RT-PCR and Gram staining were highly accurate for S. pneumoniae, N. meningitidis, and H. influenzae meningitis diagnosis with sensitivity and specificity estimates above 90% in all analytic approaches. Although CSF culture is considered the diagnostic reference standard for bacterial meningitis, its limited sensitivity often results in an inability to confirm the diagnosis and target antimicrobial therapy. The imperfect sensitivity of culture also presents a difficulty when evaluating other tests that are potentially more sensitive. Composite reference standards and LCA modeling, which can estimate test performance parameters of multiple tests simultaneously, can be helpful in addressing this problem. We are unaware of previous published studies what have used LCA modeling to evaluate the clinical accuracy of tests for acute bacterial meningitis, an approach that requires no assumptions about the accuracy of CSF culture and furthermore, allows us to also compare the sensitivity of CSF culture with the comparison tests. The sensitivities of RT-PCR and Gram staining were over 95% and far exceeded the 81% sensitivity of culture when estimated using LCA models. LCA models stratified by results of the antibiotic detection bioassay also provide evidence that RT-PCR and Gram stain sensitivity are less affected by the presence of antibiotic activity in the CSF. This is consistent with a recent study that found the presence of antibiotic activity in CSF is a strong risk factor for a RT-PCR positive, culture-negative case . These findings suggest that RT-PCR might be particularly useful in settings where patients often receive antibiotics prior to lumbar puncture.
Nucleic acid amplification tests such as PCR do not require viable bacteria for a positive assay and are generally considered to be highly sensitive . Nevertheless, the presence of PCR inhibitors in clinical specimens can compromise sensitivity , and the presence of inhibitors might explain the imperfect RT-PCR sensitivity in this and other studies . When compared to a culture reference standard, RT-PCR specificity in this study was 90%. While contamination of RT-PCR reactions leading to false-positive results is always a concern, stratification of subjects by the presence or absence of antibiotic activity in the CSF resulted in RT-PCR specificities of 56% and 96%, respectively. This suggests that the observed RT-PCR specificity is negatively biased by the imperfect sensitivity of culture, particularly when antibiotics have been administered. RT-PCR specificities calculated using a CRS or LCA modeling were higher, suggesting that these approaches can help address the bias introduced by the insensitivity of culture. Nonetheless, it appears that RT-PCR specificity is still significantly underestimated by these alternative analytic approaches. The LCA model assigned a high probability (>98%) of a disease-free status to patients with specimens testing positive with RT-PCR only. However several of these patients had additional evidence supporting a diagnosis of bacterial meningitis (i.e. positive latex agglutination tests and elevated CSF leukocyte counts). This apparent misclassification would result in falsely low RT-PCR specificity and overestimated culture and Gram stain sensitivities. This limitation could possibly be addressed with other CRS or LCA models that incorporate additional tests that are likely less affected by antibiotic administration, such as assays for bacterial antigens (i.e. latex agglutination tests and the rapid immunochromatographic test for S. pneumoniae)  or CSF indices (i.e. leukocyte, protein, and glucose counts).
Numerous PCR assays have been evaluated for bacterial meningitis diagnosis using species-specific primers [10–12, 16, 17, 21–24] or broad range bacterial PCR [25, 26]. While PCR-based assays have played an increasingly important role in meningitis surveillance in Brazil  and elsewhere, including the United Kingdom , a routine role in clinical practice has not been established. This is likely due to the resources required and empiric treatment regimens that make confirmation less critical. Current automated RT-PCR systems have advantages over traditional PCR that make it attractive in the clinical setting, including rapid availability of results within hours and a closed system that can reduce contamination risk . Although conventional antimicrobial susceptibility testing requires that the pathogen be cultured, RT-PCR identification of bacterial pathogens might allow some narrowing of antibiotic treatment in clinical situations where cultures are negative. Accuracy of RT-PCR testing could be further improved by using newer RT-PCR primers with improved sensitivity  or performing secondary testing with other RT-PCR assays when there is an indeterminate result. However, given the potentially severe consequences of misdiagnosis, the strengths and limitations of any test should be carefully considered before recommending new diagnostic strategies.
CSF Gram staining, long considered a mainstay of bacterial meningitis diagnosis , is rapid, inexpensive, and requires relatively little training. However, accurate results are highly dependent on the operator’s staining and interpretation skills. Previous studies report Gram stain sensitivities as high as 90% and specificities of 97% or more [8, 29]. In the LCA model in this study, Gram staining had a sensitivity of 98.2% and specificity of 98.7%. As previously discussed, the likely misclassification of several patients with an isolated positive RT-PCR result as disease-free suggests that the model overestimated Gram stain and culture sensitivity, and to a lesser extent, specificity. However, as an infectious diseases referral center, Couto Maia Hospital often evaluates several patients with suspected meningitis daily, and the experience of the microbiology staff likely contributed to the excellent performance of Gram staining. Given its wide availability, rapid turnaround, low cost, and high accuracy, the value of Gram staining in the clinical workup of suspected meningitis cannot be understated.
It should be noted that only S. pneumoniae, N. meningitidis, and H. influenzae were considered in this study. The performance of Gram stain and CSF culture in diagnosing other bacterial meningitis pathogens might differ significantly. Furthermore, because there were few H. influenzae cases, due to routine Hib vaccination in infants, the test parameter estimates are likely most valid for S. pneumoniae and N. meningitidis. Also, local clinical testing guidelines may have resulted in a unique study population in this study, as CSF culture was not routinely recommended for CSF specimens not meeting specific criteria. Interestingly, patients included in our final analysis were more likely to have clinical markers of severe illness consistent with bacterial meningitis when compared to those not included due to absent RT-PCR testing. The reasons for this are unclear; however, according to surveillance staff, the most common reasons for CSF not to be tested with RT-PCR were insufficient volume, loss of the specimen when discarded by hospital staff unfamiliar with the surveillance protocol, or frequent unavailability of specimens during a prolonged aseptic meningitis outbreak in 2007, due to prioritized testing at the state public health laboratory. Since test performance can vary with different patient characteristics, evaluation of these tests in other patient populations might result in different performance parameter estimates, and the clinical characteristics of our study population should be considered when interpreting the findings of this study. However, it is important to note that test sensitivity and specificities are intrinsic parameters that would not vary with changes in disease prevalence alone.