Secondary Logo

Journal Logo

Letters to the Editor

C-Reactive Protein is a Poor Predictor of Bacterial Pneumonia

Flood, Robert G. MD; Aronoff, Stephen C. MD

Author Information
The Pediatric Infectious Disease Journal: July 2008 - Volume 27 - Issue 7 - p 670-671
doi: 10.1097/INF.0b013e318174e0e8
  • Free

In Reply

Gutierrez et al raised a number of points regarding the methodology of the meta-analysis of Flood et al1 that recently appeared in the Pediatric Infectious Disease Journal. In a private communication, Dr. Korppi identified 43 patients out of 402 who may have been included in 2 studies.2,3 The duplication occurred because the population of 1 study was drawn from a population of 4 municipalities and the other study drew its subjects from hospitalized patients, of whom 43 were also residents within the 4 municipalities. This duplication represents only 11% of the total enrollment of both studies and only 3% of the entire study base; the sensitivity analysis in Table 4 shows that eliminating one or the other study does not significantly impact the resulting combined analysis: OR of 2.45 (1.03-5.85) versus 3.22 (1.57–6.61).

Second, a narrow range of values for C-reactive protein concentrations was chosen to maximize the number of studies included in the analysis. As Gutierrez et al noted, 3 studies would have been eliminated if a single cutoff value was used. Although a bright line value would have been more pleasing intellectually, we opted to include more studies to reduce the 95% confidence intervals of the result and also to provide the thoughtful reader with a complete dataset that could be re-evaluated as seen fit.

Third, a variety of outcome measures exist for expressing the results of meta-analyses.4 The odds ratio (relative odds) is optimal for generating both the Mantel-Haenszel statistic for combining the results of studies and for determining the results of homogeneity across studies;5–7 for these reasons, the odds ratio was chosen at the outset as the outcome measure for this study. We agree with Guitierrez et al, that this is not an intuitive outcome measure and its clinical applicability is open to question. The authors correctly point out that the summary of OR represents LR+/LR− and, as such, is only an approximation of LR+, the Bayesian factor used to calculate posterior probabilities. Again, we were limited by the study protocol developed before data collection that required calculation of the odds ratio. To recalculate the final results as relative risk (LR+) would have violated the protocol. It should be noted that using only 4 studies with a cutoff value of 40 mg/L, Guitierrez et al calculated a LR+ of 1.64 (1.20–2.23), which is not significantly different from the overall odds ratio reported in the study, 2.58 (1.20–5.55), and these authors reached the same conclusion regarding the use of serum C-reactive protein as a diagnostic test for bacterial pneumonia in children.

Finally, we apologize for the typographical errors and agree with Guitierrez et al that I2 is a measure of heterogeneity and was listed under the wrong heading. A cumulative forest plot was used to show when statistical difference occurred; the traditional forest plot was reiterative of data presented in the sensitivity analysis.

Robert G. Flood, MD

Department of Pediatrics

St. Louis University

St. Louis, MO

Stephen C. Aronoff, MD

Department of Pediatrics Temple University School of Medicine

Philadelphia, PA


1. Flood RG, Badik J, Aronoff SC. The utility of C-reactive protein in differentiating bacterial from non-bacterial pneumonia in children: a meta-analysis of 1230 children. PIDJ. 2008;27:95–99.
2. Korppi M, Kroger L. C-reactive protein in viral and bacterial respiratory infection in children. Scand J Infect Dis. 1992;25:207–213.
3. Heiskanen-Kosma T, Korppi M. Serum C-reactive protein cannot differentiate bacterial and viral aetiology of community-acquired pneumonia in children in primary healthcare settings. Scand J Infect Dis. 2000;32:399–402.
4. Deeks JJ, Altman DG. Effects measures for meta-analysis of trials with binary outcomes. In: Egger M, Smith GD, Altman DG, eds. Systematic Reviews in Healthcare: Meta-Analysis in Context. 2nd ed. London, England: BMJ Publishing Group; 2001:313–335.
5. Mantel N, Haenszel W. Statistical aspects of the analysis of data from retrospective studies of disease. JNCI. 1959;22:719–747.
6. Dersimionian R, Laird N. Meta-analysis in clinical trials. Controlled Clin Trials. 1986;7:177–188.
7. Laupacis A, Sackett DL, Roberts R. An assessment of clinically useful measures of the consequences of treatment. N Eng J Med. 1988;318:1728–1733.
© 2008 Lippincott Williams & Wilkins, Inc.