The review addressed a focused question and inclusion criteria were clearly defined. The literature search was adequate for published studies, but did not include specific attempts to identify unpublished studies. The review was restricted to English language studies; therefore, introducing the possibility of language and publication bias. Limited details of the review process were reported, so it was not possible to determine whether appropriate steps were taken to minimise bias and errors. Although study quality was assessed using appropriate criteria, the results of the assessment were not reported. Details of the individual studies were clearly reported in a summary table, which helped in determining the generalisability of the findings.
The decision to pool likelihood ratios using a simple random effects model from clearly heterogeneous data was questionable. As the data set included studies with a number of different thresholds, and the authors appeared to want to assess how performance varied with threshold, a summary receiver operating characteristic model would have been a more appropriate approach. Further investigation of the observed heterogeneity at the other two thresholds assessed would have been helpful, especially as there were clear outlying studies. The authors' conclusion that C-reactive protein marker could potentially be of value in ruling out a diagnosis of community acquired pneumonia in populations with a prevalence of more than 10%, seemed overly optimistic, given the pooled negative likelihood ratio for a 20mg/L threshold. The largest changes in pre- to post-test probability associated with a negative test result (i.e. the greatest informative value for ruling out) was in the pre-test probability (prevalence range) 40 to 80% (as stated by the authors).
Given these concerns, and the possibility that some relevant studies may have been missed, the authors' conclusions should be interpreted with some caution.