Ten studies (n=2,046) were included in the review. All studies were conducted in emergency departments. Fever was not defined in four studies. Withdrawals and handling of uninterpretable results were poorly reported by almost all studies. None of the studies reported sufficient information to judge whether reference standard tests were interpreted blind to the C-reactive protein result.
Seven studies (n=1,091) assessed the accuracy of C-reactive protein in differentiating between serious bacterial infection and self-limiting or non-bacterial infection; six studies reported sensitivity and specificity; and five used multivariate modelling to assess the additional information provided by C-reactive protein over clinical features. The prevalence of serious bacterial infection ranged from 11% to 29%. The summary estimates of sensitivity and specificity were 0.77 (95% CI 0.68 to 0.83) for sensitivity and 0.79 (95% CI 0.74 to 0.83) for specificity. The positive and negative likelihood ratios were 3.64 (95% CI 2.99 to 4.43) for positive and 0.29 (95% CI 0.22 to 0.40) for negative. In all five studies that used multivariate analysis, C-reactive protein was an independent predictor of serious bacterial infection.
Three studies (n=722) assessed the accuracy of C-reactive protein to differentiate between bacterial and viral infections. The reported prevalences of bacterial infection were 28%, 35% and 82%. The corresponding sensitivities were 0.44 (95% CI 0.35 to 0.54), 0.58 (95% CI 0.44 to 0.72) and 0.22 (95% CI 0.15 to 0.28). Specificities were 0.86 (95% CI 0.82 to 0.90), 0.96 (95% CI 0.91 to 1.00) and 0.94 (95% CI 0.86 to 1.00). Given the significant between-study heterogeneity, no pooled estimates were generated.