Clinical Prediction Rules for Meningitis in Children Pretreated With Antibiotics

Abstract

To the Editor.—We read with interest the retrospective cohort analysis by Nigrovic et al1 regarding the effect of antibiotic pretreatment on cerebrospinal fluid (CSF) biomarkers of bacterial meningitis. We appreciate the research committee's commitment to investigating a very rare and difficult-to-study condition. However, we do have some questions regarding the interpretation of their data.The authors previously published a prediction rule to assist clinicians in classifying patients with CSF pleocytosis, defined as a CSF white blood cell count (WBC) of ≥10 cells per μL, as being at high or low risk of having bacterial meningitis. Patients were considered to be at low risk if they did not exhibit any of the following features: a positive CSF Gram stain, a CSF absolute neutrophil count (ANC) of ≥1000 cells per μL, CSF protein count of ≥80 mg/dL, peripheral blood ANC of ≥10 000, and a history of seizure before or at the time of presentation.2 This decision rule has been retrospectively validated in patients since the introduction of conjugate bacterial vaccines against Haemophilus influenzae type B and Streptococcus pneumoniae.3In the current study, the authors noted that the Gram-stain results for patients with antibiotic pretreatment and those without pretreatment were essentially identical (62% vs 63%; P = .86). In addition, the CSF WBC and CSF ANC of patients with antibiotic pretreatment of any time interval did not statistically differ from those patients without pretreatment. The committee stated that the study “might have been underpowered to detect true differences in CSF WBC counts or CSF ANCs according to pretreatment duration.”1 Was a power analysis performed?Unlike CSF WBC and ANC, CSF glucose did significantly differ from those patients with antibiotic pretreatment and those without. However, CSF glucose was not included in their previous clinical decision rule, because it lacked sensitivity and specificity. In today's clinical practice, CSF glucose is little more than an interesting aside in regard to the diagnosis of bacterial meningitis.Like CSF glucose, CSF protein also differed significantly in patients exposed to antibiotics within 72 hours of their lumbar puncture when compared with those patients who were not. Unlike CSF glucose, a CSF protein level of ≥80 mg/dL was included as a clinical predictor of children at high risk for bacterial meningitis in the postvaccination validation of the committee's prediction rule.3 In the study, only 8 of 121 patients with bacterial meningitis had a CSF protein level of ≥80 mg/dL as the only clinical decision predictor. Therefore, isolated elevated CSF protein level only properly classified an additional 6.6% of patients with bacterial meningitis.It is interesting to note that the CSF protein first-quartile values between the pretreatment group and no-pretreatment group differed little. The no-pretreatment group first-quartile value for CSF protein was 84 mg/dL. In the pretreatment group, the values were 107 mg/dL for patients with a time interval between antibiotics and lumbar puncture of <4 hours, 73 mg/dL for patients with a time interval between 4 and 24 hours, and 70 mg/dL for patients with a time interval of ≥24 hours. Perhaps patients with antibiotic pretreatment of >4 hours require a slightly lower cutoff for CSF protein elevation.The committee felt that the study supported their conclusion “not to apply bacterial meningitis prediction rules to pretreated patients.”1 However, we feel that the question was not clearly addressed. With the information presented in the current study, one cannot determine how many patients would have been misclassified if the decision rule was applied in its entirety to patients with CSF pleocytosis and antibiotic pretreatment. What is needed is a retrospective (or prospective) validation of their decision rule in pretreated patients. The committee probably has this information in their data set. The publication of this information would be much more clinically useful.In 2006, an independent group found the Nigrovic et al prediction rule to be the most clinically useful of all existing prediction rules.4 It would be a shame to discard this potentially valuable tool in the evaluation of pretreated children with CSF pleocytosis.