Clinical Prediction Rule for Identifying Children with Cerebrospinal Fluid Pleocytosis at Very Low Risk of Bacterial Meningitis

ORIGINAL CONTRIBUTION

Clinical Prediction Rule for Identifying Children With Cerebrospinal Fluid Pleocytosis at Very Low Risk of Bacterial Meningitis

Lise E. Nigrovic, MD, MPH Context Children with cerebrospinal fluid (CSF) pleocytosis are routinely admitted Nathan Kuppermann, MD, MPH to the hospital and treated with parenteral antibiotics, although few have bacterial Charles G. Macias, MD, MPH meningitis. We previously developed a clinical prediction rule, the Bacterial Meningi- tis Score, that classifies patients at very low risk of bacterial meningitis if they lack all Christopher R. Cannavino, MD of the following criteria: positive CSF Gram stain, CSF absolute neutrophil count (ANC) Donna M. Moro-Sutherland, MD of at least 1000 cells/µL, CSF protein of at least 80 mg/dL, peripheral blood ANC of at least 10 000 cells/µL, and a history of seizure before or at the time of presentation. Robert D. Schremmer, MD Objective To validate the Bacterial Meningitis Score in the era of widespread pneu- Sandra H. Schwab, MD mococcal conjugate vaccination. Dewesh Agrawal, MD Design, Setting, and Patients Amulticenter,retrospectivecohortstudycon- Karim M. Mansour, MD ducted in emergency departments of 20 US academic medical centers through the Jonathan E. Bennett, MD Pediatric Emergency Medicine Collaborative Research Committee of the American Acad- emy of Pediatrics. All children aged 29 days to 19 years who presented at participat- Yiannis L. Katsogridakis, MD, MPH ing emergency departments between January 1, 2001, and June 30, 2004, with CSF Michael M. Mohseni, MD pleocytosis (CSF white blood cells Ն10 cells/µL) and who had not received antibiotic treatment before lumbar puncture. Blake Bulloch, MD Main Outcome Measure The sensitivity and negative predictive value of the Bac- Dale W. Steele, MD terial Meningitis Score. Ron L. Kaplan, MD Results Among 3295 patients with CSF pleocytosis, 121 (3.7%; 95% confidence Martin I. Herman, MD interval [CI], 3.1%-4.4%) had bacterial meningitis and 3174 (96.3%; 95% CI, 95.5%- 96.9%) had aseptic meningitis. Of the 1714 patients categorized as very low risk for Subhankar Bandyopadhyay, MD bacterial meningitis by the Bacterial Meningitis Score, only 2 had bacterial meningitis Peter Dayan, MD, MSc (sensitivity, 98.3%; 95% CI, 94.2%-99.8%; negative predictive value, 99.9%; 95% Uyen T. Truong, MD CI, 99.6%-100%), and both were younger than 2 months old. A total of 2518 patients (80%) with aseptic meningitis were hospitalized. Vincent J. Wang, MD Conclusions This large multicenter study validates the Bacterial Meningitis Score pre- Bema K. Bonsu, MD diction rule in the era of conjugate pneumococcal vaccine as an accurate decision sup- Jennifer L. Chapman, MD port tool. The risk of bacterial meningitis is very low (0.1%) in patients with none of the criteria. The Bacterial Meningitis Score may be helpful to guide clinical decision John T. Kanegaye, MD making for the management of children presenting to emergency departments with Richard Malley, MD CSF pleocytosis. for the Pediatric Emergency Medicine JAMA. 2007;297:52-60 www.jama.com Collaborative Research Committee of the American Academy of Pediatrics because exclusion of bacterial menin- pneumoniae,8-12 there has been a sig- gitis requires negative CSF (and nificant decrease in the incidence of LTHOUGH BACTERIAL MENIN- blood) cultures after 2 to 3 days of bacterial meningitis in US children. gitis is the greatest concern incubation,4,5 most children with This has further reduced the probabil- when evaluating and treating CSF pleocytosis are admitted to the ity that a child with CSF pleocytosis children with cerebrospinal hospital to receive broad-spectrum will have bacterial meningitis. A fluidA (CSF) pleocytosis,1 the majority antibiotics while awaiting culture test of these children have aseptic rather results. With the widespread introduc- Author Affiliations are listed at the end of this 2,3 article. than bacterial meningitis. However, tion of highly effective bacterial conju- Corresponding Author: Lise E. Nigrovic, MD, MPH, gate vaccines against Haemophilus Department of Medicine, Children’s Hospital Bos- See also Patient Page. ton, 300 Longwood Ave, Boston, MA 02115 (lise influenzae type b6,7 and Streptococcus [email protected]).

Downloaded From: http://jama.jamanetwork.com/ by a Cincinnati Children's Hospital User on 09/06/2016 VALIDATION OF A CLINICAL PREDICTION RULE FOR BACTERIAL MENINGITIS

highly accurate decision support tool and may differ if tested in a new that could identify which children patient population (model overfitting Box. Components of the with CSF pleocytosis had a near-zero to the original data set).16 Therefore, Bacterial Meningitis Score* risk of bacterial meningitis by using before implementation of a clinical Variables in the Bacterial clinical and laboratory parameters prediction rule, the model should be Meningitis Score readily available at the time of clinical validated externally using a different Positive cerebrospinal fluid Gram presentation could guide decision patient population and clinical setting stain making and limit unnecessary hospital from those on which the prediction admissions and prolonged antibiotic rule was developed.13,15-19 In practice, Cerebrospinal fluid absolute neutrophil count Ն1000 cells/µL use. these methodological standards are We previously developed a clinical seldom met.18 Cerebrospinal fluid protein Ն prediction rule, the Bacterial Meningi- We desired to validate the Bacterial 80 mg/dL tis Score,1 which classifies patients at Meningitis Score in the era of wide- Peripheral blood absolute neutro- very low risk of bacterial meningitis if spread conjugate pneumococcal phil count Ն10 000 cells/µL they lack all of the following criteria: vaccination on a large population of History of seizure before or at the positive CSF Gram stain, CSF abso- children evaluated in emergency de- time of presentation lute neutrophil count (ANC) of at least partments across the United States. To *Patients are classified as very low risk 1000 cells/µL, CSF protein of at least this end, we performed a validation if none of these variables are present. 80 mg/dL, peripheral blood ANC of at study by using a network of 20 aca- least 10 000 cells/µL, and a history of demic medical centers, as part of the Pe- seizure before or at the time of presen- diatric Emergency Medicine Collabo- tation (BOX). In the original study of rative Research Committee of the general emergency departments (n=3). 696 children hospitalized with CSF American Academy of Pediatrics. We Approval for the study and for data pleocytosis at a single institution, we planned to review the records of all chil- sharing with the coordinating institu- derived the Bacterial Meningitis Score dren with meningitis evaluated in the tion was granted by the institutional re- on a random two thirds of the chil- emergency departments of these cen- view boards at each participating insti- dren in the data set, and validated the ters over a 4-year period. Thus, our tution. Requirement for informed score on the remaining one third of the study goal was to externally validate the consent was waived by the institu- children.1,13 We found that of the 144 Bacterial Meningitis Score, focusing pri- tional review boards of each participat- patients classified as very low risk in the marily on the ability of the rule to iden- ing institution. validation set none had bacterial men- tify patients at very low risk of bacte- ingitis (negative predictive value, 100%; rial meningitis. We also determined Patient Identification 95% confidence interval [CI], 97%- whether further refinement of the Bac- We reviewed the medical records of all 100%). In the validation set, the sen- terial Meningitis Score would simplify patients aged 29 days to 19 years who sitivity of the Bacterial Meningitis Score and improve the performance of this received a diagnosis of meningitis (In- for bacterial meningitis (ie, having clinical prediction rule. ternational Classification of Diseases, Ն1predictionruleriskfactor)was Ninth Revision, Clinical Modification 100% (37/37; 95% CI, 91%-100%). METHODS [ICD-9-CM], with the following codes: For several reasons, clinical predic- Multicenter Collaborative bacterial meningitis, 320.0-320.9; vi- tion rules are often less accurate when Research Network ral meningitis, 046.0-048.9; or unspeci- tested in a new clinical setting.14 First, The Pediatric Emergency Medicine Col- fied meningitis, 321.0-322.9) at each of the assessment of either the predictor laborative Research Committee of the the participating emergency depart- or outcome variables may not be American Academy of Pediatrics re- ments between January 1, 2001, and reproducible with new patients and viewed, critiqued, and approved the June 30, 2004. For consistency, only pa- physicians.15 Second, as in the case of study protocol. We identified coinves- tients who had lumbar punctures per- bacterial meningitis since the advent tigators from 20 emergency depart- formed in the emergency department of conjugate polysaccharide vaccines, ments located across the United States, of the study institutions were in- the epidemiology of the disease or which routinely participate in this re- cluded (n=4369). We ensured com- associated diagnostic testing studied search network. Collectively, more than plete capture of children with bacte- may change over time and thus poten- 1 million children per year are evalu- rial meningitis by cross-checking each tially affect the performance of any ated in emergency departments in these institution’s microbiology logs and by prediction rule. Finally, the observed centers. Participating institutions were including all patients with CSF cul- relationships between predictors and located in 16 different states, plus the tures growing bacterial pathogens (5 pa- outcome may depend on unique char- District of Columbia, and included free- tients with bacterial meningitis were not acteristics of the derivation population standing pediatric centers (n=17) and identified by ICD-9 codes). Cultures

Downloaded From: http://jama.jamanetwork.com/ by a Cincinnati Children's Hospital User on 09/06/2016 VALIDATION OF A CLINICAL PREDICTION RULE FOR BACTERIAL MENINGITIS

that grew Staphylococcus epidermidis, pleocytosis with negative bacterial cul- stain, blood and CSF cultures), and other Streptococcus viridans,orProprionobac- tures of blood and CSF and a negative microbiology testing that was per- terium acnes were considered to repre- CSF latex agglutination test (if formed (herpes simplex virus, enterovi- sent contamination and were classi- obtained). ral or Lyme CSF polymerase chain re- fied as being negative. Patients who did not have a CSF cul- action latex agglutination testing, Borrelia ture obtained were excluded (n=15); burgdorferi serology, and viral and/or my- Inclusion Criteria however, we did include patients who cobacterial culture). We also deter- Children were classified as having men- did not have blood cultures obtained mined clinical outcome, final clinical di- ingitis and included in the study if ei- provided that a CSF culture was ob- agnosis, length of hospital stay, and ther of the following criteria applied: tained (n=342). Three of these pa- duration of parenteral antibiotics by CSF pleocytosis (CSF white blood cells tients had bacterial meningitis based on medical record review. In case of dis- Ն10 cells/µL, corrected for the pres- apositiveCSFculture.Theremaining crepancies between clinicians in the ence of CSF red blood cells using a 339 patients, all of whom had negative medical record, only the attending phy- 1:500 ratio of leukocytes to erythro- CSF cultures, had either Lyme menin- sician’s documentation was consid- cytes usually found in peripheral gitis (7 [2% of patients with aseptic men- ered. When more than 1 CSF cell count blood20,21)orapositiveCSFculturefor ingitis and no blood culture]), enterovi- was performed, the tube with the few- a bacterial pathogen. ral meningitis (79 [23%]), or unspecified est red blood cells was always used re- aseptic meningitis (253 [75%]). Ex- gardless of its order in the sequence of Exclusion Criteria cept for the patients with Lyme menin- collection. We excluded patients with CSF pleo- gitis (who each received parenteral an- cytosis who would require hospital ad- tibiotics for 21 days), none of these Bacterial Meningitis mission regardless of the risk of bacte- patients received a course of antibiot- Score Validation rial meningitis, including children with ics for bacterial meningitis (defined by In the main analysis, we evaluated the any of the following conditions or clini- acourseofantibioticsՆ7days). performance of the Bacterial Meningitis cal findings: critical illness (defined as Score for predicting patients at very low severely altered mental status, evi- Data Collection risk of bacterial meningitis. Patients pre- dence of cerebral herniation, need for Each of the coinvestigators reviewed the senting with any predictors in the Bac- respiratory or blood pressure sup- computerized medical records, written terial Meningitis Score prediction rule port), purpura, presence of ventricu- medical records, or both for all study pa- were considered not to be at very low risk lar shunt device, recent neurosurgery, tients at their site. Patient information of bacterial meningitis. Patients miss- immunosuppression, other bacterial in- was entered by each investigator either ing data for any of the predictors were fections necessitating inpatient antibi- onto a structured case report form (7 cen- excluded from this analysis unless they otic therapy (eg, urinary tract infec- ters) or directly into a computerized da- had 1 or more positive predictors among tions in infants Ͻ3months,periorbital tabase (identical in structure to the case those that could be evaluated, in which cellulitis, deep abscess, bone or joint in- report form) by using Microsoft Access case they were considered not to be at fections, or known bacteremia), or ac- database software24 (13 centers). very low risk for bacterial meningitis. We tive Lyme disease. Because antibiotic Because we were also interested in re- evaluated the performance of the rule pretreatment can alter CSF pro- fining the Bacterial Meningitis Score pre- with respect to sensitivity, specificity, files22,23 and result in falsely negative diction rule in addition to validating the negative predictive value, and positive blood cultures, CSF cultures, or both, Bacterial Meningitis Score, we collected and negative likelihood ratios for bacte- we excluded patients who had re- and recorded the following informa- rial meningitis, and calculated 95% CIs ceived oral or parenteral antibiotics tion: patient demographics (date of birth, where appropriate. within 72 hours before their lumbar date of presentation, sex), clinical data puncture. (coexisting medical conditions, antibi- Bacterial Meningitis otic pretreatment, vaccination status, tri- Score Refinement Case Definitions age temperature and duration of fever at To attempt to refine the Bacterial Men- We defined a child as having bacterial the time of presentation, occurrence and ingitis Score, we performed in a sub- meningitis if there was a positive CSF timing of seizures), physical examina- sequent analysis binary recursive par- culture, CSF pleocytosis in association findings (presence of rash, titioning using a classification tree tion with a positive blood culture for a meningeal signs, and papilledema), labo- algorithm. To approximate clinical de- bacterial pathogen, or CSF pleocyto- ratory test results (peripheral complete cision making, we assigned in the analysis in association with a positive CSF blood cell count and differential, periph- sis a relative cost of 100 for misclassi- latex agglutination test for a bacterial eral glucose, CSF white blood cell count fying a patient with bacterial meningitis pathogen. We defined a child as hav- and differential, CSF red blood cell count, as having aseptic meningitis. The re- ing aseptic meningitis if there was CSF CSF glucose, CSF protein, CSF Gram cursive partitioning algorithm gener-

Downloaded From: http://jama.jamanetwork.com/ by a Cincinnati Children's Hospital User on 09/06/2016 VALIDATION OF A CLINICAL PREDICTION RULE FOR BACTERIAL MENINGITIS

ated a decision tree classifying pa- cluded 515 patients who would have re- in TABLE 1. All patients (n=121) with tients by risk of the outcome of interest quired admission regardless of their risk bacterial meningitis and 2518 patients using standard Gini splitting rules.25 We of bacterial meningitis, 544 patients (80%) with aseptic meningitis were ad- then pruned the resulting tree for sim- who had received antibiotic treatment mitted to the hospital. The median plicity and plausibility. In the recur- before their lumbar puncture, and 15 length of parenteral antibiotic therapy sive partitioning analysis, we in- patients who had no CSF culture ob- was 14 days (interquartile range, 10-14 cluded 13 candidate predictors with tained. Patients could be excluded for days) for patients with bacterial men- biological plausibility and minimal more than 1 reason. Among 3295 re- ingitis and 2 days (interquartile range, missing data (Ͻ10%), including all of maining patients with CSF pleocyto- 1-2 days) for patients with aseptic men- the predictors of the Bacterial Menin- sis, 121 (3.7%; 95% CI, 3.1%-4.4%) had ingitis. No deaths occurred among the gitis Score (Box). These variables were bacterial meningitis and 3174 (96.3%; patients with bacterial meningitis who age, enteroviral season (June through 95% CI, 95.5%-96.9%) had aseptic met the study inclusion criteria (and October), seizure at or before presen- meningitis. thus did not present critically ill). Only tation, duration of fever, triage tem- Characteristics of patients with bac- 1patient,a17-year-oldwithaseptic perature, meningismus, peripheral terial and aseptic meningitis are shown meningoencephalitis, died. blood white blood cell count, peripheral blood ANC, CSF white blood cell count, CSF ANC, CSF glucose, CSF Figure. Patient Flow Diagram, Including the Classification Performance of the Bacterial protein, and CSF Gram stain. Continu- Meningitis Score ous variables were introduced into this 4369 Children With CSF Pleocytosis analysis as continuous predictors, as the Met Inclusion Criteria recursive partitioning algorithm iden- tifies the cutoff point for continuous 1074 Excluded∗ predictors, which maximizes the accu- 515 Required Admission Regardless racy for distinguishing patients with and of Risk of Bacterial Meningitis 218 Were Critically Ill 25 without the outcome of interest. Pa- 15 Had Purpura tients with missing variables were in- 214 Had Ventricular Shunt or Recent Neurosurgery cluded in the recursive partitioning 71 Were Immunosuppressed analysis because the algorithm substi- 85 Had Other Bacterial Infection Requiring Antibiotic Treatment tutes patient “surrogate” variables that 39 Had Active Lyme Disease most closely approximate the missing 544 Received Antibiotic Treatment Before Lumbar Puncture predictor with regard to partitioning the 15 No CSF Culture Obtained data. Continuous predictors that were

identified by the algorithm as impor- 3295 Children Eligible for Prediction Rule tant for the final decision tree were then

rounded slightly to make the resulting 392 Not Evaluated by Prediction Rule tree easier to use and interpret in the (Missing Predictor Data) clinical setting. Finally, we used 10- fold cross-validation to internally vali- 2903 Evaluated by Prediction Rule date the results of the recursive partitioning analysis.25 This iterative process results in an accurate estimate of the 1714 Very Low Risk (None of 5 Predictors†) 1189 Not Very Low Risk (Any of 5 Predictors†) misclassification rate of the decision tree on future cohorts of patients by using 1714 Had Reference Standard Test‡ 1189 Had Reference Standard Test‡ statistical resampling of the patients in the database.

We performed statistical analyses us- 2 Had Bacterial 1712 Had Aseptic 119 Had Bacterial 1070 Had Aseptic ing SPSS.26 Recursive partitioning analy- Meningitis Meningitis Meningitis Meningitis ses were performed using CART soft- 27 CSF indicates cerebrospinal fluid. ware. *Patients could be excluded for more than 1 reason. †For list of predictors, see the Box. RESULTS ‡Reference standard test was the determination of either bacterial meningitis or aseptic meningitis. We defined achildashavingbacterialmeningitisiftherewasapositiveCSFculture,CSFpleocytosisinassociationwitha Patients positive blood culture for a bacterial pathogen, or CSF pleocytosis in association with a positive CSF latex agglu- We identified 4369 children who met tination test for a bacterial pathogen. We defined a child as having aseptic meningitis if there was CSF pleocytosis with negative bacterial cultures of blood and CSF and a negative CSF latex agglutination test (if obtained). the inclusion criteria (FIGURE). We ex-

Downloaded From: http://jama.jamanetwork.com/ by a Cincinnati Children's Hospital User on 09/06/2016 VALIDATION OF A CLINICAL PREDICTION RULE FOR BACTERIAL MENINGITIS

The etiology of bacterial meningitis gram-negative rods (7 [6%]), Listeria and blood culture alone in 9 (7%). No was as follows: Spneumoniae(35 pa- monocytogenes (3 [2.5%]), and group patient had a positive CSF latex agglu- tients [29%]), Neisseria meningitidis (33 A Streptococcus (3 [2.5%]). The bac- tination test without either a positive [27%]), group B Streptococcus (24 terial pathogen was identified by both CSF or blood culture. [20%]), Escherichia coli (9 [7%]), Hin- CSF and blood culture in 65 patients Of the patients with aseptic menin- fluenzae (7 [6%, all nontypeable]), other (54%), CSF culture alone in 47 (39%), gitis, 1128 (36%) had enteroviral polymerase chain reaction testing performed, 391 (12%) herpes simplex virus Table 1. Characteristics of the 3295 Study Patients* polymerase chain reaction, 615 (19%) Bacterial Meningitis Aseptic Meningitis Characteristics (n = 121) (n = 3174) viral culture, and 231 (7%) B burgdor- Age, median (IQR), y 0.4 (0.2-2.6) 4.6 (0.2-9.8) feri serology. Fifty-two percent of tested Male sex 83 (69) 1836 (58) patients had a specific pathogen iden- Presentation during enteroviral season† 34 (28) 2174 (69) tified (enterovirus: n=839 [74% of all History of seizure before or at the time 7 (6) 80 (3) patients tested]; herpes simplex virus: of presentation n=6 [2%]; and Bburgdorferi:n=24 Peripheral blood, median (IQR), cells/µL [10%]). White blood cell count 14 400 (8600-22 000) 10 700 (8300-13 900) Absolute neutrophil count 10 176 (3759-17 315) 5890 (3604-8786) Performance of the Bacterial Cerebrospinal fluid, median (IQR) Meningitis Score White blood cell count, cells/µL 1240 (274-3435) 120 (40-300) Absolute neutrophil count, cells/µL 952 (155-2784) 29 (7-112) The Bacterial Meningitis Score was Glucose, mg/dL 30 (12-53) 56 (49-64) calculated for 2903 (88%) of 3295 Protein, mg/dL 171 (85-251) 47 (32-69) study patients and could not be calcu- Positive cerebrospinal fluid Gram stain 74 (61) 6 (0.2) lated for 392 patients (12%) due to Admitted to the hospital 121 (100) 2518 (80) missing predictor data (none of the Duration of parenteral antibiotics, median (IQR), d 14 (10-14) 2 (1-2) 392 had bacterial meningitis). Among Abbreviation: IQR, interquartile range. the 2903 patients, the frequency of SI conversion: To convert glucose to mmol/L, multiply by 0.0555. *Data are presented as number (percentage) unless otherwise specified. bacterial meningitis increased with †From June to October inclusive. greater numbers of additional Bacte- rial Meningitis Score risk factors (2 Table 2. Risk of Bacterial Meningitis for Patients With 1, 2, or 3 or More Bacterial Meningitis [0.1%] of 1714 patients with no risk Score Predictors factors; 24 [3%] of 924 with 1 risk No. of No. (%) of factor; 37 [27%] of 133 patients with Bacterial Meningitis Children With Children With Score Predictors Present CSF Pleocytosis Bacterial Meningitis 2riskfactors;40[70%]of57patients 1Predictor with 3 risk factors; and 18 [95%] of Positive CSF Gram stain 12 7 (58) 19 patients with Ն4riskfactors). CSF ANC Ն1000 cells/µL 11 1 (9) TABLE 2 demonstrates the risk of bac- CSF protein Ն80 mg/dL 445 8 (2) terial meningitis for patients with CSF Peripheral blood ANC Ն10 000 cells/µL 413 7 (2) pleocytosis with either 1, 2, or 3 or History of seizure before or at the time of presentation 43 1 (2) more Bacterial Meningitis Score pre- 2Predictors dictors. In addition, TABLE 3 demon- Positive CSF Gram stain and CSF ANC Ն1000 cells/µL 1 1 (100) strates the odds of bacterial meningi- Positive CSF Gram stain and CSF protein Ն80 mg/dL 14 14 (100) tis both for each Bacterial Meningitis Positive CSF Gram stain and peripheral blood ANC 54(80) Ն10 000 cells/µL Score predictor individually as well as Positive CSF Gram stain and seizure 0 NA after adjusting for all the other Bacte- CSF ANC Ն1000 cells/µL and CSF protein Ն80 mg/dL 30 10 (33) rial Meningitis Score predictors. All CSF ANC Ն1000 cells/µL and peripheral blood ANC 10 0 variables remained significantly asso- Ն10 000 cells/µL ciated with bacterial meningitis; CSF CSF ANC Ն1000 cells/µL and seizure 0 NA Gram stain was the most highly CSF protein Ն80 mg/dL and peripheral blood ANC 46 7 (15) associated. Ն10 000 cells/µL CSF protein Ն80 mg/dL and seizure 12 1 (8) Of 1714 patients categorized as very Peripheral blood ANC Ն10 000 cells/µL and seizure 15 0 low risk by the Bacterial Meningitis Score, Ն3Predictors 2hadbacterialmeningitisand1712had All combinations 76 58 (76) aseptic meningitis (negative predictive Total patients with Ն1predictor 1133 119(10.5) value, 99.9%; 95% CI, 99.6%-100%). Of Abbreviations: ANC, absolute neutrophil count; CSF, cerebrospinal fluid; NA, not applicable. the 1189 patients categorized as not very

Downloaded From: http://jama.jamanetwork.com/ by a Cincinnati Children's Hospital User on 09/06/2016 VALIDATION OF A CLINICAL PREDICTION RULE FOR BACTERIAL MENINGITIS

low risk by the Bacterial Meningitis Score, Table 3. Bivariate and Multivariate Adjusted Odds Ratios of Bacterial Meningitis for Each of 119 (10%) had bacterial meningitis and the Bacterial Meningitis Score Predictors 1070 (90%) had aseptic meningitis. The Odds Ratio Adjusted Odds Ratio sensitivity of the Bacterial Meningitis Predictor (95% CI) (95% CI)* Score (ie, having Ն1BacterialMeningi- Positive cerebrospinal fluid Gram stain 866.1 (358.4-2093.2) 653.7 (216.6-1972.8) tis Score risk factor) for bacterial men- Cerebrospinal fluid ANC Ն1000 cells/µL 47.2 (30.3-73.6) 8.0 (3.8-17.0) ingitis was 98.3% (119/121 patients with Cerebrospinal fluid protein Ն80 mg/dL 17.9 (11.3-28.3) 12.2 (5.7-26.0) aBacterialMeningitisScorecalculated; Peripheral blood ANC Ն10 000 cells/µL 4.8 (3.3-6.9) 4.1 (2.2-8.0) 95% CI, 94.2%-99.8%) and the specific- History of seizure before or at the time of presentation 2.4 (1.1-5.3) 3.7 (1.0-13.4) ity was 61.5% (1712/2782; 95% CI, Abbreviations: ANC, absolute neutrophil count; CI, confidence interval. *Adjusted for the other Bacterial Meningitis Score predictors. 59.7%-63.3%). The positive and negative likelihood ratios were 2.56 (95% CI, 2.43-2.69) and 0.03 (95% CI, 0.01- Table 4. Characteristics of the 2 Patients With Bacterial Meningitis Misclassified by the 0.11), respectively. Bacterial Meningitis Score Characteristics Patient 1 Patient 2 Misclassified Patients Patient age, mo 1.8 1.0 The Bacterial Meningitis Score misclas- Peripheral blood, cells/µL sified 2 patients with bacterial meningi- White blood cell count 12 300 12 300 tis as having aseptic meningitis. Both of Absolute neutrophil count 8100 6600 these patients were infants between 1 and Cerebrospinal fluid 2monthsoldwithEcolimeningitis and White blood cell count, cells/µL 23 540 urinary tract infections, but with nega- Absolute neutrophil count, cells/µL 0 497 Glucose, mg/dL 53 52 tive urinalyses at presentation (TABLE 4). Protein, mg/dL 31 65 Bacterial Meningitis Gram stain Negative Negative Score Refinement History of seizure before or at the time of presentation None None Bacterial pathogen isolated (both from cerebrospinal fluid and urine) Escherichiacoli Escherichiacoli We attempted to refine the prediction SI conversion: To convert glucose to mmol/L, multiply by 0.0555. model using recursive partitioning analysis. The resulting decision tree identified the following predictors of COMMENT ingitis Score, despite the greater num- bacterial meningitis in order of impor- We have previously derived and vali- ber, retain intuitive appeal and are eas- tance: CSF protein level of 80 mg/dL or dated the Bacterial Meningitis Score at ily applied by the clinician. Therefore, higher, positive CSF Gram stain, and asingleinstitutiontoidentifychil- we believe the Bacterial Meningitis peripheral ANC of 10 000 cells/µL or dren with CSF pleocytosis who are at Score is the more appropriate predic- more. Of 1786 patients with none of the very low risk of bacterial meningitis.1 tion rule for assisting the clinician with 3 variables in the recursive partition- In the current multicenter validation the identification of children with CSF ing model, 3 (0.2%) had bacterial men- study in the era of widespread pneu- pleocytosis at very low risk for bacte- ingitis (negative predictive value, mococcal conjugate vaccination, the rial meningitis. 99.8%; 95% CI, 99.5%-100%). The re- Bacterial Meningitis Score accurately Previously, other investigators have cursive partitioning model misclassi- identified patients at very low risk of developed clinical prediction rules to fied the same 2 infants with Ecolimen- bacterial meningitis, misclassifying only distinguish aseptic from bacterial meningitis missed by the Bacterial 0.1% of patients categorized as very low ingitis at the time of clinical presenta- Meningitis Score as well as an addi- risk for bacterial meningitis. Our at- tion. These models have included the tional patient with nontypeable H in- tempts to refine the model using recur- following parameters: age,28-30 time of fluenzae meningitis. This patient was a sive partitioning analysis led us to a the year,28 peripheral ANC,29 CSF white 3½-month-old child who presented somewhat simpler model that relied on blood cell count or ANC,28-32 CSF- with a CSF ANC of 12 cells/µL, CSF 3 variables only; however, this predic- blood glucose ratio,28,29,31,33 CSF pro- protein of 51 mg/dL, negative Gram tive model misclassified an additional tein,29,30 and CSF Gram stain.29 In our stain, peripheral blood ANC of 4500 patient with bacterial meningitis com- 20-center study, clinicians obtained pe- cells/µL, and a seizure before presen- pared with the Bacterial Meningitis ripheral glucose for only 50% of the tation. This patient was correctly iden- Score. Furthermore, the refined model study patients, making prediction rules tified by the Bacterial Meningitis Score lacks any measure of CSF white blood that rely on CSF-blood glucose ratio less as having bacterial meningitis on the ba- cell count or CSF ANC, which may applicable. Furthermore, a prediction sis of a seizure before presentation to limit its acceptability to a clinician. The rule using a complicated fractional poly- the emergency department. predictor variables of the Bacterial Men- nomial equation to calculate risk of bac-

Downloaded From: http://jama.jamanetwork.com/ by a Cincinnati Children's Hospital User on 09/06/2016 VALIDATION OF A CLINICAL PREDICTION RULE FOR BACTERIAL MENINGITIS

terial meningitis would require auto- sures and the fifth, a history of seizure, nevertheless have infections that re- mation to be applied in the clinical is a fairly objective clinical measure. Pa- quire antimicrobial therapy, such as setting.30,32 Previously published pre- tients were classified as having a sei- Lyme meningitis or herpes simplex vi- diction rules were derived before the zure for any abnormal neurological ac- rus encephalitis. Therefore, the Bacte- introduction of conjugate Hinfluen- tivity thought to possibly be a seizure to rial Meningitis Score should be used in zae type b28 and Spneumoniaevac- minimize the risk of variability in inter- concert with careful clinical assess- cines,1,29-35 and were not internally29,35 pretation. Therefore, we think our study ment of the patient, which would in- or externally validated.12,28-35 In con- is nearly equivalent to a prospective vali- clude consideration of these other im- trast, the Bacterial Meningitis Score pro- dation. Thus, based on the 2 small stud- portant treatable infections. In addition, vides a simple scoring system com- ies and our validation study, our find- the Bacterial Meningitis Score is de- posed of easily collected data elements ings are likely to be widely generalizable signed to serve as an assistive clinical pre- that could routinely be implemented by and helpful in guiding clinical decision diction rule to help guide clinical clinicians in the acute management of making, as the patient population en- decision making, and not to serve as a children with CSF pleocytosis. To our compasses a wide spectrum of ages, directive decision rule that explicitly dic- knowledge, this is the first bacterial clinical settings, geographic regions, and tates clinical care.15 We would particu- meningitis prediction model to be both seasons. larly caution against the use of the Bac- studied in the era of widespread con- Our study has some limitations. First, terial Meningitis Score for infants jugate pneumococcal vaccine use and our study was retrospective and there- younger than 2 months for whom the externally validated. fore subject to potential information Bacterial Meningitis Score may be less Our validation study was con- bias. However, the potential impact of accurate, and who may not be appro- ducted using strict methodological stan- this limitation is minimal because the priate candidates for outpatient man- dards.13,18,36 Our inclusion and exclu- Bacterial Meningitis Score includes only agement. In this subgroup of the 792 sion criteria were such that only those objective clinical characteristics and children younger than 2 months (of patients with CSF pleocytosis who laboratory parameters. Furthermore, we whom 26 had bacterial meningitis), the could be reasonably considered for out- used strict criteria to define the out- classification performance of the Bacte- patient management were included in come variable (bacterial meningitis) to rial Meningitis Score was sensitivity of the study (ie, we excluded patients with minimize misclassification bias. Cere- 92.3% (95% CI, 74.9%-99.1%), speci- other reasons for hospital admission or brospinal fluid cultures were available ficity of 56.3% (95% CI, 52.7%- with critical illness). Both the predic- for all included patients and blood cul- 59.8%), and negative predictive value of tors and outcome measure were clearly tures were available for 90% of the study 99.5% (95% CI, 98.3%-99.9%). Fi- and objectively defined. By careful stan- patients. Although it is conceivable that nally, the Bacterial Meningitis Score dardized chart reviews, we had mini- some of the patients with no blood cul- should also not be used to guide deci- mal missing data for the included clini- ture obtained may have had bacterial sion making for children pretreated with cal and laboratory predictors and were meningitis, this seems unlikely given antibiotics in whom the diagnosis of able to assign an appropriate outcome that CSF cultures were negative in all aseptic meningitis is difficult and whose (aseptic vs bacterial meningitis) for all of these patients. In addition, none of pretreatment may have affected CSF pro- patients. the patients who did not have blood cul- files.22,23 The Bacterial Meningitis Score has al- tures drawn (except for those with For patients with at least 1 Bacterial ready been independently validated Lyme meningitis) received a standard Meningitis Score risk factor or who are (prospectively and retrospectively) in 2 course of antibiotics for bacterial men- younger than 2 months, we suggest ad- small pediatric studies in France and Bel- ingitis or a diagnosis of bacterial men- mission to the hospital and adminis- gium (166 and 277 study patients with ingitis by the treating clinician. It is also tration of parenteral antibiotics. For the meningitis, respectively), and shown to possible that we may have missed po- 2111 patients older than 2 months in perform very well.37,38 None of the pa- tentially eligible study patients due to our study (of whom 95 had bacterial tients classified by the Bacterial Menin- errors in emergency department diag- meningitis), the Bacterial Meningitis gitis Score in the very low risk category nosis coding. However, we attempted Score was highly accurate. The classi- had bacterial meningitis (negative pre- to capture all cases of children with bac- fication performance of the Bacterial dictive value, 100% for both studies). Al- terial meningitis by cross-checking the Meningitis Score for identifying bacte- though a large prospective validation institution’s microbiology test results rial meningitis for these children was would be preferable to a retrospective and by including all patients with CSF sensitivity of 100% (95% CI, 96.9%- validation, such a study would be diffi- cultures growing bacterial pathogens. 100%), specificity of 63.5% (95% CI, cult to accomplish given the very low in- Because our model was designed to 61.4%-65.6%), and negative predic- cidence of bacterial meningitis. Four of identify patients at very low risk for bac- tive value of 100% (95% CI, 99.8%- the 5 predictors in the Bacterial Menin- terial meningitis, some patients with no 100%). For patients older than 2 gitis Score are objective laboratory mea- predictors of bacterial meningitis may months with a Bacterial Meningitis

Downloaded From: http://jama.jamanetwork.com/ by a Cincinnati Children's Hospital User on 09/06/2016 VALIDATION OF A CLINICAL PREDICTION RULE FOR BACTERIAL MENINGITIS

Score score of 0 and who are well ap- ferson Medical College, Wilmington, Del (Dr Ben- data, or in the preparation, review, or approval of the nett); Department of Pediatrics, Children’s Memorial manuscript. All members of the Scientific Review Com- pearing, physicians could consider 2 op- Hospital and Northwestern University Feinberg School mittee of the Pediatric Emergency Medicine Collabo- tions: admission for observation or, in of Medicine, Chicago, Ill (Dr Katsogridakis); Depart- rative Research Committee of the American Acad- ment of Emergency Medicine, Children’s Medical Cen- emy of Pediatrics reviewed the study protocol. the proper clinical context and if ad- ter and Medical College of Georgia, Augusta (Dr Moh- Acknowledgment: We thank other members of the equate follow-up is available, outpa- seni); Department of Emergency Medicine, Phoenix study group, including those who helped with chart tient management. Because the conse- Children’s Hospital and University of Arizona Col- abstraction, data entry, and database management: lege of Medicine, Phoenix (Dr Bulloch); Departments Elizabeth R. Alpern, MD, MSCE (Children’s Hospital quences of missing bacterial meningitis of Emergency Medicine and Pediatrics, Hasbro Chil- of Philadelphia and University of Pennsylvania, Phila- could be devastating, however, we dren’s Hospital and Brown Medical School, Provi- delphia), Troy Bush (Texas Children’s Hospital and Bay- dence, RI (Dr Steele); Department of Emergency Medi- lor College of Medicine, Houston), Joseph M. Cam- would recommend serious consider- cine, Children’s Hospital and Regional Medical Center, pos, PhD (Children’s National Medical Center, ation of administration of a long- and University of Washington School of Medicine, Se- Washington, DC), Murray Edelberg, PhD (Carlisle, attle (Dr Kaplan); Department of Pediatrics, Le Bon- Mass), Kim Fisher, PhD (Center for Pediatric Re- acting parenteral antibiotic if the pa- heur Children’s Medical Center and University of Ten- search, University of Tennessee, Memphis), Marissa tient is to be discharged from the nessee Health Science Center, College of Medicine, Hauptman (Brown Medical School, Providence, RI), emergency department. Memphis (Dr Herman); Department of Emergency Paul Ishimine, MD (Rady Children’s Hospital San Di- Medicine, Children’s Hospital of Wisconsin and Medi- ego Medical Center and University of California, San In the conjugate H influenzae type b cal College of Wisconsin, Milwaukee, and Pediatric Diego School of Medicine, San Diego), Daniel M. Kap- and pneumococcal vaccines era, bac- Emergency Medicine Associates LLC, Children’s Health- lan (Children’s National Medical Center and George care of Atlanta at Scottish Rite, Atlanta, Ga (Dr Ban- Washington University School of Medicine, Wash- terial meningitis has become an un- dyopadhyay); Department of Emergency Medicine, ington, DC), John Leake, MD, and R. Ian McCaslin, common disease in US children.39 Morgan Stanley Children’s Hospital of New York- MD, MPH (Rady Children’s Hospital San Diego Medi- Presbyterian and Columbia University College of Phy- cal Center and University of California, San Diego Therefore, the majority of children with sicians and Surgeons, New York, NY (Dr Dayan); De- School of Medicine, San Diego), Umair Salim (Center CSF pleocytosis have aseptic rather than partment of Pediatrics, Children’s Hospital Los Angeles for Pediatric Research, University of Tennessee, Mem- bacterial meningitis. Furthermore, our and Keck School of Medicine, University of Southern phis), James Wilde, MD (Children’s Medical Center California, Los Angeles (Dr Wang); and Department and Medical College of Georgia, Augusta), and Xiaohi study confirms that most children with of Emergency Medicine, Columbus Children’s Hos- Zhao (Brookline, Mass). None of the persons acknowl- CSF pleocytosis are admitted to the hospital and The Ohio State University, Columbus (Drs edged received any financial compensation for their Bonsu and Chapman). work. pital to receive parenteral antibiotics Author Contributions: Dr Nigrovic had full access to while awaiting bacterial culture test re- all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data sults. Using the Bacterial Meningitis analysis. REFERENCES Score prediction rule to assist with clini- Study concept and design: Nigrovic, Kuppermann, Malley. 1. Nigrovic LE, Kuppermann N, Malley R. Develop- cal decision making could substan- ment and validation of a multivariable predictive model Acquisition of data: Nigrovic, Macias, Cannavino, to distinguish bacterial from aseptic meningitis in chil- tially reduce unnecessary hospital ad- Moro-Sutherland, Schremmer, Schwab, Agrawal, Man- dren in the post-Haemophilus influenzae era. sour, Bennett, Katsogridakis, Mohseni, Bulloch, Steele, missions for children with CSF Pediatrics.2002;110:712-719. Kaplan, Herman, Bandyopadhyay, Dayan, Truong, 2. Centers for Disease Control and Prevention. Out- pleocytosis at very low risk of bacte- Wang, Bonsu, Chapman, Kanegaye. breaks of aseptic meningitis associated with echovi- Analysis and interpretation of data: Nigrovic, rial meningitis. Future investigations ruses 9 and 30 and preliminary surveillance reports on Kuppermann, Malley. should study the clinical implementa- enterovirus activity—United States, 2003. MMWR Drafting of the manuscript: Nigrovic, Kuppermann, tion of the Bacterial Meningitis Score Morb Mortal Wkly Rep.2003;52:761-764. Malley. 3. Fleisher GR. Central nervous system infections. In: as a guide to help care for children with Critical revision of the manuscript for important intel- Fleisher GR, Ludwig SL, Henretig FM, eds. Textbook CSF pleocytosis. lectual content: Macias, Cannavino, Moro-Sutherland, of Pediatric Emergency Medicine. 5th ed. Philadel- Schremmer, Schwab, Agrawal, Mansour, Bennett, phia, Pa: Lippincott Williams & Williams; 2006:788- Author Affiliations: Department of Medicine, Chil- Katsogridakis, Mohseni, Bulloch, Steele, Kaplan, Herman, 795. dren’s Hospital Boston and Harvard Medical School, Bandyopadhyay, Dayan, Truong, Wang, Bonsu, 4. Jansen GJ, Mooibroek M, Idema J, Harmsen HJ, Boston, Mass (Drs Nigrovic and Malley); Depart- Chapman, Kanegaye. Welling GW, Degener JE. Rapid identification of bac- ments of Emergency Medicine (Dr Kuppermann) and Statistical analysis: Nigrovic, Kuppermann, Malley. teria in blood cultures by using fluorescently labeled Pediatrics (Drs Kuppermann and Truong), University Obtained funding: Nigrovic. oligonucleotide probes. JClinMicrobiol.2000;38:814- of California, Davis Medical Center, Sacramento, and Financial Disclosures: None reported. 817. University of California, Davis School of Medicine, Members of the Pediatric Emergency Medicine Col- 5. Poppert S, Essig A, Stoehr B, et al. Rapid diagnosis Davis; Department of Pediatrics, Texas Children’s Hos- laborative Research Committee of the American Acad- of bacterial meningitis by real-time PCR and fluores- pital and Baylor College of Medicine, Houston (Dr Ma- emy of Pediatrics Scientific Review Committee are cence in situ hybridization. JClinMicrobiol.2005;43: cias); Department of Emergency Medicine, Rady Chil- Denise Dowd, MD, Children’s Mercy Hospitals and 3390-3397. dren’s Hospital San Diego Medical Center (Dr Clinics and University of Missouri-Kansas City School 6. Schuchat A, Robinson K, Wenger JD, et al. Bacte- Kanegaye) and Department of Pediatrics, University of Medicine, Kansas City; Marc Gorelick, MD, Chil- rial meningitis in the United States in 1995: active of California, San Diego School of Medicine, San Di- dren’s Hospital of Wisconsin and Medical College of surveillance team. NEnglJMed.1997;337:970- ego (Drs Cannavino and Kanegaye); Wake Emer- Wisconsin, Milwaukee; Louis C. Hampers, MD, MBA, 976. gency Physicians, PA and WakeMed Health and Hos- University of Colorado School of Medicine and The 7. Peltola H, Salo E, Saxen H. Incidence of Haemo- pitals, Raleigh, NC (Dr Moro-Sutherland); Department Children’s Hospital, Denver; Philip Scribano, MD, Co- philus influenzae type b meningitis during 18 years of Pediatrics, Children’s Mercy Hospitals and Clinics lumbus Children’s Hospital and The Ohio State Uni- of vaccine use: observational study using routine hos- and University of Missouri-Kansas City School of Medi- versity College of Medicine, Columbus; and Milton pital data. BMJ.2005;330:18-19. cine, Kansas City (Dr Schremmer); Department of Pe- Tenenbein, MD, Children’s Hospital Winnipeg and Uni- 8. Black S, Shinefield H, Fireman B, et al; Northern Cali- diatrics, Children’s Hospital of Philadelphia and Uni- versity of Manitoba, Winnipeg. fornia Kaiser Permanente Vaccine Study Center Group. versity of Pennsylvania, Philadelphia (Dr Schwab); Funding/Support: This work was supported by the Am- Efficacy, safety and immunogenicity of heptavalent Department of Emergency Medicine, Children’s Na- bulatory Pediatric Association Young Investigator Grant pneumococcal conjugate vaccine in children. Pediatr tional Medical Center and George Washington Uni- and the National Research Service Award (T32 Infect Dis J.2000;19:187-195. versity School of Medicine, Washington, DC (Dr HD40128-01; Research Training in Pediatric Emer- 9. Black S, Shinefield H, Baxter R, et al. Postlicensure Agrawal); Department of Emergency Medicine, Chil- gency Medicine), which was awarded to Dr Nigrovic. surveillance for pneumococcal invasive disease after dren’s Hospital and Research Center Oakland, Oak- Role of the Sponsors: The sponsors did not partici- use of heptavalent pneumococcal conjugate vaccine land, Calif (Dr Mansour); Department of Pediatrics, pate in the design or conduct of the study, in the col- in Northern California Kaiser Permanente. Pediatr In- Alfred I. duPont Hospital for Children and Thomas Jef- lection, management, analysis, or interpretation of the fect Dis J.2004;23:485-489.

Downloaded From: http://jama.jamanetwork.com/ by a Cincinnati Children's Hospital User on 09/06/2016 VALIDATION OF A CLINICAL PREDICTION RULE FOR BACTERIAL MENINGITIS

10. Whitney CG, Farley MM, Hadler J, et al. Decline 19. Wasson JH, Sox HC. Clinical prediction rules: have terial meningitis from acute viral meningitis among chil- in invasive pneumococcal disease after the introduc- they come of age? JAMA.1996;275:641-642. dren with cerebrospinal fluid pleocytosis: a multivari- tion of protein-polysaccharide conjugate vaccine. 20. Rubenstein JS, Yogev R. What represents pleo- able regression model. Pediatr Infect Dis J.2004;23: NEnglJMed.2003;348:1737-1746. cytosis in blood-contaminated (“traumatic tap”) cere- 511-517. 11. Kaplan SL, Mason EO Jr, Wald ER, et al. De- brospinal fluid in children? JPediatr.1985;107:249- 31. Oostenbrink R, Moons KG, Twijnstra MJ, Grobbee crease of invasive pneumococcal infections in 251. DE, Moll HA. Children with meningeal signs: predict- children among 8 children’s hospitals in the United 21. Mayefsky JH, Roghmann KJ. Determination of leu- ing who needs empiric antibiotic treatment. Arch Pe- States after the introduction of the 7-valent pneumo- kocytosis in traumatic spinal tap specimens. Am J Med. diatr Adolesc Med.2002;156:1189-1194. coccal conjugate vaccine. Pediatrics.2004;113:443- 1987;82:1175-1181. 32. Bonsu BK, Harper MB. Explanation of mathemati- 449. 22. Blazer S, Berant M, Alon U. Bacterial meningitis: cal model. Pediatr Infect Dis J.2004;23:893. 12. Kyaw MH, Lynfield R, Schaffner W, et al. Effect effect of antibiotic treatment on cerebrospinal fluid. 33. Oostenbrink R, Moons KG, Derksen-Lubsen AG, of introduction of the pneumococcal conjugate vac- Am J Clin Pathol.1983;80:386-387. Grobbee DE, Moll HA. A diagnostic decision rule for cine on drug-resistant Streptococcus pneumoniae. 23. Kaplan SL, Smith EO, Wills C, Feigin RD. Asso- management of children with meningeal signs. Eur J NEnglJMed.2006;354:1455-1463. ciation between preadmission oral antibiotic therapy Epidemiol.2004;19:109-116. 13. Wasson JH, Sox HC, Neff RK, Goldman L. Clini- and cerebrospinal fluid findings and sequelae caused 34. Oostenbrink R, Moll HA, Moons KG, Grobbee cal prediction rules: applications and methodological by Haemophilus influenzae type b meningitis. Pedi- DE. Predictive model for childhood meningitis. Pedi- standards. NEnglJMed.1985;313:793-799. atr Infect Dis.1986;5:626-632. atr Infect Dis J.2004;23:1070-1071. 14. Martin TP, Hanusa BH, Kapoor WN. Risk strati- 24. Microsoft Access [computer program]. Version 35. Negrini B, Kelleher KJ, Wald ER. Cerebrospinal fluid fication of patients with syncope. Ann Emerg Med. 2003. Seattle, Wash: Microsoft; 2003. findings in aseptic versus bacterial meningitis. Pediatrics. 1997;29:459-466. 25. Breiman L, Friedman JH, Olshen RA, Stone CJ. Clas- 2000;105:316-319. 15. Reilly BM, Evans AT. Translating clinical research sification and Regression Trees. Washington, DC: 36. Stiell IG, Wells GA. Methodologic standards into clinical practice: impact of using prediction rules Chapman & Hall; 1994. for the development of clinical decision rules in to make decisions. Ann Intern Med.2006;144: 26. SPSS for Windows [computer program]. Version emergency medicine. Ann Emerg Med.1999;33:437- 201-209. 12.0.1. Chicago, Ill: SPSS Inc; 2003. 447. 16. McGinn TG, Guyatt GH, Wyer PC, et al; Evidence- 27. CART for Windows [computer program]. Ver- 37. Dubos F, Lamotte B, Bibi-Triki F, et al. Clinical de- Based Medicine Working Group. Users’ guides to the sion 5.0. Stanford, Calif: Statistical Software Inc; 2005. cision rules to distinguish between bacterial and asep- medical literature, XXII: how to use articles about clini- 28. Spanos A, Harrell FE Jr, Durack DT. Differential tic meningitis [published ahead of print April 4, 2006]. cal decision rules. JAMA.2000;284:79-84. diagnosis of acute meningitis: an analysis of the pre- Arch Dis Child.2006;91:647-650. 17. Charlson ME, Ales KL, Simon R, MacKenzie CR. dictive value of initial observations. JAMA.1989;262: 38. Pierart J, LePage P. Utilite du “Bacterial Menin- Why predictive indexes perform less well in valida- 2700-2707. gitis Score (BMS)” dans le diagnostic differentiel des tion studies: is it magic or methods? Arch Intern Med. 29. Freedman SB, Marrocco A, Pirie J, Dick PT. Pre- meningites baceteriennes et virales [in French]. Rev 1987;147:2155-2161. dictors of bacterial meningitis in the era after Haemo- Med Liege.2006;61:581-585. 18. Laupacis A, Sekar N, Stiell IG. Clinical prediction philus influenzae. Arch Pediatr Adolesc Med.2001; 39. Chavez-Bueno S, McCracken GH Jr. Bacterial men- rules: a review and suggested modifications of meth- 155:1301-1306. ingitis in children. Pediatr Clin North Am.2005;52:795- odological standards. JAMA.1997;277:488-494. 30. Bonsu BK, Harper MB. Differentiating acute bac- 810.

Downloaded From: http://jama.jamanetwork.com/ by a Cincinnati Children's Hospital User on 09/06/2016