Objective. To determine the utility of plasma levels of tumor necrosis factor-α (TNF), interleukin 1β (IL-1), and interleukin 6 (IL-6) in the prediction of occult bacteremia in febrile, young children.
Study Design. Prospective, case-control study conducted in a large, urban, children's hospital emergency department. Eligibility criteria were: 0 to 36 months of age, febrile, nontoxic appearing, immunocompetent, no apparent bacterial source for fever on physical examination, and blood culture obtained. Additional blood, procured at the time of the blood culture, was analyzed by enzyme-linked immunosorbent assay for TNF, IL-1, and IL-6. Children with positive blood cultures for pathogenic bacteria served as cases. Two age-matched controls for each case were selected from the children with negative cultures.
Results. Out of 1329 enrollees, 33 cases and 66 controls were evaluated. IL-6 levels were significantly higher for the cases than controls but with moderate overlap in their ranges. TNF and IL-1 levels were not significantly different between cases and controls. Height of fever, duration of fever, acute illness observation score, absolute band count, and white blood cell count were all much less predictive of bacteremia than either IL-6 or absolute neutrophil count (ANC). The optimum IL-6 threshold value had a sensitivity of 88%, a specificity of 70%, a positive predictive value (PPV) of 7.0%, a negative predictive value (NPV) of 99.6%, and an odds ratio (OR) of 16.7 (95% confidence interval [CI], 4.8–71.6). The optimum ANC threshold value had a sensitivity of 82%, a specificity of 74%, a PPV of 7.5%, a NPV of 99.4%, and an OR of 12.8 (95% CI, 3.2–59.7). The best predictor was a combination of IL-6 and ANC. It had a sensitivity of 100%, a specificity of 78%, a PPV of 10.4%, a NPV of 100%, and an OR which is undefined because of the 100% sensitivity (95% CI, 33.0-∞). For comparison, a WBC >15 × 109 cells/L had a sensitivity of 48%, a specificity of 79%, a PPV of 5.5%, a NPV of 98.3%, and an OR of 3.5 (95% CI, 1.1–10.7).
Conclusions. In febrile children 0 to 36 months of age, IL-6 levels may be helpful in the prediction of occult bacteremia, but TNF and IL-1 levels are not. IL-6 levels alone or notably when combined with an ANC were more predictive of occult bacteremia than traditional tests and clinical criteria. The wide range in the IL-6 values for cases and controls detracts from the precision of the findings. The lack of rapid processing and clinical availability of IL-6 assays hampers its present application. However, despite these drawbacks and given the poor ability of traditional clinical and laboratory criteria to predict occult bacteremia, these results suggest a possible future role for IL-6 in predicting occult bacteremia when rapid assays become available.
The precise prediction of bacteremia in a febrile, nontoxic appearing, young child at the time of an outpatient visit continues to be a difficult challenge. White blood cell counts (WBC), absolute band counts (ABC), height of fever, duration of fever, and clinical scoring systems are standard criteria to predict occult bacteremia, but they are imprecise.1–3 Automated blood culture techniques4,,5 offer more precision than traditional laboratory tests and clinical scoring systems, but still are unable to detect bacteremia early enough to allow diagnosis at the time of the initial outpatient visit. For swift and precise detection of bacteremia, an in vitro study using a polymerase chain reaction (PCR) for bacterial DNA seems promising.6 However, Isaacman and co-workers7 reported their PCR-based assay for pneumococcal DNA to be less sensitive and specific for predicting pneumococcal bacteremia than the sensitivity and specificity reported by other authors for the WBC.1,,3,8 Consequently, a more reliable method for predicting occult bacteremia in this population is still needed.
Tumor necrosis factor-α (TNF), interleukin 1β (IL-1), and interleukin 6 (IL-6) are well known and important early mediators in the host's initial response to bacterial infection.9–13Because occult bacteremia is considered the first stage of invasive disease, TNF, IL-1, and IL-6 may serve as markers for occult bacteremia. The role of TNF, IL-1, and IL-6 in predicting bacteremia and/or survival in toxic appearing children has been evaluated numerous times, but we are aware of just one study involving occult bacteremia.14 That study assessed TNF and IL-1 levels in young children with occult Streptococcus pneumoniaebacteremia and reported improved predictive values over traditional tests when TNF or IL-1 levels were used in conjunction with a WBC. The study was limited by small numbers, addressed only S pneumoniae bacteremia, and did not include IL-6, which seems to be a major cytokine involved in the release of other acute phase reactants.12,,13,15
In this study, we hypothesize that TNF, IL-1, and IL-6 will serve as predictors of occult bacteremia. To test our hypothesis, we measured TNF, IL-1, and IL-6 levels in the plasma of nontoxic appearing, febrile, young children, with and without bacteremia, and determined the predictive values of these cytokines for occult bacteremia in this population. We compared these results to the predictive values of traditional tests and clinical data obtained on these same children.
Patient and Plasma Samples
This study is a prospective, nested, case-control study conducted between June 1995 and June 1998 in the emergency department of the Children's Hospital of Wisconsin, a large, freestanding, tertiary, children's hospital. Eligibility criteria for the study included: 0 to 36 months of age, rectal temperature in the emergency department or a reliable home temperature ≥39°C (≥38°C for children 0–2 months of age), immunocompetent, nontoxic appearing,3, no significant bacterial source for their fever on physical examination (otitis media was not considered a significant bacterial source1–3), no steroids or antibiotics within the preceding 24 hours, and having a blood culture obtained. The need for additional tests was left to the discretion of the treating physician and no additional tests were required to be eligible for the study. After written informed consent, as approved by the institutional human research review board, and at the same time as phlebotomy for the blood culture, an additional 1 to 5 mL of blood was obtained for the study. The additional blood was drawn into standard ethylenediaminetetraacetic acid glass tubes with 1.0 trypsin inhibitor unit (Aprotinin; Sigma, St Louis, MO) preinjected. The plasma was separated, placed into 500-μL polypropylene tubes (Baxter, McGraw Park, IL), and stored at −80°C within 30 minutes. Demographic data, home and emergency department temperatures, time course of fever, recent use of antibiotics or steroids, and acute illness observation score16 were recorded. The results of any additional tests, such as WBC, urine culture, and spinal fluid cultures, were recorded.
Cases were defined as those patients whose blood culture grew pathogenic bacteria. Based on the published definition of occult bacteremia17–20 and eligibility criteria for the study, all our cases were considered to have occult bacteremia. Controls were chosen from among the children with negative blood cultures. Children were excluded from being a control if a chest radiograph demonstrated pneumonia, as interpreted by a pediatric radiologist, or any additional bacterial cultures were done (eg, urine and cerebrospinal fluid cultures) and were positive. Two age-matched (by days old) controls for each case were chosen from among the children eligible to be controls. If >2 potential controls were the same age as a case, priority was given to the ones enrolled closest to the time/date of enrollment of the case. If <2 potential controls were the same age as a case, those closest in age to the case were given preference. All cases and controls were chosen before determination of cytokine levels.
TNF, IL-1, and IL-6 levels were quantified using a standard double-sandwich, enzyme-linked immunosorbent assay protocol similar to that recommended by R & D Systems Inc (Minneapolis, MN). All cytokine standards and matched-pair anticytokine antibodies were purchased from R & D Systems Inc. The 96-well Immulon-2 microtiter plates were purchased from Fischer Scientific (Itasca, IL). All capture antibodies were diluted to a concentration of 4 μg/mL in phosphate buffered saline (PBS) containing 1% bovine serum albumin (Sigma, St Louis, MO). The detecting antibodies were used at concentrations of: 0.2 μg/mL for TNF, 0.1 μg/mL for IL-1, and 0.025 μg/mL for IL-6. Duplicate standards were run in serial 2-fold dilutions. The IL-6 and IL-1 standards were started at a concentration of 1 ng/mL and serially diluted to 0.00195 ng/mL. The TNF standard was started at a concentration of 5 ng/mL and serially diluted to 0.01 ng/mL. The plasma specimens were diluted as necessary to between 2- and 200-fold and added to the wells in duplicate. The standards, detecting antibodies, and plasma specimens were all diluted in PBS with 1% bovine serum albumin and 0.05% Tween 20 (Sigma, St Louis, MO). Four washes were done between each step with a PBS and 0.05% Tween 20 solution. A 1:5000 (v/v) streptavidin peroxidase (Sigma, St Louis, MO) solution, chromogen substrate (0.06% o-phenylene diamine, 0.015% of hydrogen peroxide in 0.1 M of citrate buffer, pH 4.5) and 2 M of sulfuric acid solution comprise the color development steps. The absorbency at 490 nm was read in an automated MR 700 enzyme-linked immunosorbent assay microplate reader (Dynatech Lab, Chantilly, VA). The minimum detectable concentration of cytokines was 10 pg/mL for TNF, 2 pg/mL for IL-1, and 4 pg/mL for IL-6. Standard curves were constructed and cytokine concentrations in the specimens were determined using GraphPad Prism version 2.0 (GraphPad Software, Inc, San Diego, CA).
Nonparametric data were analyzed using the Mann WhitneyU test. Normally distributed data were examined using thet test. The acute illness observation score was analyzed as continuous data21 with the t test. Dichotomous data were analyzed using χ2. ORs and their 95% CIs were calculated using Epi Info version 5.0. To evaluate and compare the limits of IL-6, absolute neutrophil count (ANC), and WBC in discriminating bacteremic and nonbacteremic children, receiver-operator characteristic curves were constructed in the usual manner and the area under the curves calculated.22 Scatter plots were used to evaluate combinations of laboratory and clinical data in predicting bacteremia. Positive predictive values (PPV) and negative predictive values (NPV) were calculated in the standard method23using a prior predictive value of occult bacteremia of 2.5%. Where appropriate an α of .05 and a β of .2 were applied. A Pvalue <.05 was considered significant. Power analysis for the nonparametric data of cytokine levels was approximated using nQuery Advisor Release 3.0 (Statistical Solutions Ltd., Cork, Ireland).
From 1329 children enrolled, 33 met our definition of cases, yielding an incidence of occult bacteremia of 2.5%. Two controls were selected for each case, generating a total of 66 controls. Eight of the 33 cases were <60 days old. The cases grew the following bacteria in their blood cultures: 19 S pneumoniae, 5 Escherichia coli, 2 Salmonella typhi, 2 Enterococcus faecalis, and 1 each of: Staphylococcus aureus,Moraxella catarrhalis, Yersinia enterocolitica, group B streptococcus, and Citrobacter diversus. One child with pneumococcal bacteremia subsequently developed meningitis within 24 hours of enrollment and was treated without apparent sequela. Of the 5 children with E coli bacteremia, 2 were <90 days old and all 5 grew E coli from their urine. The child with S aureus bacteremia also had a urine culture positive for S aureus and a negative bone scan. The 14-month-old child withM catarrhalis bacteremia had a repeat negative blood culture and had no sequela without antibiotic treatment. The neonate withC diversus bacteremia suddenly deteriorated 12 hours after enrollment and subsequently died 10 days later with meningitis and sepsis from this same organism. The 3 neonates with bacteremia involving E faecalis or Y enterocolitica had negative urine and cerebrospinal fluid cultures. All 3 children were treated with antibiotics and subsequent blood cultures were negative.
From the 1329 children enrolled, 80 were diagnosed with otitis media. Two of the 80 met criteria for serving as cases and 6 met criteria for serving as controls. Gastroenteritis was diagnosed in 20 children (1 serving as a case, 1 as a control), pharyngitis in 1 child (no cases, no controls), and bronchiolitis in 27 (no cases, 2 controls). All the rest of the children had fever without source as their primary emergency department discharge diagnosis. No enrolled children were diagnosed with croup, varicella, stomatitis, or other easily identifiable viral infection in the emergency department.
The demographic data and routinely reported clinical and laboratory results used in distinguishing bacteremic and nonbacteremic children for the cases and controls are presented in Table 1. There was no significant difference between them for age, gender, race, duration of fever, and ABC. Cases had higher fevers, acute illness observation score, ANC, and WBC as compared with controls. The treating physician for the enrolled children did not always request additional tests, such as a WBC with a differential cell count. Therefore, 25 of the 33 cases and 57 of the 66 controls had a WBC available for analysis and an ABC and ANC were available for analysis in 22 of the 33 cases and 46 of the 66 controls.
Median IL-6 levels of the cases were significantly higher than those of controls (P < .001). However, median TNF and IL-1 levels were not significantly different between cases and controls (Table 2). In addition, in evaluating scatter plots (data not shown) no useful trends were appreciated between the various cytokine levels to each other (eg, IL-6 versus TNF) or between the various cytokine levels to other individual demographic, clinical data, and ABC (eg, IL-6 versus age).
The cases were also assessed for subgroup trends. No significant differences were found between the cytokine levels for the cases with Gram-positive versus Gram-negative bacteria (data not shown). However, given the relatively small sample size of Gram-positive and Gram-negative bacteria, the median levels of the cytokines and wide range of cytokine levels, the power to detect a significant difference was <20%. The comparison of the cytokine levels, demographic, clinical, and laboratory data for S pneumoniae cases versus cases without S pneumoniae bacteremia were also not significant except S pneumoniae bacteremic patients were significantly older than the bacteremic patients without S pneumoniae (mean 412 days versus 211 days, P = .012).
Receiver-operator characteristic curves were constructed for the ANC and IL-6 levels and areas under each curve calculated (Fig 1). These curves were then used to determine the optimal threshold value for IL-6 and ANC. The optimal threshold value is defined as the value represented by the point on the curve at which the curve passes closest to the left upper corner of the graph.22 Based on this definition and using calipers to determine the ideal point on each curve, the optimal threshold value for IL-6 and ANC are >95 pg/mL and >7200 cells/mm3, respectively. The sensitivity, specificity, PPV, and NPV of an IL-6 > 95 pg/mL, an ANC > 7200 cells/mm3 are presented in Table 3. The odds ratios (ORs) and 95% confidence intervals (CIs) are included as a measure of precision and strength of association between the test and occult bacteremia.
The WBC is a widely used laboratory test for predicting occult bacteremia at the time of an outpatient visit. To allow for comparison of IL-6 and ANC results with the WBC, a receiver-operator characteristic curve was also constructed for the WBC and area under each curve calculated (Fig 1). For further comparison, the sensitivity, specificity, PPV, and NPV for the extensively used threshold value of a WBC > 15 × 109 cells/L are presented in Table 3.
From visualization of a scatter plot of ANC versus IL-6 levels, the optimum combination of threshold values for a positive test was found to be an IL-6 level of >65 pg/mL and an ANC of >5000 cells/mm3. These values are represented by the dotted lines in Fig 2 and the right upper quadrant of this figure represents those patients with a positive test for the combination of IL-6 > 95 and ANC > 5000. The sensitivity, specificity, PPV, and NPV of this combination are reported in Table 3.
It is possible that the 3 children with either E faecalis orY enterocolitica in their blood cultures may represent contamination instead of a true bacteremia. Therefore, the data were reanalyzed with the 3 children and their 6 age-matched controls eliminated. No significant change was demonstrated in any of the demographic, clinical, or laboratory parameters evaluated for predicting bacteremia. The largest change was in the PPV of the ANC which decreased only 0.4%. Of interest the predictive values of the combination of IL-6 and ANC remained unchanged.
S pneumoniae was the etiologic bacterium in the majority of the cases in our study population. In addition, one previous study evaluated the utility of TNF and IL-1 levels in predicting occultS pneumoniae bacteremia in children with fever.14 Therefore our data were reanalyzed using only the cases with S pneumoniae and their age-matched controls. The overall findings were no different from for all cases and controls except that the WBC had improved predictive values, PPV = 7.2% and NPV = 99.1%. Of special interest, the IL-6, ANC, and their combination remained the best predictors of occult bacteremia (IL-6: PPV = 7.1%, NPV = 99.8%; ANC: PPV = 8.9%, NPV = 99.7%; and combination: PPV = 9.7%, NPV = 100%). TNF and IL-1 levels were not significantly different between cases and controls; hence, they were not as predictive as the WBC, IL-6, or ANC. The power to detect a significant difference between the medians of the cases and controls for either TNF or IL-1 was <20%. In contrast the power for IL-6 in these same cases and controls was ∼80%.
The bacteremic cases without S pneumoniae and their age-matched controls were also reanalyzed. The overall findings were again no different from all cases and controls except that the WBC had decreased predictive values (PPV = 3.7% and NPV = 98.0%). The best predictor was again the combination of IL-6 and ANC, PPV = 9.7% and NPV = 100%.
The results presented above indicate that in our study population IL-6 levels, but not IL-1 and TNF levels, are significantly higher in the children with occult bacteremia than in those with apparent viral infection. Interleukin 6 seems more advantageous in predicting bacteremia than WBC, and equivalent to the ANC. This observation is supported by the equality of the receiver-operator characteristic curves and areas under the curves for IL-6 and ANC and the lower curve and lesser area for WBC as shown Fig 1. However, there is overlap of the IL-6 levels between the bacteremic and nonbacteremic groups. This overlap detracts from the utility of IL-6 in distinguishing bacteremic from nonbacteremic patients and results in the modest sensitivity and specificity presented in Table 3.
From the traditional clinical and laboratory data evaluated in this study, the ANC was found to be the best means of predicting occult bacteremia at the time of the initial visit. The sensitivity and specificity for the WBC in our study was similar to those reported in other studies,24–26 but was not as predictive as the ANC in our study population. When only S pneumoniae cases and their matched controls were analyzed the advantage of the ANC over the WBC was lessened. Three other recent reports1,,8,14 on occult S pneumoniae bacteremia describe similar findings for the ANC and WBC.
When IL-6 and ANC were evaluated in combination, there was some improvement in the predictive values over either alone. As described earlier, the optimum combination of IL-6 and ANC was established using visualization of a scatter plot and the positive test result for this combination is represented by the right upper quadrant in Fig 2. The combination of an IL-6 > 65 pg/mL and ANC > 5000 cells/mm3 offers an increase in the PPV and NPV over either test alone. The receiver-operator characteristic areas under the curve22 and the statistically significant difference in median values between cases and controls support the significant role that IL-6 and ANC play in the prediction of occult bacteremia. However, the overlap of IL-6 and ANC values between cases and controls, the retrospective nature of choosing cutoff values and the wide range of the confidence levels of the ORs as shown in Table 3detracts from the significance of this determination.
Our findings are similar to those of other studies in regard to the significance of height of fever,1,,14,24,27 duration of fever,2,,14 and acute illness observation score14,,18,28 as predictors of occult bacteremia. The clinical utility of any of these factors for distinguishing occult bacteremia is inadequate. Although the height of fever in our study was statistically significant, the temperature difference between the 2 groups (means of 40.1°C for cases and 39.8°C for controls) was so small that the clinical utility beyond the study entry eligibility criteria remains poor. Similarly, the difference between acute illness observation scores for cases and controls (means of 9 and 8, respectively) was so minor as to be clinically indistinguishable.
There are 2 potential problems with this study's inclusion criterion. First, there is no way to ensure that every single patient that was eligible for enrollment was approached. Therefore, it may be prudent to view the study as a nonconsecutive series. Nonetheless, the incidence of bacteremia and the clinical and traditional laboratory findings in this study are consistent with previous studies1–3,14suggesting that the study group offers a good representation of the target population. Second, the inclusion criterion does not exclude children with presumably easily identifiable viral sources for their fever (eg, varicella, bronchiolitis, or pharyngitis) as other studies have done.1–3 However, all our findings for the clinical and traditional laboratory data for the cases and controls were comparable to previous reports. Therefore, differences in inclusion criterion had no demonstrable impact on the study results.
There are 2 issues regarding age for this study that warrant remark. First, the influence of age on cytokine levels within our study population has never been addressed. No significant trend was noted in evaluation of the scatter plot of the 3 cytokines versus age. However, the limited number of cases and controls and the wide range of cytokine levels make formulating any association between age and cytokine levels inconclusive. Future studies may need to focus on this question.
The second issue for age is about which age groups should be included in the study and/or analyzed separately. We elected to include not only the 60-days-old to 36-months-old children but also the younger children. Occult bacteremia (ie, bacteremia which occurs in the nontoxic appearing child without bacterial source on physical examination17–20) is a significant issue for both the 60-days- to 36-months-old age group and the 0- to 60-days-old age group. The limited number of cases <60 days old does not lend itself to useful separate analysis. However, now that this study has shown a potentially promising role for IL-6 in predicting occult bacteremia future studies may need or want to address the various age groups separately.
The 3 neonates with Y enterocolitica or E faecalis bacteremia represent unusual but known pathogens. Yet, the lack of additional positive cultures and clinical courses of these patients makes it difficult to ascertain whether these organisms represented true pathogens or contaminants. However, separate analysis of the other 30 cases without Y entercolitica or E faecalis and their matched controls did not change any of our conclusions. Therefore, the 3 unusual pathogens are included in the final analysis and conclusions.
An unusual finding in our study is that S pneumoniaerepresents only 57% of the cases. This is in comparison to the more commonly reported apportionment of 85%.1–3 The cited studies included much larger sample sizes and excluded children 0 to 3 months of age. However, if only children 3 to 36 months of age are addressed in our study the percentage of S pneumoniaechanges, but remains low at 70%. The reasons for this difference are unknown.
Our controls were chosen from among the enrolled children who had negative blood cultures. In addition we also required that the controls not have any other positive bacterial cultures or a chest radiograph consistent with bacterial pneumonia. We specifically eliminated bacterial sources from serving as controls to create a more homogeneous control group. However, now that this study has shown a potentially promising role for IL-6 in predicting occult bacteremia, future, more focused studies may want to address, in a similar manner to this study, the IL-6 response of bacteremic and nonbacteremic children having these focal bacterial sources.
A previous study involving a population similar to our present study found TNF and IL-1 levels to be significantly different between children with and without occult S pneumoniae bacteremia and as predictive of occult S pneumoniae bacteremia as a WBC.14 Our study failed to confirm their results. Our study involved 50% more cases of S pneumoniae bacteremia than the previous study (19 and 12, respectively) and similar to this study, the previous study also reported significant overlap in the TNF and IL-1 values for their children with and without bacteremia. The authors acknowledged the limitations of small sample size and significant overlap and even commented that a larger study would be needed before acceptance of their results as conclusive. The power analysis of our TNF and IL-1 levels would agree with the need for a much larger study. However, the excessively large overlap in either TNF or IL-1 levels for cases and controls makes the finding of a clinically significant difference unlikely and the need for future investigation of TNF and IL-1 for this purpose most likely fruitless.
In summary, IL-6 may be helpful in the prediction of occult bacteremia, but TNF and IL-1 are not. In predicting occult bacteremia in our study population, IL-6 alone or in combination with an ANC offered a significant increase over clinical data and ABC, and a modest increase over WBC. The retrospective nature of choosing optimum IL-6 threshold values, overlap in IL-6 values between cases and controls, and the wide range in IL-6 values for both cases and controls detract from the strength of IL-6 as a predictor of occult bacteremia. To date, assays for IL-6 take several hours to complete, thus eliminating it from practical application. Despite these limitations, given the poor ability of traditional clinical and laboratory criteria to predict occult bacteremia, these findings suggest a possible role for IL-6 in predicting occult bacteremia when rapid assays become clinically available.
This work was supported by Grant No. 3301.862 from the Children's Hospital of Wisconsin Foundation and by Grant No. 557 from the Clinical Research Center of the Medical College of Wisconsin.
We thank Abe Resnick for his assistance with the enzyme-linked immunosorbent assays and the emergency department staff of the Children's Hospital of Wisconsin for their help with enrollment of children.
- Received April 5, 1999.
- Accepted June 4, 1999.
Reprint requests to (R.T.S.) Children's Hospital Medical Center, Division of Emergency Medicine, 3333 Burnet Ave, OSB-4, Cincinnati, OH 45229. E-mail:
Presented, in part, at the AAP annual meeting; November 1, 1997; New Orleans, LA.
Dr Strait is a former faculty member of the Department of Pediatrics at the Medical College of Wisconsin.
- WBC =
- white blood cell count •
- ABC =
- absolute band count •
- PCR =
- polymerase chain reaction •
- TNF =
- tumor necrosis factor-α •
- IL-1 =
- interleukin 1β •
- IL-6 =
- interleukin 6 •
- PBS =
- phosphate buffered saline •
- ANC =
- absolute neutrophil count •
- PPV =
- positive predictive value •
- NPV =
- negative predictive value •
- OR =
- odds ratio •
- CI =
- confidence interval
- Rohner P,
- Pepey B,
- Auckenthaler R
- Stevens CM,
- Swaine D,
- Butler C,
- et al.
- McCabe KM,
- Khan G,
- Zhang YH,
- Mason EO,
- McCabe ERB
- Isaacman DJ,
- Zhang Y,
- Reynolds EA,
- Ehrlich GD
- Mantovani A
- ↵Richards CD, Gauldie J. Role of cytokines in acute-phase response. In: Aggarwal BB, Puri RK, eds. Human Cytokines: Their Role in Disease and Therapy. Cambridge, MA: Blackwell Science, Inc; 1995:253–269
- ↵Abbas AK, Lichtman AH, Pober JS. Cellular and Molecular Immunology. 3rd ed. Philadelphia, PA: WB Saunders; 1997:250–277, 342–361
- McCarthy PL,
- Sharpe MR,
- Spiesel SZ,
- et al.
- Rothrock SG,
- Harper MB,
- Green SM,
- et al.
- Garcia Pena BM, Harper MB, Fleisher GR
- ↵Browner WS, Newman TB, Cummings SR. Designing a new study: III. Diagnostic tests. In: Hulley SB, Cummings SR, eds. Designing Clinical Research. Baltimore, MD: Williams & Wilkins; 1988:87–97
- Jaffe DM,
- Fleisher GR
- Liu CH,
- Lehan C,
- Speer ME,
- et al.
- McCarthy PL,
- Grundy GW,
- Spiesel SZ,
- Dolan TF
- Baker MD,
- Avner JR,
- Bell LM
- Copyright © 1999 American Academy of Pediatrics