Reassessing the Value of CSF Protein and Glucose Measurement in Pediatric Infectious Meningitis Free

The study was approved by the university’s institutional review board. All patients aged 0 to 18 years old with CSF studies performed at our institution between October 2015 and August 2017 were identified from an internal laboratory database. We included all CSF studies performed to evaluate for infectious meningitis or in the context of infectious work-up for unclear primary site of infection (such as a febrile infant) as determined by chart review of the chief complaint; patients with CSF studies clearly performed for noninfectious concerns were excluded, as were patients with incomplete CSF (missing WBC, protein, or glucose) testing data (Fig 1). For patients who had multiple lumbar punctures during a single illness course, only the initial CSF studies were included in the analysis.

Data collected for all patients included age, clinical characteristics, CSF studies (WBC, RBC, protein, glucose, Gram stain, culture, and PCR testing if performed), and final diagnosis. The availability of different PCR assays in our institution changed during the study period: assays for enterovirus and parechovirus (internal microbiology laboratory) and 16S ribosomal RNA sequencing (University of Washington, Seattle, WA) were available throughout the study period and a multiplex PCR assay targeting 14 viral, bacterial, and fungal pathogens (FilmArray Meningitis and Encephalitis Panel [MEP], BioFire, Salt Lake City, UT) was available from January 2017 to August 2017. We used our internal laboratory reference norms for CSF WBC, RBC, protein, and glucose. (Supplemental Table 4). Abnormal protein was defined as values above the reference range (>80 mg/dL), and abnormal glucose was defined as values below the normal range (<20 mg/dL). CSF pleocytosis was defined as CSF WBC count >5/cm³ in all age groups and traumatic or hemorrhagic LP was defined as CSF RBC >500/cm³.

Patients were categorized in to 3 final diagnostic categories. The “No IM” category was defined by any final diagnosis in the chart other than meningitis. “Microbiologically proven IM” required positive culture, MEP, or other molecular test. “Clinical diagnosis of IM” was defined as a discharge diagnosis of meningitis with negative microbiological testing.

Chart review was performed on all patients with microbiologically confirmed or clinical IM to identify and characterize instances in which CSF protein or glucose were explicitly documented in clinical notes as included in medical decision-making.

Patient demographic, test results and clinical diagnoses were described. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for CSF WBC, protein, and glucose were calculated for microbiologically confirmed IM. These values were calculated for the entire study population and stratified by age ≤2 or >2 months (since this age represents a threshold below which lumbar puncture is often performed on well-appearing infants with fever). Test characteristics were not calculated for the MEP as a positive result on the MEP was part of the case definition. The analysis was performed including (Table 1) and excluding (Table 2) patients with traumatic or hemorrhagic LPs as well as on patients with bacterial pathogen identified (Table 3). For our sensitivity and specificity calculations, we considered only patients with culture- or molecular assay-proven meningitis to truly have disease. The sensitivity, specificity, PPV, and NPV for the WBC, protein, and glucose were compared with combinations of these tests in parallel. We used McNemar’s test to compare diagnostic test performance in 2 × 2 tables between WBC alone and WBC plus protein or WBC plus glucose. Among patients with WBC > 5 we used Wilcoxon tests to compare rates of elevated protein and positive MEP between patients with bacterial IM, viral IM, and no pathogen identified. All analyses were performed using SAS version 9.4, Cary, NC.

Of 887 patients with CSF testing performed, we identified 735 patients with CSF testing for IM over the 22-month period; 446 patients were age <2 months. The number of patients screened, excluded, and included, as well as distribution of CSF results and final classification are shown in Fig 1. A total of 69 patients were diagnosed with IM: 45 (6.1%) microbiologically confirmed and 24 (3.3%) with clinical diagnosis (Fig 2). Twenty-three microbiologically confirmed patients were <2 months (5.2% of total patients <2 months). Twelve of the microbiologically confirmed cases were with a bacterial pathogen. Bacterial pathogens identified were Escherichia coli (2 cases), Haemophilus influenzae (3 cases), Streptococcus pneumoniae (2 cases), Group B Streptococcus (2 cases), Streptococcus anginosus, Streptococcus viridans and Sneathia sanguinegens. Viral pathogens included 30 cases of enterovirus, 2 human parechovirus, and 1 human herpes virus-6. There were no cases of HSV meningitis among the study population. All patients diagnosed with clinical IM were >2 months old. One patient with a ventriculoperitoneal shunt was identified, but the sample of CSF was obtained by lumbar puncture.

There were 5 microbiologically confirmed cases in which a positive MEP was the only CSF abnormality (WBC, protein and glucose normal), all either human parechovirus or enterovirus. Three microbiologically confirmed cases had organisms not included in the MEP (Streptococcus anginosus, Streptococcus viridans and Sneathia sanguinegens); all of these had abnormal WBC. Thus, 100% of microbiologically confirmed cases had either positive MEP, CSF WBC >5, or both. CSF WBC was abnormal in all patients with clinical meningitis cases and, based on medical record documentation, was the major driver of clinicians making a clinical diagnosis in the absence of microbiologic proof. CSF WBC >5 had sensitivity of 88.9%, specificity of 63.5%, PPV of 13.7% and NPV of 98.9% for microbiologically proven meningitis (Table 1) and 100% sensitivity for bacterial pathogens (Table 3).

There were 195 elevated CSF protein values, 44 low values, and 496 normal for age. For microbiologically proven meningitis, the sensitivity (44.4%), PPV (10.3%), and NPV (95.4%) were all notably lower than values for WBC; specificity (74.6%) was slightly higher than for WBC. In parallel testing with abnormal WBC or protein combined, there was no change in test sensitivity and small decreases in specificity (−9.4%), PPV (−2.5%), and NPV (−0.2%) compared with CSF WBC alone (Table 1). For infants <2 months these values were (−13.7%) for specificity, (−2.1%) for PPV and (−0.6%) for NPV.

Traumatic or hemorrhagic LP (RBC > 500) was common in the study cohort, occurring in 170 of 735 patients (23%). Abnormal CSF protein was approximately 3 times more likely in patients with traumatic or hemorrhagic tap (53.5% vs. 18.4%, P < .0001 χ² test). However, exclusion of patients with traumatic or hemorrhagic tap did not alter the observations of unchanged sensitivity and decreased specificity, PPV, and NPV when protein plus WBC were compared with WBC alone (Table 2).

Abnormally low glucose occurred in only 8 of 735 patients, all with a value of 19 mg/dL. Data on concurrent serum glucose measurements were not available. Low glucose had sensitivity of 13.3%, specificity of 99.7%, PPV of 75.0%, and NPV of 94.6% for microbiologically confirmed meningitis. In parallel testing, abnormal CSF glucose and WBC combined had zero impact on test characteristics compared with using CSF WBC alone across all cases and age subgroups. Because our institutional normal value for CSF glucose may be lower than some published in the literature, we repeated the analysis with lower limits of 30 and 40 mg/dL.⁷  These changes predictably increased the sensitivity and slightly decreased the specificity of CSF glucose alone, but resulted in no meaningful change in the sensitivity, specificity, PPV, or NPV of glucose plus WBC when compared with WBC alone.

CSF WBC and protein were both abnormal in all cases of bacterial IM (WBC 82–8790/cm³, protein 116–1801 mg/dL), and CSF glucose was abnormal in 5 of 12 cases (all 19 mg/dL). As all cases of bacterial meningitis had elevated WBC count, CSF protein and glucose measurements did not add to the sensitivity of testing in these patients. When MEP was performed, results were positive in all patients with IM due to organisms contained in the assay. Five patients were identified by CSF culture alone, all prior to MEP becoming available at our institution: 2 with Streptococcus pneumoniae, and 1 each with Streptococcus anginosus, Streptococcus viridans and Sneathia sanguinegens (the latter 3 etiologies are not contained on the MEP). Gram stain was positive in 7 of 12 (58%) of bacterial IM cases.

Among patients with WBC >5, the median CSF protein was higher among patients with microbiologically proven bacterial (286 mg/dL) than viral (59 mg/dL) (P < .0001, Wilcoxon test), and 12 of 12 (100%) of bacterial IM had protein >80 compared with 9 of 29 (32%) (P value .007) with viral IM and 135 of 292 (46%) with no etiology. In contrast, among patients with WBC >5 and who had an MEP sent, the MEP was positive in 6 of 6 (100%) of proven bacterial IM, 8 of 8 (100%) proven viral IM (and 0 of 80 [0%] with no microbiologically proven etiology—as expected based on the case definition).

Among the 24 patients diagnosed clinically with IM, all had elevated WBC count, 15 (63%) had elevated protein, and 1 (7%) had low glucose. Abnormal protein was documented in MDM in 6 patients (0.81% of the total patient population), 3 of whom were <2 months. In all 6 both WBC and protein together were cited as reasons to continue antibiotics despite negative PCR or culture. All 6 patients had been pretreated with antibiotics before CSF collection. In no case was a clinical decision-making tool incorporating protein documented and in no case was the glucose noted in clinical decision-making.

The 3 patients noted above with IM caused by bacterial organisms not detected on the MEP had elevated protein and 2 had low glucose in addition to all 3 having elevated WBC; 2 had positive Gram stain. Clinicians in these cases cited both elevated CSF WBC count and protein as contributing to the decision to continue antibiotics despite negative initial microbiologic testing. CSF glucose was not mentioned.

In a large cohort of children undergoing testing for infectious meningitis, abnormal WBC or MEP identified all patients with a final diagnosis of either microbiologically proven or clinical IM; in no patient was an abnormal protein or glucose the only indicator of IM. CSF protein and glucose had very low PPV for prediction of meningitis in isolation, and when combined with WBC have at best no impact on the prediction of meningitis, and in the case of CSF protein, actually reduced the specificity, positive predictive value, and negative predictive value. CSF protein was cited in less than 1% of cases as contributing to MDM and never cited in isolation as the sole reason for a clinical choice. Glucose was never mentioned in MDM. In the current era in which meningitis (particularly severe forms including bacterial and herpes simplex virus) is rare and multiplex PCR testing is increasingly available, our observations suggest that clinicians should be intentional about selecting these tests in the evaluation of patients with possible IM, rather than routinely ordering them on all CSF samples.

Measurement of CSF protein and glucose for evaluation of infectious meningitis represents both the technology and the epidemiology of a past era. Examination of CSF was first performed in 1885, and chemical composition of CSF, including standard values for CSF protein and glucose, was described in 1912.^8,9  The first CSF profiles for infectious meningitis were described in 1937⁸ . Significant declines in pediatric IM occurred after the introduction of the Hib vaccine in 1990, which reduced incidence of bacterial meningitis by 55% alone,¹  the pneumococcal conjugate vaccines (PCV7 in 2000 and PCV13 in 2010), which reduced invasive pneumococcal disease in children by more than 90% (CDC 2020),²  a 90% reduction in meningococcal disease between 1995 and 2019, partly attributable to introduction of the meningococcal vaccine in 2005,^3,10  and the introduction of intrapartum Group B strep (GBS) prophylaxis in 1996, which is estimated to have reduced the incidence of early onset invasive GBS disease by 87% between 1990 and 2015.^4,11  With declining incidence by the early 2000s, the positive predictive value of CSF white blood cells, protein, and glucose for bacterial causes have also declined.

An even more significant impact on the utility of these tests has resulted from the introduction of highly sensitive and specific multiplex PCR assays such as the MEP. These tests have an approximately 1-hour assay time, and turnaround time from sample collection to report back to clinicians in published studies ranges from 2.2 to 6.2 hours.^5,12,13  Prior studies have shown MEP has comparable diagnostic accuracy and more rapid time-to-diagnosis compared with standard culture and molecular reference methods.^5,6  Though the MEP has higher upfront cost than standard of care tests, studies have shown a cost-justification for MEP both at community and tertiary care hospitals with benefits including reduced empirical antibiotic use and shorter hospital length of stay, that offset higher initial costs of testing.^14–17 

Our data suggest that routine or universal use of CSF protein and glucose is low value care and that clinicians, particularly those with access to multiplex PCR, should be intentional and consider if results will impact clinical decision-making before ordering them. This consideration should incorporate the clinical context as well as the availability and turn-around-time of multiplex assays.

Lumbar punctures are frequently performed on patients with overall low suspicion of meningitis, perhaps most commonly among well-appearing febrile infants. In these cases, clinicians are typically testing CSF to determine if there is evidence of IM or not. In such infants, decisions to initiate antibiotic coverage may already be predetermined based on age or blood biomarker results, and in lower-risk infants there is frequently time for decision-making to be made iteratively over a period of several hours; a timeframe that would allow incorporation of multiplex PCR testing. In neither of these scenarios is measurement of protein or glucose likely to alter clinical decisions.

In children with higher suspicion of meningitis, either because of systemic illness or meningeal signs, more rapid decision-making is required, and in addition to the question of whether ME is present or not, clinicians often require rapid information to help discriminate benign (mostly viral) causes of IM from more serious causes (bacterial or HSV). In this context, clinicians may still prefer to order CSF protein at the time of initial testing if rapid multiplex PCR assays are unavailable, or after initial result review if high clinical suspicion remains despite negative molecular testing. However, if protein is measured, clinicians should consider deliberate application of an established clinical decision tool that incorporates CSF protein. Nigrovic et al developed a Bacterial Meningitis Score in which a CSF protein <80 was 1 of 5 clinical and laboratory parameters (Gram stain, peripheral absolute neutrophil count ≥10 000 cells/mm³, seizure before or at time of presentation, and CSF absolute neutrophil count ≥1000 cells/mm³) which if all negative, predicted a very low risk of bacterial meningitis.^18,19  More recently, Mintegi et al developed the Meningitis Score for Emergencies to differentiate bacterial from aseptic meningitis in which CSF protein >80 was more likely to predict bacterial over aseptic meningitis among patients with pleocytosis.²⁰  Our data do support the potential for elevated protein to discriminate viral from bacterial causes in patients with pleocytosis: protein was abnormal in 100% of bacterial cases, statistically higher than in viral (32%) or aseptic cases (46%). However, in no case in our study was the use of such a tool documented and addition of multiplex PCR in the group of patients with elevated WBC dramatically lowers the level of uncertainty between viral and bacterial etiologies that such tools were originally intended to help navigate. Our observations are consistent with both meningitis tools in that measurement of CSF glucose did not contribute any predictive benefit and was excluded. Older studies characterizing potential utility of CSF glucose in predicting bacterial ME used a simultaneous blood glucose to determine abnormal values, a practice which was employed neither in our cohort nor in the large retrospective datasets that were used by Nigrovic or Mintegi.

Finally, there remain circumstances in which CSF protein (and to a lesser extent, glucose) may still be of diagnostic utility, such as when autoimmune etiologies and atypical meningitis or encephalitis pathogens such as tuberculosis are part of the differential diagnosis. Clinicians should be aware of when it would be appropriate for CSF protein and glucose to be added to the sample in the laboratory when a specific need arises, rather than ordering routinely on every sample.²¹ 

In a 2016 study of 96 000 infants in northern California, the incidence rate of fever in infants <90 days was 1.4%, and of 1380 febrile infants identified over 18 months, approximately one third had a CSF culture.²²  Extrapolating this data to the 3.7 million annual births suggests that as many as 16 000 lumbar punctures in infants may occur annually in the United States, and thus measurement of protein and glucose, though not extremely costly (in our institution laboratory cost [not charge] for protein is $163 and glucose is $40), may represent significant resource expenditure. Furthermore, given the poor specificity of CSF protein, false-positive values may contribute to inappropriate medical decision-making and excessive subsequent testing or treatment.

Perhaps more compelling is that routine ordering of protein and glucose require additional volumes of CSF to be collected, which may have direct impact on patient care. Many clinicians will be familiar with the scenario in which it is difficult or slow to obtain CSF and reducing the volume of CSF required might help reduce repeat LP attempts or shorten difficult or sedation-requiring procedures. At our institution, the MEP requires 200 µL of CSF, whereas CSF protein and glucose combined require 500 µL of CSF.

Limitations to our study include retrospective data analysis, which can result in loss of information and inability to assess clinical decision-making beyond what is documented. We selected a study population undergoing work up for infectious concerns only so these observations should not be extrapolated to clinical scenarios in which autoimmune, metabolic, or other neurologic etiologies are under consideration. Our study data come from a single institution and incorporation of protein and glucose in provider decision-making may differ in other centers. Finally, the number of cases of microbiologically confirmed and especially bacterial IM was relatively low, limiting more precise analysis of these patients, though the low incidence is also a key underlying premise of the study.

In conclusion, our results demonstrate that routine measurement of CSF protein and glucose provide low and diminishing value in the diagnosis of pediatric infectious meningitis in the era of reduced incidence of bacterial meningitis and rapid molecular testing to identify causative pathogens.