Hospital Practice: Volume 37: No.1

Benjamin R. Doolittle, MD And Amy Alias, MD

Abstract: Introduction Differentiating between aseptic and bacterial meningitis presents a difficult diagnostic challenge. Accurately ruling out bacterial meningitis may reduce unnecessary hospitalization, patient morbidity, and utilization of resources. This study applies a prediction rule previously developed in the pediatric population to an adult cohort. Methods We performed a retrospective chart review of all patients admitted to an urban community hospital in the United States between 1994 and 2007. Results One hundred eleven patients met the inclusion criteria. Twenty-two (20%) had bacterial meningitis and 89 (80%) had aseptic meningitis. The prediction rule generated a negative predictive value of 100% and a sensitivity of 100%, successfully ruling out all patients with aseptic meningitis. Conclusions: The prediction rule, previously validated in a pediatric population, was accurate in ruling out bacterial meningitis in an adult cohort. Prospective validation in an adult population is warranted.

Keywords: meningitis; prediction rule; aseptic; bacterial

Differentiating between bacterial and aseptic meningitis presents a difficult diagnostic challenge. Bacterial meningitis has an incidence of 1.5 per 100 000 person-years and a 17% mortality rate even if treated.^1,2 Among survivors, 18% have long-term sequelae such as deafness, dizziness, fatigue, and ataxia.² Aseptic meningitis is much more common with an incidence of 10 per 100 000 person-years.² Several studies document a broad overlap in cerebrospinal fluid (CSF) results between bacterial and aseptic meningitis.^3-7 Thus, conventional practice is to err on the side of caution and admit many more patients than necessary based on the concern for bacterial meningitis. An accurate prediction rule that effectively rules out bacterial meningitis would be helpful to guide appropriate therapy.

In the adult literature, there have been several attempts to develop such a prediction rule.^6,8,9 Hoen et al⁸ and Spanos et al⁹ devised a probability rule based on CSF results and patient age. They also incorporated the seasons of the year to account for variation of enterovirus infection. The negative predictive value (NPV) of Hoen et al’s model was 0.99 among 500 patients for a probability of 0.1 for bacterial meningitis.⁸ This model has been affirmed in at least 2 other studies.^10,11 However, the formula is meant to be derived in the laboratory by clinicians and is complicated to use. As such, this model has never gained widespread use.

In 2005, Brivet et al⁶ published a retrospective cohort study of 144 adult patients who presented with meningitis—90 with confirmed bacterial meningitis and 54 with aseptic meningitis. Brivet et al⁶ discovered that severity of illness—altered consciousness, seizures, focal neurological findings, and shock—along with a CSF absolute neutrophil count (ANC) > 1000 were predictive of bacterial meningitis. This study did not seek to rule out bacterial meningitis, but rather sought to identify those at highest risk. While this is important, ruling out bacterial meningitis is a slightly different clinical question since it identifies those patients who might not need hospitalization.

Another recent study among adults by Tokuda et al¹² derived a model using data from 167 patients presenting with meningitis (66 bacterial, 101 aseptic), and validating it on 28 patients (5 bacterial, 23 aseptic). Their model revealed 4 criteria as having the best discriminating variables for aseptic meningitis: gram stain, ANC < 15%, CSF ANC < 150, and mental status change. With all 4 criteria being negative, the sensitivity in the derivation group was 99% and in the validation group was 100%. While this study uses discrete, easily obtainable variables, it also relies on the subjectivity of mental status change, which opens the study to interobserver variability. The authors further acknowledge the weakness of the very small number of study subjects in the validation cohort.¹²

In the pediatric literature, there are 5 prediction algorithms to evaluate bacterial versus aseptic meningitis.^7,13-17 These studies use varying laboratory cut-offs and presenting symptoms to differentiate between bacterial and aseptic meningitis. Dubos et al¹⁸ compared each of these prediction rules to a cohort of 166 pediatric patients. Several studies had a sensitivity of 100%. Both Bonsu and Harper⁷ and Jaeger et al¹⁴ use complicated fractional polynomial equations, which impede clinical utility. Jaeger et al and Oostenbrink et al also had unacceptably low sensitivities to rule out bacterial meningitis—94% and 83%, respectively.¹⁸ In addition, Oostenbrink et al¹⁷ incorporated subjective clinical criteria into their score—vomiting, meningeal irritation, cyanosis, and petechiae. The subjectivity of these criteria opens the rule to interobserver variability. Among those studies with a sensitivity of 100%, the Nigrovic study had the highest specificity of 66%.¹⁸

In 2002, Nigrovic et al¹⁵ published a prediction rule that incorporated 5 easily identifiable features—positive gram stain, CSF protein > 80 mg/dL, peripheral ANC > 10000, CSF ANC > 1000 cells/mm³, and seizure at or before presentation. Her rule has several advantages over previously published studies. First, the 5 criteria are easy to obtain. Second, the data are discrete points without subjective interpretation. Third, the sensitivity and NPV are very effective at ruling out bacterial disease. In this initial study, her NPV was 100%.¹⁵

In 2007, Nigrovic et al¹⁶ applied the Bacterial Meningitis Score (BMS) to a prospective cohort of 3295 patients from 20 medical centers. The prevalence of bacterial meningitis was 3.7% and the NPV was 99.9%.¹⁶ Given that her BMS has been prospectively validated in a large multicenter cohort, her rule has a provocative utility in the adult population. This study applies the Nigrovic et al BMS to a retrospective cohort of adult patients to discern its accuracy in ruling out bacterial meningitis.

All patients aged ≥ 18 years admitted to St. Mary’s Hospital in Waterbury, CT between January 1994 and April 2007 were evaluated for inclusion in the study with the following International Classification of Diseases (ICD)-9 codes for meningitis: bacterial meningitis (320.0–320.9), meningitis due to organisms other than bacteria (321.0–321.8), and unspecified meningitis (322.0–322.9).

Patients were considered to have bacterial meningitis if their CSF was positive for gram stain or if their CSF culture was positive. The diagnosis of aseptic meningitis was considered if viral cultures were positive or no other etiology was found.

In addition to demographic criteria, information pertinent to the BMS was collected: peripheral white blood cells (WBC) and ANC; CSF WBC and ANC; CSF glucose; protein; and gram stain. Additionally, concomitant diagnoses of cancer, HIV, or other comorbities were noted but not excluded.

One hundred eleven patients met the inclusion criteria. Twenty-two (20%) were diagnosed with bacterial meningitis based on CSF or blood culture results. Eighty-nine (80%) were diagnosed with aseptic meningitis. Table 1 details patient characteristics as well as values for gram stain, CSF ANC > 1000 cell/mm, CSF protein > 80 mg/dL, and peripheral WBC ANC > 10000 cells/mm.

Of the 22 patients with bacterial meningitis, the breakdown was as follows: 15 (68%) Streptococcus pneumonia, 2 (9%) Staphylococcus aureus, 1 (5%) Neisseria meningitis, 1 (5%) Hemophilus influenza, and 3 (14%) other Streptococcus species.

A BMS was generated for each patient using positive gram stain, CSF protein > 80 mg/dL, peripheral ANC > 10000, and CSF ANC > 1000. Two points are given to positive gram stain and 1 point for each of the other predictors. A BMS score of 0 accurately identified only those patients with aseptic meningitis (Figure 1). The sensitivity of this rule in our patient cohort was 100% with an NPV of 100%, a PPV of 33%, and a specificity of 49%. Figure 2 details the recursive partitioning of the patient cohort.

This study demonstrates that Nigrovic et al’s BMS was effective to rule out bacterial meningitis in a retrospective cohort of adult patients with a sensitivity and negative predictive value of 100%. Given its ease of use, application of this rule to the adult population is compelling.

The pediatric population is different from adults in several respects. First, pediatric patients are usually healthy at baseline and present with a new-onset illness. Adults often contend with comorbidities, such as cancer and HIV, which predispose them to severe illness. Second, pediatric patients in the United States are very often vaccinated. In particular, the widespread use of conjugate vaccine against Streptococcus pneumonia has dramatically decreased pneumococcal meningitis.¹⁹ Black et al¹⁹ revealed the efficacy of the 7-valent pneumococcal conjugate vaccine to be 100% against invasive pneumococcal disease for fully vaccinated children and 89.1% efficacious for nonvac-cine serotypes. Similarly, Bisgard et al²⁰ demonstrate that invasive disease from Haemophilus influenzae disease has been nearly eliminated among fully vaccinated children. Data from the Centers for Disease Control and Prevention show a decrease from 11 000 cases in 1987—prior to the widespread implementation of the vaccine—to fewer than 300 cases in 2000. In Nigrovic et al’s¹⁶ JAMA cohort, only 3.7% of patients had bacterial meningitis, reflecting the widespread use of vaccines. In this adult cohort, 22% had bacterial meningitis. Despite this much higher prevalence, Nigrovic et al’s prediction rule was accurate.

An important strength to the bacterial meningitis score is that no subjective criteria were included. Many studies include mental status change and meningismus, which may have a subjective interpretation between providers.⁶ This study affirms the work of Tokuda et al¹² without incorporating the subjective criteria of mental status change. While the Nigrovic et al model does incorporate seizure among her JAMA cohort of 3295 patients, only 2 had seizures. In our cohort, only 1 presented with a seizure and was appropriately characterized with the BMS as having bacterial meningitis.

In this cohort, no patients were excluded because of comorbidities, reflecting a “real-world” cohort of patients presenting to the hospital where concomitant illness, such as cancer and HIV, may not be known. Among the 111 patients, 4 had HIV. Among these 4 patients, 1 had bacterial meningitis and was appropriately characterized. Four patients had cancer, 3 of whom had bacterial meningitis. All 3 were correctly partitioned. Despite this accuracy, given the higher prevalence for bacterial infection among those with significant comorbidity, a higher clinical suspicion for bacterial meningitis is warranted.

Important weaknesses of this study are its retrospective nature and lack of a prospective validation. Because the stakes are high in accurately ruling out bacterial meningitis, prospective validation is important before widespread use can be employed. It is also important to note that “aseptic” meningitis does not mean “benign” meningitis. Nor does it mean that a patient with aseptic meningitis need not be admitted to the hospital. In our opinion, the BMS is not a substitution for clinical judgment, but rather a complement. The use of such a prediction tool is consistent with the Infectious Diseases Society of America guidelines, which affirm the importance of early recognition and maintain that prompt treatment of bacterial meningitis can lead to improved outcome.²¹ These results are provocative. Further evaluation among a larger cohort of adult patients, in a prospective manner, is indicated.

Hello!


Back to the table of contents for the December 2009 issue