boston University School of Medicine, Boston, MA; Joel Steinberg, MD, Department of Pediatrics, University of Texas Southwestern Medical School, Dallas; Keith Powell, MD, and Cynthia Christy, MD, Department of Pediatrics, University of Rochester School of Medicine and Dentistry, Rochester, NY; Douglas A Boenning, MD, Department of Pediatrics, George Washington School of Medicine, Washington DC; Gary Overturf, MD, Department of Pediatrics, University of New Mexico School of Medicine, Albuquerque; and Richard Platt, MD, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA.
REFERENCES 1. McGowan JE, Bratton L, Klein JO, et al: Bacteremia in febrile children seen in a “walk-in” pediatric clinic. N Engl J Med 1973;288:1309-1312. 2. McCarthy PL, Grundy GW, Spiesel SZ, et al: Bacteremia in children: An outpatient clinical review. Pediatrics 1976;57:861-869. 3. Krauss BS, Harakal T, Fleisher GR: The spectrum and frequency of illness presenting to a pediatric emergency department. Pediatr Emerg Care 1991;7:67-71. 4. Nelson DS, Walsh K, Fleisher GR: Spectrum and frequency of pediatric illness presenting to a general community hospital emergency department. Pediatrics 1992;90:5-10. 5. Teele DW, Pelton SI, Grant MJA, et al: Bacteremia in febrile children under 2 years of age: Results of cultures of blood of 600 consecutive febrile children seen in a “walk-in” clinic. J Pediatr 1975;87:227-230. 6. Jaffe DM, Tanz RR, Davis AT, et al: Antibiotic administration to treat possible occult bacteremia in febrile children. N Engl J Med 1987;317:1175-1180. 7. Harper MB, Fleisher GR: Occult bacteremia in the 3-month-old to 3-year-old age group. Pediatr Ann 1993;22:484, 487-493. 8. Fleisher GR, Rosenberg N, Vinci R, et al: Intramuscular versus oral antibiotic therapy for the prevention of meningitis and other bacterial sequelae in young, febrile children at risk for occult bacteremia. J Pediatr 1994;124:504-512. 9. Myers MG, Wright PF, Smith AL, et al: Complications of occult pneumococcal bacteremia in children. J Pediatr 1974;84:656-660. 10. Bratton L, Teele DW, Klein JO: Outcome of unsuspected pneumococcemia in children not initially admitted to the hospital. J Pediatr 1977;90:703-706. 11. Feder HM Jr: Occult pneumococcal bacteremia and the febrile infant and young child. Clin Pediatr 1980;19:457-462. 12. Carroll WL, Farrell MK, Singer JI, et al: Treatment of occult bacteremia: A prospective randomized clinical trial. Pediatrics 1983;72:608-612. 13. Baron MA, Fink HD, Cicchetti DV: Blood cultures in private pediatric practice: An eleven-year experience. Pediatr Infect Dis J 1989;8:2-7. 14. Woods ER, Merola JL, Bithoney WG, et al: Bacteremia in an ambulatory setting. Improved outcome in children treated with antibiotics. Am J Dis Child 1990;144:1195-1199. 15. Baraff LJ, Bass JW, Fleisher GR, et al: Practice guideline for the management of infants and children 0 to 36 months of age with fever without source. Pediatrics 1993;92:1-12. 16. Harper MB, Bachur R, Fleisher GR: Effect of antibiotic therapy on the outcome of outpatients with unsuspected bacteremia. Pediatr Infect Dis J 1995;14:760-767. 17. Rothrock SG, Harper MB, Green SM, et al: Do oral antibiotics prevent meningitis and serious bacterial infections in children with Streptococcus pneumoniae occult bacteremia? A meta-analysis. Pediatrics 1997;99:438-444. 18. Bass JW, Steele RW, Wittler RR, et al: Antimicrobial treatment of occult bacteremia: A multicenter cooperative study. Pediatr Infect Dis J 1993;12:466-473. 19. Baraff LJ, Lee SI: Fever without source: Management of children 3 to 36 months of age. Pediatr Infect Dis J 1992;11:146-151.
ANNALS OF EMERGENCY MEDICINE
31:6 JUNE 1998
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
20. Baraff LJ, Oslund S, Prather M: Effect of antibiotic therapy and etiologic microorganism on the risk of bacterial meningitis in children with occult bacteremia. Pediatrics 1993;92:140-143. 21. Todd JK: Childhood infections: Diagnostic value of peripheral white blood cell and differential cell counts. Am J Dis Child 1974;127:810-816. 22. Weitzman M: Diagnostic utility of white blood cell and differential cell counts. Am J Dis Child 1975;129:1183-1189.
Reprint no. 47/1/90125 Address for reprints: Nathan Kuppermann, MD, MPH Division of Emergency Medicine and the Department of Pediatrics
23. McCarthy PL, Jekel JF, Dolan TF Jr: Temperature greater than or equal to 40˚ C in children less than 24 months of age: A prospective study. Pediatrics 1977;59:663-668.
University of California, Davis, Medical Center
24. McCarthy PL, Jekel JF, Dolan TF Jr: Comparison of acute-phase reactants in pediatric patients with fever. Pediatrics 1978;62:716-720.
PSSB Building
2315 Stockton Boulevard
25. Morens DM: WBC count and differential: Value in predicting bacterial diseases in children. Am J Dis Child 1979;133:25-27.
Sacramento, CA 95817
26. Teele DW, Marshall R, Klein JO: Unsuspected bacteremia in young children: A common and important problem. Pediatr Clin North Am 1979;26:773-784.
Fax 916-734-7950
27. Waskerwitz S, Berlelhamer JE: Outpatient bacteremia: Clinical findings in children under two years with initial temperatures of 39.5˚ C or higher. J Pediatr 1981;99:231-233.
916-734-1535 E-mail [email protected]
28. Rasmussen NH, Rasmussen LN: Predictive value of white blood cell count and differential cell count to bacterial infections in children. Acta Paediatr Scand 1982;71:775-778. 29. McCarthy PL, Sharpe MR, Spiesel SZ, et al: Observation scales to identify serious illness in febrile children. Pediatrics 1982;70:802-809. 30. Bennish M, Beem MO, Ormiste V: C-reactive protein and zeta sedimentation ratio as indicators of bacteremia in pediatric patients. J Pediatr 1984;104:729-732. 31. Liu C-H, Lehan C, Speer ME, et al: Early detection of bacteremia in an outpatient clinic. Pediatrics 1985;75:827-831. 32. Crocker PJ, Quick G, McCombs W: Occult bacteremia in the emergency department: Diagnostic criteria for the young febrile child. Ann Emerg Med 1985;14:1172-1177. 33. Jaffe DM, Fleisher GR: Temperature and total white blood cell count as indicators of bacteremia. Pediatrics 1991;87:670-674. 34. Kramer MS, Tange SM, Mills EL, et al: Role of the complete blood count in detecting occult focal bacterial infection in the young febrile child. J Clin Epidemiol 1993;46:349-357. 35. StataCorp: Stata Statistical Software. Release 5.0. College Station, TX: Stata Corporation, 1997. 36. Browner WS, Newman TB, Cummings SR. Designing a new study: III. Diagnostic tests, in Hulley SB, Cummings SR (eds): Designing clinical research. Baltimore: Williams & Wilkins, 1988:90. 37. Murray DL, Zonana J, Seidel JS, et al: Relative importance of bacteremia and viremia in the course of acute fevers of unknown origin in outpatient children. Pediatrics 1981;68:157-160. 38. Baker RC, Tiller T, Bausher JC, et al: Severity of disease correlated with fever reduction in febrile infants. Pediatrics 1989;83:1016-1019. 39. Hosmer DW, Lemeshow S: Applied logistic regression. New York: John Wiley & Sons, 1989. 40. Teach SJ, Fleisher GR: Efficacy of an observation scale in detecting bacteremia in febrile children three to thirty-six months of age, treated as outpatients. J Pediatr 1995;126:877-881. 41. McCarthy PL, Lembo RM, Fink HD, et al: Observation, history, and physical examination in diagnosis of serious illnesses in febrile children ≤24 months. J Pediatr 1987;110:26-30. 42. Dershewitz RA, Wigder HN, Wigder CM, et al: A comparative study of the prevalence, outcome, and prediction of bacteremia in children. J Pediatr 1983;103:352-358. 43. Adams WG, Deaver KA, Cochi SL, et al: Decline of childhood Haemophilus influenzae type b (Hib) disease in the Hib vaccine era. JAMA 1993;269:221-226.
JUNE 1998
31:6
ANNALS OF EMERGENCY MEDICINE
6 8 7
Predictors of Occult Pneumococcal Bacteremia in Young Febrile Children From the *Division of Emergency Medicine, Department of Internal Medicine and the Department of Pediatrics, University of California, Davis School of Medicine, Davis, CA; the ‡Division of Emergency Medicine, Department of Pediatrics, Harvard Medical School, Boston, MA; and the § Division of Emergency Medicine, Department of Pediatrics, Washington University School of Medicine, St Louis, MO.
Nathan Kuppermann, MD, MPH* Gary R Fleisher, MD‡ David M Jaffe, MD§
Received for publication May 9, 1997. Revision received October 27, 1997. Accepted for publication November 24, 1997. Presented in part at the meeting of the Society for Pediatric Research, Washington DC, May 8, 1996. Copyright © 1998 by the American College of Emergency Physicians.
Study objective: Occult pneumococcal bacteremia (OPB) occurs in 2.5% to 3% of highly febrile children 3 to 36 months of age, and 10% to 25% of untreated patients with OPB experience complications, including 3% to 6% in whom meningitis develops. The purpose of this study was to identify predictors of OPB among a large cohort of young, febrile children treated as outpatients using multivariable statistical methods. Methods: We derived and validated a logistic regression model for the prediction of OPB. We evaluated 6,579 outpatients 3 to 36 months of age with temperatures of 39° C or higher who previously had been enrolled in a study of young febrile patients at risk of OPB in the emergency departments of 10 hospitals in the United States between 1987 and 1991; 164 patients (2.5%) had OPB. We randomly selected two thirds of this population for the derivation of the model and one third for validation. In the derivation set, we analyzed the univariate relationships of six variables with OPB: age, temperature, clinical score, WBC count, absolute neutrophil count (ANC), and absolute band count (ABC). All six variables were then entered into a logistic regression equation and those retaining statistical significance were considered to have an independent association with OPB. Results: Patients with OPB were younger, more frequently illappearing, and had higher temperatures, WBC, ANC, and ABC than patients without bacteremia. Only three variables, however, retained statistically significant associations with OPB in the multivariate analysis: ANC (adjusted odds ratio [OR] 1.15 for each 1,000 cells/mm3 increase, 95% confidence interval [CI] 1.06, 1.25), temperature (adjusted OR 1.77 for each 1° C increase, 95% CI 1.21, 2.58), and age younger than 2 years (adjusted OR 2.43 versus patients 2 to 3 years old, 95% CI interval 1.11, 5.34). In the derivation set, 8.1% of patients with ANCs greater than or equal to 10,000 cells/mm3 had OPB (95% CI 6.3, 10.1%) versus .8% of patients with ANCs less than 10,000 cells/mm3 (95% CI .5, 1.2%). When tested on the validation set, the model performed similarly.
JUNE 1998
31:6
ANNALS OF EMERGENCY MEDICINE
6 7 9
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
Conclusion: Independent predictors of OPB in children 3 to 36 months of age with temperatures of 39° C or higher treated as outpatients include ANC, temperature, and age younger than 2 years. These predictors may be used to develop clinical strategies to limit laboratory testing and antibiotic administration to those children at greatest risk of OPB. [Kuppermann N, Fleisher GR, Jaffe DM: Predictors of occult pneumococcal bacteremia in young febrile children. Ann Emerg Med June 1998;31:679-687.]
INTRODUCTION
Fever in children accounts for 10% to 20% of pediatric visits to emergency departments.1-4 In 2.5% to 3% of highly febrile children 3 to 36 months of age without physical signs of invasive bacterial disease, fever is due to occult bacteremia, caused primarily by Streptococcus pneumoniae.1,2,5-8 Among children with occult pneumococcal bacteremia (OPB) who are not treated with antibiotics at the time of initial evaluation, complications including cellulitis, pneumonia, and sepsis develop in 10% to 25%, and meningitis develops in 3% to 6%.6,9-17 Previous studies demonstrate that timely diagnosis and treatment of OPB with antibiotics reduces the risk of many of these complications. 8,12-18 Although controversial, several studies show that the risk of meningitis may also be reduced by antibiotic treatment of OPB.8,12,19,20 Laboratory evaluation and expectant antibiotic treatment of young febrile children at risk of OPB, however, requires time and transient patient discomfort, adds cost, and may play a role in the emerging resistance of S pneumoniae to selected antibiotics. Therefore it is important for clinicians choosing to administer antibiotics expectantly to limit their use to patients at greatest risk of OPB. During the past two decades, several investigators have explored the utility of the clinical examination and laboratory tests for predicting which young febrile patients have bacterial illnesses, including bacteremia, at the time of the evaluation.1,2,5,13,21-34 However, these studies have been limited by small numbers of patients with bacteremia, and many have included hospitalized patients with focal bacterial diseases. Furthermore, most of the analyses have been univariate in nature, and thus have not been adjusted for the presence of other important predictive and confounding variables. The objective of this study was to identify clinical and laboratory predictors of OPB in young highly febrile children treated as outpatients by deriving and validating a
6 8 0
multivariable model. These predictors are then used to stratify patients according to risk of OPB and to make suggestions for utilization of laboratory studies and antibiotics on an expectant basis. METHODS Patient Selection
The study sample was drawn from 6,680 pediatric patients participating in a multicenter, randomized trial of antibiotic use in young febrile children at risk of occult bacteremia. Patients were prospectively enrolled from 1987 to 1991 in the EDs of 10 hospitals in the United States.8 Inclusion criteria were as follows: age between 3 and 36 months, temperature greater than or equal to 39° C, and no apparent focal infection (other than uncomplicated otitis media). Exclusion criteria were the following: patients with a toxic clinical appearance requiring hospitalization, the presence of a specific viral infection (eg, croup, varicella) or focal bacterial infection other than otitis media (eg, cellulitis, urinary tract, meningitis), a known immunodeficiency or chronic illness that would affect the approach to a febrile illness, or immunization or antibiotic therapy within the preceding 48 hours. The study was approved by the institutional review board at each participating institution, and informed consent was obtained from the legal guardian of each participating child. Clinical Assessment
Before review of laboratory results, an attending pediatrician examined each patient and assigned a clinical score according to the Yale Observation Scale (YOS).29 Laboratory Assessment
Blood samples were obtained from each patient for culture. A CBC was strongly encouraged but not required. Other laboratory tests were performed at the discretion of the treating physicians. Of the 6,680 eligible patients in the original study, 101 were excluded from the current analysis. These included: (1) 43 patients whose specimens for blood cultures were lost, (2) 28 patients with non-pneumococcal bacteremia, and (3) 30 patients who were inadvertently enrolled despite age younger than 3 months or older than 36 months.8 The clinical and laboratory data of the remaining 6,579 patients were the subject of the current analysis; 164 of these patients had pneumococcal bacteremia and 6,415 had sterile blood cultures or growth of contaminant bacteria.
ANNALS OF EMERGENCY MEDICINE
31:6 JUNE 1998
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
A CBC was performed for 5,695 (89%) of the 6,415 patients without and 141 (86%) of the 164 patients with pneumococcal bacteremia (P=.26). Manual differential counts of the peripheral blood smears were performed for 5,340 (83%) of the 6,415 patients without and on 137 (84%) of the 164 patients with pneumococcal bacteremia (P=.92). Statistical Analysis
Generation of derivation and validation sets. To derive a model for the prediction of OPB, we randomly selected two thirds of the patients in our study sample to represent the derivation set, using a computer-generated randomization procedure.35 The remaining one third of the study sample represented the validation set, on which we tested the model’s performance. Univariate analysis of the derivation set. Six variables were evaluated for their univariate association with OPB in the derivation set: three clinical variables (age, YOS, temperature) and three laboratory variables (WBC count, absolute neutrophil count [ANC], and absolute band count [ABC]). The ANC and ABC were calculated on only those patients who had manual differential counts performed on their peripheral blood smears (5,477 [83%] of 6,579 patients in the complete database). The frequency distributions for the continuous variables were examined to determine whether to include the variables as dichotomous or continuous predictors in the multivariable analysis. Student’s t test was used to compare the variable age between patients with and without pneumococcal bacteremia. Because the risk of pneumococcal bacteremia increased from 2.4% in the first to 2.9% in the second year of life, but decreased to 1.6% in the third year of life, analysis of the age variable in a dichotomous manner (<24 months and ≥24 months) was considered more appropriate. The risk of OPB in these categories was compared using the χ2 test. The multiple-category variable YOS (values ranging from 6 to 30) was compared between patients with and without bacteremia using the Wilcoxon rank sum test. Because there were few patients with high scores (score >12, n=57, 1 with bacteremia; score >10, n=145, 6 with bacteremia), this variable was further analyzed by dividing the patients into two groups: patients with a YOS equal to 6 (the best possible score) or greater than 6. This dichotomous variable was analyzed using the χ2 test. The remaining continuous variables (temperature, WBC count, ANC, and ABC) were compared between patients with and without pneumococcal bacteremia using Student’s t test. These variables were not divided into dichotomous groups because of the continuous increase in the risk of
JUNE 1998
31:6
ANNALS OF EMERGENCY MEDICINE
bacteremia with increasing values. All statistical tests were conducted based on two-tailed alternatives. Finally, we compared the receiver operating characteristic (ROC) curves between WBC count, ANC, and ABC to determine which of these tests best distinguished patients with OPB from those without OPB.36 Multivariable analysis of the derivation set. A multiple logistic regression analysis was performed on the patients from the derivation set who had complete information on the six variables. All six variables from the univariate analysis were entered into the logistic regression model as each of these variables has been reported to be associated with bacteremia in previous reports.1,2,5,21,23-26,30,32,33,37,38 Variables that retained statistical significance (P<.05) were considered to have an independent association with OPB. An ROC curve was constructed for the final model identified in the regression analysis and compared with the ROC curve for the model applied to the validation set. In addition, model pseudo-R2 (the proportion of variation explained by the model) was compared. Clinical prediction of pneumococcal bacteremia. Risk profiles for pneumococcal bacteremia were constructed based on the significant predictive variables identified in the regression analysis of the derivation set. These risk profiles were then applied to the validation set. All statistical analyses were performed using Stata statistical software, version 5.0.35 R E S U LT S Study Sample
A total of 6,579 patients qualified for the current analysis, of whom 164 (2.5%) had OPB; 351 (5.3%) of these patients had otitis media.8 Of the entire study sample, 5,451 (82.9%) Table 1.
Comparison of patients in the derivation and validation sets. Characteristic* No. (%) of subjects No. (%) with OPB Age (mo) Median YOS (range) Temperature (° C) WBC (×103/mm3)† ANC (×103/mm3)† ABC (×103/mm3)†
Derivation Set
Validation Set
P Value
4,384 (67%) 109 (2.5%) 14.2±8.0 6 (6, 24) 39.8±0.6 13.1±6.7 7.4±5.2 .99±1.3
2,195 (33%) 55 (2.5%) 14.3±8.2 6 (6, 18) 39.8±0.6 13.1±6.6 7.5±5.1 .95±1.1
.96 .73 .39 .30 .91 .75 .26
*Values shown are means±SD unless noted otherwise. †WBC obtained on 89% of patients; ANC and ABC obtained on 83% of patients.
6 8 1
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
had complete information on all six variables. Compared with patients without complete information on all six variables, patients with complete information had a similar rate of bacteremia (2.5% versus 2.4%, odds ratio [OR] 1.05, 95% confidence interval [CI] .69, 1.59, P=.81, patients with complete information versus those without, respectively). The two groups also had similar mean age (14.22 versus 14.47 months, difference 0.25 months, 95% CI –.26, .77 months, P=.34) and mean temperature (39.78 versus 39.77° C, difference .01° C, 95% CI –.03, .04° C, P=.61). Patients with complete information, had slightly lower YOS scores than those patients without complete information (mean 6.6, median 6, range 6 to 24 versus mean 6.7, median 6, range 6 to 18, P<.01); however, there was substantial overlap between groups. Generation of Derivation and Validation Sets
From the study sample, 4,384 (66.6%) patients were randomly selected to represent the derivation set, of whom 109 (2.5%) had bacteremia. The remaining 2,195 (33.4%) patients, of whom 55 (2.5%) had bacteremia, represented the validation set. The patients in the derivation and validation sets were also similar in mean age, median YOS, mean temperature, and mean WBC count, ANC, and ABC (Table 1). Univariate Analysis of the Derivation Set
Patients with pneumococcal bacteremia were more frequently younger than 24 months, more frequently had elevated YOS scores, and had higher mean temperatures, WBC count, ANC, and ABC than patients without bacteremia (Table 2). Patients with YOS greater than 6 had a higher risk of pneumococcal bacteremia than patients with
scores of 6 (4.3 versus 2.1%, OR 2.1, 95% CI 1.4, 3.2, P<.001); however, the median YOS was 6 for patients with and without pneumococcal bacteremia, and 69% of the patients with bacteremia had the lowest possible score of 6. Examination of the frequency distributions of the continuous variables revealed a steady increase in the risk of pneumococcal bacteremia with increasing temperature, from a risk of 1.2% at a temperature less than 39.5° C to a risk of 4.4% at a temperature greater than or equal to 40.5° C (Figure 1, A). Similarly, there were steady increases in risk of pneumococcal bacteremia with increasing WBC count and ANC (Figure 1, B and C). Of note, there were no cases of pneumococcal bacteremia in the 175 patients with WBC counts less than or equal to 5,000 cells/mm3 (one-sided 95% CI 0, 1.7%) and only three cases in the 1,419 patients with ANCs less than or equal to 5,000 cells/mm3 (.2%, 95% CI 0, .6%). The risk of pneumococcal bacteremia also increased with higher ABC (Figure 1, D). Examination of the ROC curves for WBC count, ANC, and ABC revealed that the ANC had the greatest ability to distinguish patients with pneumococcal bacteremia from those without bacteremia, although the area under the curve for ANC (.835) was only slightly greater than the area under the curve for WBC (.812) (Figure 2). An ANC of 9,000 to 10,000 cells/mm3 appeared to have the best sensitivity/ specificity profile (the point at which each incremental increase in sensitivity produces a large loss of specificity). An ANC cutoff of 10,000 cells/mm3 had a sensitivity of 76% for detecting pneumococcal bacteremia (95% CI 66%, 84%) and a specificity of 78% (95% CI 76%, 79%); a WBC count cutoff of 15,000 cells/mm3 had a sensitivity of 80% (95% CI 70%, 87%) and a specificity of 69% (95% CI 68%, 71%).
Table 2.
Univariate analysis of the derivation set.
Characteristic* Age (mo) Age <2 yr (No., %) Median YOS (range) YOS >6 (No., %) Temperature (° C) WBC (×103/mm3) ANC (×103/mm3) ABC (×103/mm3)
OPB (n=109)
Non-OPB (n=4,275)
Difference Between Means or Odds Ratios for %† (95% CI)
P Value
14.17±6.94 99 (91%) 6 (6, 14) 34 (31%) 40.04±.58 21.49±8.21 14.70±7.06 2.13±2.32
14.23±8.05 3,670 (86%) 6 (6, 24) 751 (18%) 39.78±.55 12.90±6.54 7.25±4.97 .96±1.26
–.06 (–1.40, 1.28) 1.63 (0.86, 3.11)† — 2.12 (1.41, 3.20)† .26 (.16, .37) 8.59 (6.89, 10.29) 7.45 (5.99, 8.93) 1.17 (.68, 1.64)
.93 .14 <.001 <.001 <.001 <.001 <.001 <.001
*Values
shown are means±SD unless noted otherwise. ratios denoting the increased odds of OPB are given for categorical variables (age <2 years versus 2 to 3 years, YOS>6 versus YOS=6); differences in mean values between patients with and without OPB are given for continuous variables. †Odds
6 8 2
ANNALS OF EMERGENCY MEDICINE
31:6 JUNE 1998
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
Multivariable Analysis of the Derivation Set
For purposes of the multivariable analysis, only the 3,626 (83%) patients in the derivation set who had complete information on all 6 variables were included. Of these patients, 92 (2.5%) had pneumococcal bacteremia. All six variables were entered into the regression analysis. Only ANC, temperature, and age retained significant independent associations with pneumococcal bacteremia (P<.05, Table 3). The area under the curve for this model was .843, and the pseudo-R2 value was .158. When the three nonsignificant variables (YOS, WBC count, and ABC) were removed from the model, the area under the curve remained almost identical (.842), as did the pseudo-R2 (.156). Furthermore, both forward and backward stepwise modelbuilding techniques selected the same three-variable model with age, temperature, and ANC.39 The three-variable model was then tested on the 1,825 (83%) patients in the validation set who had complete in-
formation on all 6 variables, 45 (2.5%) of whom had pneumococcal bacteremia. When applied to this validation set, the model with age, temperature, and ANC had an area under the curve of .846, and a pseudo-R2 of .155, nearly identical to that of the same model applied to the derivation set (Figure 3). This suggests little if any degradation in the model’s performance between data sets. Clinical Prediction of Pneumococcal Bacteremia
Using the three significant predictors identified in the regression analysis (ANC, age, temperature), we developed and tested a simple clinical algorithm for the prediction of OPB (Figure 4). The patients in the derivation set could be divided into three groups with increasing risks of bacteremia: (1) patients 2 to 3 years of age with temperatures less than 39.5° C, with a 1.1% (2/185) risk of bacteremia; (2) patients 2 to 3 years of age with temperatures greater than or equal to 39.5° C, and patients 3 to 24 months of
Figure 1.
Univariate associations of continuous variables with OPB in the derivation set. A, Temperature; B, WBC count; C, ANC; D, ABC. The error bars represent the 95% CIs of the point estimates. % OPB
% OPB 10 8
10
4.4%
8
6 2.5%
4
3.2%
4.0%
6 4
1.2%
2 0
8.2%
12
1.3%
0%
.3%
2 39.0–39.4
A
39.5–39.9
40.0–40.4
0
≥40.5
10–14.9
15–19.9
≥20
WBC (×103/mm3)
% OPB
% OPB 10 12.2%
5.2%
6.3%
1.5–1.9
≥2
8 6
5.8% 4 .2% <5
C
JUNE 1998
5–9.9
B
Temperature (˚ C)
20 18 14 12 10 8 6 4 2 0
<5
1.4% 5–9.9
1.7%
1.7%
.5–.99
1–1.49
2 10–14.9
ANC (×103/mm3)
31:6
1.5%
ANNALS OF EMERGENCY MEDICINE
≥15
0
0–.49
D
ABC (×103/mm3)
6 8 3
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
age with temperatures greater than or equal to 39° C, with a 2.6% (107/4,185) risk of bacteremia ; (3) patients from group 2 who had ANCs greater than or equal to 10,000 cells/mm3, with an 8.3% (69/827) risk of pneumococcal bacteremia. When evaluated in the validation set, the risks of bacteremia in each of these three patient risk groups were very similar to those in the derivation set (Figure 4). Overall, the CBC allowed for effective stratification of patient risk. For the total study sample (combined derivation and validation sets), the risk of pneumococcal bacteremia for patients with ANCs greater than or equal to 10,000 cells/ mm3 was 7.8% (102/1,301, 95% CI 6.4%, 9.4%). For patients with ANCs less than 10,000 cells/mm3, however, the risk was only 0.8% (35/4,176, 95% CI .6%, 1.2%). Patients with WBC counts greater than or equal to 15,000 cells/mm3 had a 6% risk of pneumococcal bacteremia (112/ 1,869, 95% CI 5.0%, 7.2%) compared with a risk of .7% for patients with WBC counts less than 15,000 cells/mm3 (29/3,967, 95% CI 0.5%, 1.0%). DISCUSSION
The benefits of screening children at risk of OPB and administering antibiotics on an expectant basis must be balanced against the costs and inconvenience of excessive testing and the risks of indiscriminate antibiotic usage, including the potential for development of resistant bacterial strains. The identification of reliable predictors of OPB will increase
Figure 2.
Comparison of WBC count, ANC, and ABC ROC curves in the derivation set. AUC, Area under the curve.
diagnostic accuracy and decrease unnecessary antibiotic usage. In this large study of young, highly febrile children, we have identified ANC as the major predictor of OPB of the six variables we evaluated, and this predictor can be further refined on the basis of two other independent predictors, age and temperature. This study adds several new pieces of information to what is currently known about occult bacteremia. Because we used a very large cohort of febrile children at risk of OPB and performed a multivariable analysis, we were able to quantify the independent contribution of various risk factors accurately. By means of ROC analysis, we found that no factor was more important than the ANC. Furthermore, we found that the band count in the peripheral blood smear does not contribute predictive information after adjusting for the ANC, thus obviating the need for a manual differential count on these patients. Our results agree with previous research that identified age younger than 2 years1,2,26,37 and high fever1,2,5,26,33 as risk factors for pneumococcal bacteremia and further characterize these associations. In our study sample, the overall risk of pneumococcal bacteremia in the 2- to 3-year age group in children with a temperature of 39.0 to 39.4° C was only .7%. The risk of bacteremia in this older age group at temperatures greater than or equal to 39.5° C, however, was somewhat higher (2.1%), and similar to the overall risk of bacteremia of children younger than 2 years with temperatures greater than or equal to 39° C (2.6%, P=0.4). Furthermore, in patients younger than 2 years, the risk of pneumococcal bacteremia increased with increasing temperature after adjusting for the ANC. This study demonstrates, by means of ROC curve analysis, that the ANC is a somewhat better screening tool for OPB than is the WBC count, and both the ANC and WBC
Sensitivity (True Positive) 1 ANC≥10K
Table 3.
ANC≥9K
Logistic regression analysis of the derivation set.
.75 WBC≥15K
.50 ABC≥1.5K
.25
ROC curve for ANC: AUC=.835 ROC curve for WBC: AUC=.812 ROC curve for ABC: AUC=.668
0 0
.25
.50
1–Specificity (False Positive)
6 8 4
.75
1
Predictor
OR*
95% CI
P Value
ANC Temperature Age <2 years YOS >6 WBC count ABC
1.15 1.77 2.43 1.23 1.01 1.02
1.06, 1.25 1.21, 2.58 1.11, 5.34 .74, 2.04 .95, 1.08 .91, 1.14
.001 .003 .03 .42 .77 .71
*The odds ratios describe the increased odds of OPB for the following increments in the predictive variables: (1) an increase in ANC, WBC count, and ABC of 1,000 cells/mm3, (2) a 1° C increase in temperature, (3) patients younger than 2 years versus patients 2 to 3 years of age, and (4) patients with YOS scores greater than 6 versus patients with YOS scores of 6.
ANNALS OF EMERGENCY MEDICINE
31:6 JUNE 1998
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
count are better than the ABC. In addition, in the multivariable analysis we found that the ABC did not contribute significant predictive information after adjusting for the ANC (or WBC count). Several previous studies have compared the WBC count with the ANC, ABC, or both as predictors of bacterial illness, including bacteremia, with discrepant results.21,23-25,30,32,38 Our study differs from these previous studies in several ways. Most importantly, we included only young, highly febrile children without a focus of infection (other than otitis media) who were treated as outpatients, and identified a large number of patients with OPB. Many of the previous studies included patients hospitalized with focal bacterial infections, varying widely in age and temperature, and with bacteremias caused by many different pathogens (including H influenzae type b, which is now rarely seen). In addition, previous studies generally used univariable rather than multivariable statistical methods, and thus failed to adjust for the presence of other important predictive and confounding variables. This study also demonstrates that the YOS does not reliably detect patients with OPB and does not contribute significant predictive information after adjusting for the ANC, age, and temperature. These results have been previously reported, in a study including patients with bac-
teremias caused by a variety of pathogens40 and are not unexpected given that the majority of young, febrile children on whom the YOS was derived and validated, had overt (meningitis, pneumonia, cellulitis) rather than occult disease.29,41 Our study has several potential limitations. The patients were enrolled in EDs only. The similarity of OPB in this setting to the clinic and office settings, however, is well documented.1,5,13,42 In addition, we did not evaluate bacteremia caused by pathogens other than S pneumoniae. Therefore the model derived in this study does not apply to these other uncommon causes of occult bacteremia. In the current analysis, we excluded the 28 patients with nonpneumococcal bacteremia from the study sample. Of these, nine were infected with H influenzae, a cause of bacteremia now rarely observed in the fully immunized child.43 Of the remaining 19, 7 had Salmonella bacteremia, 2 each had Neisseria meningitidis, Staphylococcus aureus, and Escherichia coli bacteremias, and the remaining 6 had other bacteremias.8 The number of patients with each of these types of bacteremias was too small to permit us to reach meaningful conclusions about the relationship of fever and ANC to these bacteremias. Previous investigators, however, have noted that the WBC count may not be elevated in children with gram-negative infections.21,22,25 In the current era of
Figure 3.
ROC curves for the three-variable predictive model (ANC, temperature, and age younger than 2 years) in the derivation and validation sets. The 45-degree line through the origin represents the ROC curve of a test whose decision ability is no greater than chance. AUC, Area under the curve. Sensitivity (True Positive)
Figure 4.
Clinical preduction of OPB by risk group in the derivation and validation sets. Risk Group 1: Patients 24 to 36 months of age with temperatures <39.5° C. Risk Group 2: Patients 24 to 36 months of age with temperatures ≥39.5° C, and patients 3 to 24 months of age with temperatures ≥39.0° C. Risk Group 3: Patients from Risk Group 2 with ANCs ≥10,000/mm3. The error bars represent the 95% CIs of the point estimates.
1 % OPB .75
12 10
.50 8 ROC curve for the model applied to the Derivation set: AUC=.842 ROC curve for the model applied to the Validation set: AUC=.846
.25
0
Validation Set
6 4
2/185
0/101
107/4185 55/2086
2 0
.25
.50
.75
1–Specificity (False Positive)
JUNE 1998
69/827 32/411 Derivation Set
31:6
ANNALS OF EMERGENCY MEDICINE
1 0 Risk Group 1
Risk Group 2
Risk Group 3
6 8 5
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
widespread immunization against H influenzae type b, however, S pneumoniae is responsible for at least 90% of the cases of occult bacteremia in patients 3 to 36 months of age, and for the majority of complications of occult bacteremia.7 Applying the three independent predictors of OPB identified among our patients allows for the development of more efficient clinical strategies for the evaluation of young, nontoxic-appearing, highly febrile children. Such children could be divided into three groups with varying risks of pneumococcal bacteremia (based on the data from the combined derivation and validation sets), suggesting different clinical approaches: (1) patients 2 to 3 years of age with temperatures less than 39.5° C, who have a very low overall risk of bacteremia (.7%) and therefore do not merit routine laboratory evaluation or expectant antibiotics for OPB; (2) patients 2 to 3 years of age with temperatures greater than or equal to 39.5° C, and patients 3 to 24 months of age with temperatures greater than or equal to 39° C, with an overall risk of bacteremia of 2.6%, for whom an ANC (or WBC count) should be considered for stratification into high- and low-risk groups, and from whom blood cultures should be considered, depending on the ANC; (3) patients from group 2 with an ANC greater than or equal to 10,000 cells/mm3, and an 8.2% risk of pneumococcal bacteremia, for whom antibiotics should be considered on an expectant basis after samples for blood cultures have been obtained. Use of this clinical strategy for testing and treating young, highly febrile children at risk of OPB obviates the need for a manual differential of the peripheral blood smear and results in a decrease in blood culture collection and antibiotic administration compared with previous algorithms.15 This strategy would result in the administration of antibiotics on an expectant basis to less than 24% of patients 3 to 36 months of age with temperatures greater than or equal to 39° C without source (representing a small fraction of the total population of febrile pediatric patients evaluated as outpatients). This figure is substantially smaller than that for a strategy based on age and fever alone in which 100% of patients are treated with antibiotics or a strategy based on a WBC cutoff of greater than or equal to 15,000 cells/mm3 in which 32% are treated.15 We conclude that predictors independently associated with OPB in children 3 to 36 months of age with temperatures greater than or equal to 39° C without source (or with otitis media) treated as outpatients include the absolute neutrophil count, temperature, and age younger than 2 years. These predictors may be used to develop clinical strategies to limit laboratory testing and antibiotic administration to those children at greatest risk of OPB.
6 8 6
We gratefully acknowledge the helpful guidance of Wesley O Johnson, PhD, in the statistical analysis of this study. Contributing investigators from the Occult Bacteremia Study Group are Norman Rosenberg, DO, Department of Pediatrics, Wayne State University School of Medicine, Detroit, MI; Robert Vinci, MD, Department of Pediatrics, Boston University School of Medicine, Boston, MA; Joel Steinberg, MD, Department of Pediatrics, University of Texas Southwestern Medical School, Dallas; Keith Powell, MD, and Cynthia Christy, MD, Department of Pediatrics, University of Rochester School of Medicine and Dentistry, Rochester, NY; Douglas A Boenning, MD, Department of Pediatrics, George Washington School of Medicine, Washington DC; Gary Overturf, MD, Department of Pediatrics, University of New Mexico School of Medicine, Albuquerque; and Richard Platt, MD, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA.
REFERENCES 1. McGowan JE, Bratton L, Klein JO, et al: Bacteremia in febrile children seen in a “walk-in” pediatric clinic. N Engl J Med 1973;288:1309-1312. 2. McCarthy PL, Grundy GW, Spiesel SZ, et al: Bacteremia in children: An outpatient clinical review. Pediatrics 1976;57:861-869. 3. Krauss BS, Harakal T, Fleisher GR: The spectrum and frequency of illness presenting to a pediatric emergency department. Pediatr Emerg Care 1991;7:67-71. 4. Nelson DS, Walsh K, Fleisher GR: Spectrum and frequency of pediatric illness presenting to a general community hospital emergency department. Pediatrics 1992;90:5-10. 5. Teele DW, Pelton SI, Grant MJA, et al: Bacteremia in febrile children under 2 years of age: Results of cultures of blood of 600 consecutive febrile children seen in a “walk-in” clinic. J Pediatr 1975;87:227-230. 6. Jaffe DM, Tanz RR, Davis AT, et al: Antibiotic administration to treat possible occult bacteremia in febrile children. N Engl J Med 1987;317:1175-1180. 7. Harper MB, Fleisher GR: Occult bacteremia in the 3-month-old to 3-year-old age group. Pediatr Ann 1993;22:484, 487-493. 8. Fleisher GR, Rosenberg N, Vinci R, et al: Intramuscular versus oral antibiotic therapy for the prevention of meningitis and other bacterial sequelae in young, febrile children at risk for occult bacteremia. J Pediatr 1994;124:504-512. 9. Myers MG, Wright PF, Smith AL, et al: Complications of occult pneumococcal bacteremia in children. J Pediatr 1974;84:656-660. 10. Bratton L, Teele DW, Klein JO: Outcome of unsuspected pneumococcemia in children not initially admitted to the hospital. J Pediatr 1977;90:703-706. 11. Feder HM Jr: Occult pneumococcal bacteremia and the febrile infant and young child. Clin Pediatr 1980;19:457-462. 12. Carroll WL, Farrell MK, Singer JI, et al: Treatment of occult bacteremia: A prospective randomized clinical trial. Pediatrics 1983;72:608-612. 13. Baron MA, Fink HD, Cicchetti DV: Blood cultures in private pediatric practice: An eleven-year experience. Pediatr Infect Dis J 1989;8:2-7. 14. Woods ER, Merola JL, Bithoney WG, et al: Bacteremia in an ambulatory setting. Improved outcome in children treated with antibiotics. Am J Dis Child 1990;144:1195-1199. 15. Baraff LJ, Bass JW, Fleisher GR, et al: Practice guideline for the management of infants and children 0 to 36 months of age with fever without source. Pediatrics 1993;92:1-12. 16. Harper MB, Bachur R, Fleisher GR: Effect of antibiotic therapy on the outcome of outpatients with unsuspected bacteremia. Pediatr Infect Dis J 1995;14:760-767. 17. Rothrock SG, Harper MB, Green SM, et al: Do oral antibiotics prevent meningitis and serious bacterial infections in children with Streptococcus pneumoniae occult bacteremia? A meta-analysis. Pediatrics 1997;99:438-444. 18. Bass JW, Steele RW, Wittler RR, et al: Antimicrobial treatment of occult bacteremia: A multicenter cooperative study. Pediatr Infect Dis J 1993;12:466-473. 19. Baraff LJ, Lee SI: Fever without source: Management of children 3 to 36 months of age. Pediatr Infect Dis J 1992;11:146-151.
ANNALS OF EMERGENCY MEDICINE
31:6 JUNE 1998
OCCULT PNEUMOCOCCAL BACTEREMIA Kuppermann, Fleisher & Jaffe
20. Baraff LJ, Oslund S, Prather M: Effect of antibiotic therapy and etiologic microorganism on the risk of bacterial meningitis in children with occult bacteremia. Pediatrics 1993;92:140-143. 21. Todd JK: Childhood infections: Diagnostic value of peripheral white blood cell and differential cell counts. Am J Dis Child 1974;127:810-816. 22. Weitzman M: Diagnostic utility of white blood cell and differential cell counts. Am J Dis Child 1975;129:1183-1189.
Reprint no. 47/1/90125 Address for reprints: Nathan Kuppermann, MD, MPH Division of Emergency Medicine and the Department of Pediatrics
23. McCarthy PL, Jekel JF, Dolan TF Jr: Temperature greater than or equal to 40˚ C in children less than 24 months of age: A prospective study. Pediatrics 1977;59:663-668.
University of California, Davis, Medical Center
24. McCarthy PL, Jekel JF, Dolan TF Jr: Comparison of acute-phase reactants in pediatric patients with fever. Pediatrics 1978;62:716-720.
PSSB Building
2315 Stockton Boulevard
25. Morens DM: WBC count and differential: Value in predicting bacterial diseases in children. Am J Dis Child 1979;133:25-27.
Sacramento, CA 95817
26. Teele DW, Marshall R, Klein JO: Unsuspected bacteremia in young children: A common and important problem. Pediatr Clin North Am 1979;26:773-784.
Fax 916-734-7950
27. Waskerwitz S, Berlelhamer JE: Outpatient bacteremia: Clinical findings in children under two years with initial temperatures of 39.5˚ C or higher. J Pediatr 1981;99:231-233.
916-734-1535 E-mail [email protected]
28. Rasmussen NH, Rasmussen LN: Predictive value of white blood cell count and differential cell count to bacterial infections in children. Acta Paediatr Scand 1982;71:775-778. 29. McCarthy PL, Sharpe MR, Spiesel SZ, et al: Observation scales to identify serious illness in febrile children. Pediatrics 1982;70:802-809. 30. Bennish M, Beem MO, Ormiste V: C-reactive protein and zeta sedimentation ratio as indicators of bacteremia in pediatric patients. J Pediatr 1984;104:729-732. 31. Liu C-H, Lehan C, Speer ME, et al: Early detection of bacteremia in an outpatient clinic. Pediatrics 1985;75:827-831. 32. Crocker PJ, Quick G, McCombs W: Occult bacteremia in the emergency department: Diagnostic criteria for the young febrile child. Ann Emerg Med 1985;14:1172-1177. 33. Jaffe DM, Fleisher GR: Temperature and total white blood cell count as indicators of bacteremia. Pediatrics 1991;87:670-674. 34. Kramer MS, Tange SM, Mills EL, et al: Role of the complete blood count in detecting occult focal bacterial infection in the young febrile child. J Clin Epidemiol 1993;46:349-357. 35. StataCorp: Stata Statistical Software. Release 5.0. College Station, TX: Stata Corporation, 1997. 36. Browner WS, Newman TB, Cummings SR. Designing a new study: III. Diagnostic tests, in Hulley SB, Cummings SR (eds): Designing clinical research. Baltimore: Williams & Wilkins, 1988:90. 37. Murray DL, Zonana J, Seidel JS, et al: Relative importance of bacteremia and viremia in the course of acute fevers of unknown origin in outpatient children. Pediatrics 1981;68:157-160. 38. Baker RC, Tiller T, Bausher JC, et al: Severity of disease correlated with fever reduction in febrile infants. Pediatrics 1989;83:1016-1019. 39. Hosmer DW, Lemeshow S: Applied logistic regression. New York: John Wiley & Sons, 1989. 40. Teach SJ, Fleisher GR: Efficacy of an observation scale in detecting bacteremia in febrile children three to thirty-six months of age, treated as outpatients. J Pediatr 1995;126:877-881. 41. McCarthy PL, Lembo RM, Fink HD, et al: Observation, history, and physical examination in diagnosis of serious illnesses in febrile children ≤24 months. J Pediatr 1987;110:26-30. 42. Dershewitz RA, Wigder HN, Wigder CM, et al: A comparative study of the prevalence, outcome, and prediction of bacteremia in children. J Pediatr 1983;103:352-358. 43. Adams WG, Deaver KA, Cochi SL, et al: Decline of childhood Haemophilus influenzae type b (Hib) disease in the Hib vaccine era. JAMA 1993;269:221-226.
JUNE 1998
31:6
ANNALS OF EMERGENCY MEDICINE
6 8 7