C-Reactive Protein?
Aaron E. Carroll, MDMichael Silverstein, MD
Robert Wood Johnson Clinical Scholars Program and Department of Pediatrics
University of Washington
Seattle, WA 98195-7183
To the Editor.—
Although we appreciate the recent publication of Pulliam et al,1 several issues need clarification. Consistent with much of the existing literature on the febrile infant, Pulliam et al employ a receiver operating characteristic analysis for quantitative C-reactive protein (CRP); from this, they find that CRP has better sensitivity, specificity, and predictive value than tests currently available. They conclude, " Quantitative CRP concentration is a valuable laboratory test in the evaluation of febrile young children who are at risk for OB [occult bacteremia] and SBI [serious bacterial infection] ... " We, however, conclude that CRP represents yet another laboratory test that offers little help to the clinician evaluating the febrile infant.
As the authors note, the 18% prevalence of SBI among their subjects is higher than that found in previous studies. Most previous studies have found a 2% to 3% prevalence of OB2 and 2% to 8% of urinary tract infection (UTI)3 among similar series of febrile children. The higher prevalence reported by Pulliam et al is of concern for a significant selection bias towards sicker children. Although the authors acknowledge this limitation, they use their 18% prevalence in reporting the study’s most important finding: the posttest probability of SBI after CRP quantification. Because these calculations are based on an 18% prevalence, they may be applicable only to this atypical population.
Although the positive and negative predictive values of the CRP assay must, by definition, be calculated using the 18% prevalence, we point out 2 additional issues. First, the reported 95% confidence intervals of the positive and negative predictive values for CRP, white blood cell count (WBC), and absolute neutrophil count (ANC) overlap. This implies that if the study were repeated, the results could easily show CRP to be equivalent or even less useful than WBC or ANC. Second, the reported posttest probability is—in strict Bayesian terms—identical to the positive predictive value when the 18% prevalence is used as pretest probability. Therefore, although the reported likelihood ratios are, as claimed, insulated from prevalence, the reported posttest probabilities are as much governed by prevalence as the predictive values.
Although the authors’ receiver operator characteristic (ROC) curve suggests that CRP has the greatest area under the curve (AUC), it is unclear how the individual data points for any of the 3 parameters were derived. The authors include a number of data points to define the CRP curve, but far fewer for the WBC and ANC curves. Although it is unlikely that more points would have made the other 2 ROC curves as impressive as the CRP curve, it may have changed them somewhat. This is important because the 95% confidence interval for all 3 AUCs overlap, suggesting no statistically significant difference among them.
Most concerning, however, is the authors’ willingness to follow convention and rely inappropriately on the ROC analysis to show the worthiness of quantitative CRP in this setting. We argue that an ROC analysis is of limited use in this particular arena. The authors state, based on their ROC curve, that a CRP of 7 "simultaneously maximizes both sensitivity and specificity" of the test. This is untrue. By definition, the sensitivity of a test increases at the expense of its specificity, and vice versa. The authors have merely chosen a cutoff point at which the total number of misclassification errors is minimized—that is, the point at which the sum of false-positives (a function of specificity) and false-negatives (a function of sensitivity) is less than at any other point.
Underlying the use of an ROC analysis, therefore, is the assumption that the scientific or clinical value of a false-positive is worth the same as a false-negative. In the setting of the febrile infant, this assumption implies that missing a case of SBI is of equal clinical significance to unnecessarily treating a child falsely suspected of having one. Although minimizing unnecessary treatment is important, most clinicians would agree the errors are of different importance. Therefore, although an ROC analysis is important when evaluating a test’s overall performance characteristics, we argue that in this case—in which ruling out bad disease is of paramount importance—sensitivity is the more important test characteristic.
Finally, we question the authors’ decision to analyze together children with UTIs and OB. Urinalysis has been shown to be a highly sensitive and specific test for diagnosing UTIs.4 Most practicing physicians would obtain a urinalysis in a young febrile child from whom blood is going to be drawn for culture, complete blood count, or perhaps CRP. This means that a large proportion of UTIs will be diagnosed at the point of initial contact, leaving only those with OB requiring a rapid, sensitive prognostic test. These are the patients for whom CRP might be useful. It would perhaps be more useful to study this subset of patients.
According to the data presented, CRP should not change our current practice. Until a test approaches the sensitivity of blood culture and is significantly faster, none will.
REFERENCES
- Pulliam PN, Attia MW, Cronan KM. C-reactive protein in febrile children 1 to 36 months of age with clinically undetectable serious bacterial infection.
Pediatrics.2001; 108
:1275
–1279
[Abstract/Free Full Text] - Alpern EA, Alessandrini EA, Bell LM, Shaw KN, McGowan KL. Occult bacteremia from a pediatric emergency department; current prevalence, time to detection, and outcome.
Pediatrics.2000; 106
:505
–511
[Abstract/Free Full Text] - Shaw KN, Gorelick MH, McGowan KL, McDaniel Yakscoe N, Schwartz JS. Prevalence of urinary tract infection in febrile young children in the emergency department. Pediatrics.2001; 102(2) . Available at: http://www.pediatrics.org/cgi/content/full/102/2/e16
- Shaw KN, McGowan KL, Gorelick MH, Schwartz JS. Screening for urinary tract infection in infants in the emergency department: which test is best? Pediatrics.1998; 101(6) . Available at: http://www.pediatrics.org/cgi/content/full/101/6/e1
C-Reactive Protein?
Patrick N. Pulliam, MD*Department of Pediatrics
Temple University School of Medicine
Temple University Children’s Medical Center
Philadelphia, PA
Magdy W. Attia, MD
Department of Pediatrics
Division of Emergency Medicine
Jefferson Medical College
duPont Hospital for Children
Wilmington, DE
Kathleen M. Cronan, MD
Department of Pediatrics
Division of Emergency Medicine
Jefferson Medical College
duPont Hospital for Children
Wilmington, DE
In Reply.—
We appreciate the interest in and thoughtful comments on our article1 by Drs Carroll and Silverstein. We would like to address several of their concerns.
The authors are correct in stating that the posttest probability depends on prevalence. However, likelihood ratios are insulated from prevalence. Likelihood ratios are becoming the preferred method for evaluating and using diagnostic tests for this very reason. As we pointed out in the article, for any particular patient a clinician has some idea as to the pretest probability of the presence of serious bacterial infection (SBI). The clinician’s estimate may be based on clinical appearance, other laboratory studies, historical information, knowledge of the population, epidemiologic factors, etc. For example, the pretest probability of SBI in a 6-week-old infant with a fever of 105°F and irritability in April is much higher than that of a cheerful 30-month-old toddler with fever of 102°F and copious rhinorrhea in January. Based on the clinician’s estimate of pretest probability, the posttest probability of SBI can be calculated based on the laboratory result and knowledge of the likelihood ratio. The use of multilevel likelihood ratios allows clinicians to calculate the likelihood of a diagnosis based on a range of laboratory values, not just on a positive or negative scale based on a particular cutoff value. Likelihood ratios also offer the opportunity to compare the usefulness of diagnostic tests with each other in a much more objective fashion than sensitivity, specificity, and predictive values, which are based on prevalence.2–4
Receiver operator characteristic (ROC) curve analysis was used to determine a cutoff point that, as the authors correctly point out, minimizes the total number of false-positives and false-negatives. We are not suggesting that this value should be used in every clinical situation. On the contrary, ROC curves allow clinicians to choose a value to the left or to the right of the " optimal" value, depending on whether sensitivity or specificity is more important in a given clinical situation. However, it would be preferable to use the likelihood ratios in actual clinical situations, allowing for a more accurate estimate of the likelihood of SBI. In our graph, we showed the ROC curves for white blood cell count (WBC) and absolute neutrophil count (ANC) with fewer data points than on the curve for C-reactive protein (CRP). This was intended for the ease of producing the graph and of interpretation. Adding more data points to the WBC and ANC curves resulted in nearly identical area under the curve and a 95% confidence interval.
The authors are correct in stating that the 95% confidence intervals for CRP, WBC, and ANC overlap. Larger studies are needed to more accurately determine the usefulness of these screening tests and, as we stated, to determine the usefulness of these tests for specific bacterial illnesses, eg, occult bacteremia and urinary tract infections (UTIs). We have elaborated in the "Discussion" section on our reasoning to think of occult bacteremia, UTI, and occult pneumonia as a single entity when the clinician is confronted with a young child who has fever with clinically undetectable source. We refer the authors back to that section.
Finally, we are not suggesting that CRP or any other screening test should replace blood culture. Until a test approaches the sensitivity of blood culture and is significantly faster, screening tests may aid in the overall management of these patients pending the outcome of definitive laboratory studies. Perhaps one day, some test will reach this elusive goal.
FOOTNOTES
* Dr Pulliam’s current address is ABC Pediatrics, Fayetteville, GA 30214-2049. 
REFERENCES
- Pulliam PN, Attia MW, Cronan KM. C-reactive protein in febrile children 1 to 36 months of age with clinically undetectable serious bacterial infection. Pediatrics.2001; 108 :1275 –1279
- Sackett DL, Haynes RB, Guyatt GH, Tugwell P. Clinical Epidemiology: A Basic Science for Clinical Medicine. 2nd ed. Boston, MA: Little, Brown and Company; 1991:119–13
- Jaeschke R, Guyatt GH, Sackett DL for the Evidence-Based Medicine Working Group. Users’ guides to the medical literature. VI. How to use an article about a diagnostic test. A. Are the results of the study valid?
JAMA.1994; 271
:389
–391
[Abstract/Free Full Text] - Jaeschke R, Guyatt GH, Sackett DL for the Evidence-Based Medicine Working Group. Users’ guides to the medical literature. VI. How to use an article about a diagnostic test. B. What are the results and will they help me in caring for my patients?
JAMA.1994; 271
:703
–707
[Abstract/Free Full Text]
PEDIATRICS (ISSN 1098-4275). ©2002 by the American Academy of Pediatrics
CiteULike 
Connotea 
Del.icio.us 
Digg 
Facebook 
Reddit 
Technorati 
Twitter What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
