PEDIATRICS Vol. 106 No. 3 September 2000, pp. 477-482

Predicting the Outcome of Neonatal Bacterial Meningitis

Gil Klinger, MD^*, Choy-Nyok Chin, MBBS, MRCP^*, Joseph Beyene, MSc, and Max Perlman, MB, BS, FRCP(Lond), FRCPC^*

From the ^* Division of Neonatology, Hospital for Sick Children; and the  Division of Neonatology, Mount Sinai Hospital and University of Toronto, Toronto, Ontario, Canada.

	ABSTRACT

Top Abstract Methods Results Discussion References

Objective.  To build predictive models of severe adverse outcome at various times in the course of neonatal bacterial meningitis.

Study Design.  Retrospective cohort study with follow-up to a minimum age of 1 year of term and near-term infants, admitted between 1979 and 1998 to a regional tertiary care center. Predictors of adverse outcome detectable at 1 year of age (death or moderate or severe neurosensory impairment) were identified by univariate analysis. Independent predictors of adverse outcome were identified by multivariate analysis. Predictive tree models were constructed at 12, 24, 48, and 96 hours after admission and at discharge.

Results.  Of 101 infants admitted with definitive bacterial meningitis, 13 died and 17 had moderate or severe disability at 1 year of age. Outcomes are known for all patients, to 1 year of age. Twelve hours after admission the important predictors of adverse outcome were presence of seizures, presence of coma, use of inotropes, and leukopenia (sensitivity: 68%; specificity: 100%). At 96 hours the predictors were seizure duration of >72 hours, presence of coma, use of inotropes, and leukopenia (sensitivity: 88%; specificity: 99%).

Conclusions.  Most infants at risk for adverse outcome can be identified within 12 hours of admission. Duration of seizures for >72 hours, presence of coma, use of inotropes, and leukopenia were the most important predictors of adverse outcome. Although these models have good predictive accuracy, they need to be validated in a contemporary cohort in large multicenter studies.bacterial meningitis, neonate, prognostic model.

Long-term morbidity rates of neonatal meningitis (21%-56%) have remained high in the last 2 decades, during which time mortality rates have declined.^1-8 The severity of illness and outcomes of infectious diseases are related to the virulence of the pathogen and to various host factors, including timeliness of diagnosis and treatment initiation.⁸ Recognized associations of severe adverse outcome in neonatal bacterial meningitis include presence of seizures, duration of seizures for >72 hours, presence of coma, hypotension, respiratory distress, markedly elevated cerebrospinal fluid (CSF) protein, hypoglycorrhachia, leukopenia, and thrombocytopenia.^9-16 Because predictive models are not available for neonatal meningitis, estimating the prognosis of the individual patient is based on intuitive processing of the above information.

Our goal was to construct objective user-friendly guides to the prognosis of neonatal bacterial meningitis at a number of time points in the course of the illness. Such guides can help in the provision of relatively accurate prognostic information to parents and in clinical decision-making. Prognostic models are also used to characterize patients for inclusion and for stratification in trials of new adjuvant therapies (eg, pharmacological manipulation of the inflammatory cascade). Predictions also can be used to identify individuals who warrant early follow-up and intervention.

	METHODS

Top Abstract Methods Results Discussion References

Patients

Infants 1 to 28 days of age were admitted through the emergency department or referred from peripheral hospitals between 1979 and 1998 to the Hospital for Sick Children. This hospital is the regional referral center for ~70 000 annual births.

Eligibility Criteria

Inclusion criteria were: 1) diagnosis of bacterial meningitis confirmed by CSF culture; 2) gestational age (GA) 35 weeks; and, 3) age at diagnosis 28 days. The exclusion criteria were coexisting intrapartum asphyxia (American College of Obstetricians and Gynecologists criteria¹⁷), major congenital anomalies, any congenital central nervous system anomaly, known syndrome, genetic condition, or chromosomal abnormality. Premature infants born at GA of <35 weeks were considered at risk for neurologic disability and were not included in this study.

Data Collection

Data on >40 variables were collected retrospectively from the patient records. These were based on maternal and perinatal history, demographic data, neonatal history before diagnosis of meningitis, vital signs at admission, presence of seizures and their characteristics, neurological examination, laboratory data, and treatment given (time to initiation, duration, and type). The reference time (zero time) was time at initial admission.

Outcome data were determined from records of neonatal follow-up and neurology clinics and from readmissions. Where follow-up was incomplete, the infant's family physician or pediatrician was contacted by telephone and asked whether there was any reason why the family should not be contacted. If not, then a form letter requesting a telephone interview, including a consent form and a stamped, addressed envelope to the authors was mailed to the family. This procedure was approved by the Research Ethics Board of the Hospital for Sick Children.

The primary outcome variableadverse outcomewas defined as death or moderate or severe disability at 1 year of age. Moderate or severe disability was defined as severe cerebral palsy (severe impairment of daily activities), moderate to severe developmental delay (determined by a Bayley score of under 2 standard deviations [SDs] for age), blindness, and/or deafness.¹⁸ Data about outcomes for children who had follow-up to school age were also recorded. Adverse outcome to school age was defined as above but also included moderate to severe intellectual disability (Wechsler Intelligence Scale for Children or McCarthy score under 2 SD for age).

Description of Data and Analysis

Frequencies of clinical and laboratory variables, causative organisms, and treatments were computed. These variables and outcome rates were compared for infants born between 1989 and 1998 and those born between 1979 and 1988.

Statistical analysis was performed using SPSS software (SPSS for Windows, Version 9.0, SPSS Inc, Chicago, IL). Univariate analyses were performed using the 2-tailed Fisher's exact test for dichotomous variables and the Student's t test for continuous variables. Multivariate logistic regression analyses were performed entering variables found by univariate analyses to be associated with adverse outcome with P < .05.

Prognostic trees were constructed based on the classification and regression tree method¹⁹ using Answer Tree software (SPSS). Clinically relevant predictors found by univariate analyses to be associated with adverse outcome (P < .10) were analyzed. The prognostic trees were constructed at each of 4 time points after admission: 12, 24, 48, and 96 hours and at the time of discharge. For each prognostic tree, the sensitivity, specificity, positive and negative predictive values, and predictive accuracy were computed.

	RESULTS

Top Abstract Methods Results Discussion References

Of 228 charts reviewed, 101 cases fulfilled the eligibility criteria. Most exclusions were for unconfirmed meningitis or GA <35 weeks. The characteristics of the study population were male:female ratio of 1.3:1; mean GA, 38.3 weeks; mean birth weight, 3.23 kg; mean age at diagnosis, 10.8 days; and mean duration of symptoms before admission, 31 hours. The causative organisms are listed in Table 1.

Forty-six infants were admitted during the period 1979-1988, and 55 infants during the subsequent 10 years. No statistically significant differences were found between these 2 periods regarding age at admission (P = .43), symptom duration (P = .81), presence of seizures (P = .70), coma (P = 1.00), use of inotropes (P = .51), hypotension (P = .80), and duration of ventilation (P = .21). There was a trend toward decreased incidence of streptococcal meningitis in the second period (P = .09).

Outcomes

Patient information was obtained from hospital records for 85 children; for 16 children it was obtained by the procedure previously described. Outcome information was available for all survivors to 1 year of age; the mean age at which the latest outcome information was available was 4.0 (4.1) years (range: 1-19 years). Thirteen infants died; 4 on life support and 7 after its discontinuation. Two infants died at 8 and 26 weeks of age of complications of meningitis. Twelve survivors had adverse outcomes at 1 year of age; most of these had multiple neurological disabilities (Table 2). Four additional children had mild neurological or developmental delay at 1 year of age.

The adverse outcome rates were 11/46 (23.9%) for the period 1979-1988 and 14/55 (25.5%) for the period 1989-1998 (P = .86). Mortality rate was 8/46 (17.4%) for the first period and 5/55 (9.0%) for the second period (P = .23).

Follow-up after 1 year of age identified 2 additional children with adverse outcomes, 1 with moderate and the other with severe disability. Follow-up information was available for 27 children at school age; 7/27 (26%) had moderate to severe disabilities and 4/27 (15%) had isolated learning disabilities.

Univariate Analyses

The results of univariate analyses are summarized in Tables 3 and 4. The main associations with adverse outcome were CSF:blood glucose ratio <.5, hypotension, presence of coma, presence of seizures, use of inotropes, need for ventilation, number of anticonvulsants used, leukopenia, and abnormal neurological examination on discharge. The causative organism was not related to adverse outcome (Table 3).

Multiple Logistic Regression Analyses

Multivariate logistic regression analyses indicated that the independent predictors of adverse outcome at 12 hours postadmission were presence of seizures, presence of coma, need for ventilation, and leukopenia. At 24 hours the independent predictors were seizure duration for >12 hours, presence of coma, and need for ventilatory support. At 48 hours, 96 hours, and at time of discharge, the independent predictors were seizure duration and presence of coma.

Prognostic Trees

Prognostic trees at 12 and 96 hours are presented in Figs 1 and 2, respectively. The characteristics of the prognostic tree models are presented in Table 5. Twelve hours after admission the important predictors of adverse outcome were presence of seizures, presence of coma, use of inotropes, and leukopenia (sensitivity: 68%; specificity: 100%). At 96 hours the predictors were seizure duration of >72 hours, presence of coma, use of inotropes, and leukopenia (sensitivity: 88%; specificity: 99%).

View larger version (16K): [in this window] [in a new window]   Fig. 1.   Predicting outcome of infants with bacterial meningitis at 12 hours postadmission. Each branch of the predictive tree shows the remaining number of infants with adverse outcome (1) and without adverse outcome (0). Each branch divides into 2 branches (eg, the first branch is based on treatment or nontreatment with inotropes), 1 branch fulfilling a specific criterion and the other not. Progression continues towards the terminal branches of the tree. Any individual patient can be compared with our group of patients by following the tree according to the patient's criteria. Each terminal branch shows the percentage of infants with and without adverse outcome.

View larger version (20K): [in this window] [in a new window]   Fig. 2.   Predicting outcome of infants with bacterial meningitis at 96 hours postadmission (see Fig 1 for explanation).

	DISCUSSION

Top Abstract Methods Results Discussion References

We have shown that the relationship between risk factors and outcomes is well-captured by prognostic tree methodology, and that fairly accurate outcome predictions can be made within 12 hours of admission. The risk factors identified by our study are similar to those found in previous studies, with 1 exception, Although seizures were clearly associated with adverse outcomes, infants with seizures that lasted <12 hours did not have these outcomes. Our study reinforces the limited number of published observations from previous prognostic studies of well-defined neonatal bacterial meningitis in term and near-term newborn infants to 28 days of age.

Bacterial meningitis continues to be a serious disease with an unchanging rate of adverse outcome of ~30%, despite a worldwide decline in mortality rate.¹ A recently observed worldwide decline in mortality rate¹ was reflected in our cohort also; however, this decline may have been at the expense of increased numbers of survivors with disability.

Previous attempts to predict the outcome of bacterial meningitis have been limited by small sample size,⁴ wide age ranges,^11,14,16 inadequate follow-up,⁵ and equivocal definitions of bacterial meningitis.¹¹ We included infants with positive CSF cultures and not those with questionable diagnoses. The spectrum of pathogens,^1,2,6-8 the clinical presentation,^11,6-8 and the rate of neurological sequelae¹ in our cohort are similar to those reported by others. These similarities between our study and previous studies suggest that predictions based on our models are likely to be robust and generalizable to referral centers similar to ours.

The majority of infants had neurological follow-up in our hospital. We chose to focus on easily identifiable sequelae that would be unlikely to be missed at 1 year of age. Two children, however, were identified as having adverse outcomes at later ages. One child with a mild developmental delay at 1 year of age was later found to have severe cognitive impairment with autistic features; the second was considered normal at 1 year of age and was identified to have a moderate cognitive impairment at school age. Because this study was designed to detect severe or moderately severe adverse outcomes, we have not attempted to assess the incidence of minor neurological deficits or learning disabilities.

Prognostic trees provide a graphic model that can be grasped intuitively by clinicians. Classification and regression tree methodology creates binary decision trees.¹⁹ At each branch, determined by a single predictor variable, the patients are divided into 2 subgroups. Variables are selected automatically by computer program and can be either categorical or continuous; if the latter, the predictor is evaluated by the program to determine the best cutpoint for predictions. The variable providing the best degree of separation of outcomes is chosen at each level. Selection of variables can be modified at each branch by the investigators' experience.²⁰ The process of constructing a tree is a recursive one, and continues until the best predictive model is created. When prognostic variables are correlated, the choice of one will cause the second to be rejected because it does not provide additional predictive information. In the case of leukopenia and hypotension for example, both are manifestations of septic shock and both are strongly associated with outcome in univariate analysis, but only the former appears in the prognostic tree. All clinically relevant patient data should be tested for inclusion in the prognostic tree model.^19,20 Because >40 variables were collected and analyzed, it is likely that all data known to be relevant were included. These models may be used to stratify patients by severity of illness in therapeutic trials. Useful predictions should be made in a timely fashion. We chose to create the first model at 12 hours, because this is a conventional time for first predictions.²¹ The models were built with the intent to minimize false-positives; such models may aid decisions to discontinue life-sustaining therapy. The prognostic model at 96 hours failed to predict an adverse outcome in 2 infants. These infants were the least severely affected of those with adverse outcome at 1 year of age.

Bacterial meningitis initiates a cascade of events, which include altered cerebral perfusion, disruption of the blood-brain barrier, cerebral edema, intracranial hypertension, and neurological damage.²² Agents that modify mediators of the inflammatory cascade (eg, reactive oxygen species, nitric oxide, and excitatory amino acids) that have been shown in animal models to reduce neurological damage are candidates for future clinical trials of adjuvant therapy in this disease.^23-26

Our prognostic tree models use data that are readily accessible to the clinician, are simple to apply, are intuitively comprehensible, and seem to be relatively accurate. To further develop reliable contemporary models, it will be necessary to perform a prospective, geographically based multicenter observational study. The blueprint should include provisions to predict outcome within an hour or 2 of admission.

	FOOTNOTES

Received for publication Nov 5, 1999; accepted Jan 4, 2000.

Reprint requests to (M.P.) Neonatal Intensive Care Unit, Hospital for Sick Children, University Ave, Toronto, Ontario, Canada, M5G 1X8. E-mail: mperlman{at}sickkids.on.ca

	ABBREVIATIONS

CSF, cerebrospinal fluid; GA, gestational age; SD, standard deviation.

	REFERENCES

Top Abstract Methods Results Discussion References

1. Harvey D, Holt DE, Bedford H Bacterial meningitis in the newborn: a prospective study of mortality and morbidity. Semin Perinatol 1999; 23:218-225 [CrossRef][Medline]
2. de Louvois J, Blackbourn J, Hurley R, Harvey D Infantile meningitis in England and Wales: a two-year study. Arch Dis Child 1991; 66:603-607 [Abstract]
3. Unhanand M, Mustafa MM, McCracken GH, Nelson JD Gram-negative enteric bacillary meningitis: a twenty-one-year experience. J Pediatr 1993; 122:15-21 [Medline]
4. Chin KC, Fitzhardinge PM Sequelae of early-onset group B hemolytic streptococcal neonatal meningitis. J Pediatr 1985; 106:819-822 [CrossRef][Medline]
5. Edwards MS, Rench MA, Haffar AA, Murphy MA, Desmond MM, Baker CJ Long-term sequelae of group B streptococcal meningitis in infants. J Pediatr 1985; 106:717-722 [CrossRef][Medline]
6. Hickey SM, McCracken G. Postnatal bacterial infections. In: Fanaroff AA, Martin RJ, eds. Neonatal-Perinatal Diseases of the Fetus and Infant. 6th ed. St Louis, Mo: Mosby-Year Book, Inc; 1997:731-797
7. Klein JO, Marcy M. Bacterial sepsis and meningitis. In: Remington JS, Klein JO, eds. Infectious Diseases of the Fetus and Newborn Infant. 4th ed. Philadelphia, PA: WB Saunders Co; 1993:835-890
8. Volpe JJ. Bacterial and fungal intracranial infections. In: Volpe JJ, ed. Neurology of the Newborn. 3rd ed. Philadelphia, PA: WB Saunders Co; 1995:730-766
9. Anderson SG, Gilbert GL Neonatal gram-negative meningitis: a 10-year review, with reference to outcome and relapse of infection. J Pediatr Child Health 1990; 26:212-216
10. Kaaresen PI, Flaegstad T Prognostic factors in childhood bacterial meningitis. Acta Paediatr 1995; 84:873-878 [Medline]
11. Kornelisse RF, Westerbeek CM, Spoor AB, Pneumococcal meningitis in children: prognostic indicators and outcome. Clin Infec Dis 1995; 21:1390-1397 [Medline]
12. Pikis A, Kavaliotis J, Tsikoulas J, Andrianopolous P, Venzon D, Manios S Long-term sequelae of pneumococcal meningitis in children. Clin Pediatr (Phila) 1996; 35:72-78
13. Madagme ET, Havens PL, Bresnahan JM, Babel KL, Splaingard ML Survival and functional outcome of children requiring mechanical ventilation during therapy for acute bacterial meningitis. Crit Care Med 1995; 23:1279-1283 [CrossRef][Medline]
14. Grimwood K, Nolan TM, Bond L, Anderson VA, Catroppa C, Keir EH Risk factors for adverse outcomes of bacterial meningitis. J Pediatr Child Health 1996; 32:457-462
15. Rosenberg NM, Meert K, Marino D, De Baker K Seizures associated with meningitis. Pediatr Emerg Care 1992; 8:67-69 [Medline]
16. Valmari P, Makela M, Kataja M, Peltola H Multivariate prognostication in bacterial meningitis of childhood. Scand J Infect Dis 1987; 19:29-34 [Medline]
17. American Academy of Pediatrics, American College of Obstetricians and Gynecologists. Relationship between perinatal factors and neurologic outcome. In: Poland RL, Freeman RK, eds. Guidelines for Perinatal Care. 3rd ed. Elk Grove Village, IL: American Academy of Pediatrics; 1992:221-224
18. Ekert P, Perlman M, Steinlin M, Hao Y Predicting the outcome of postasphyxial hypoxic-ischemic encephalopathy within 4 hours of birth. J Pediatr 1997; 131:613-617 [CrossRef][Medline]
19. Zhang H, Singer B. Recursive Partitioning in the Health Sciences. 1st ed. New York, NY: Springer-Verlag; 1999:7-19
20. Wyatt JC, Altman D Prognostic models: clinically useful or quickly forgotten? Br Med J 1995; 311:1539-1540 [Free Full Text]
21. Richardson D, Tarnow-Mordo WO, Lee SK Risk adjustment for quality improvement. Pediatrics 1999; 103:255-265
22. Hauck W, Samlalsingh-Parker J, Glibetic M, Deregulation of cyclooxygenase and nitric oxide synthase gene expression in the inflammatory cascade triggered by experimental group B streptococcal meningitis in the newborn brain and cerebral microvessels. Semin Perinatol 1999; 23:250-260 [CrossRef][Medline]
23. Boje KM Inhibition of nitric oxide synthase partially attenuates alterations in the blood-cerebrospinal fluid barrier during experimental meningitis in the rat. Eur J Pharmacol 1995; 272:297-300 [CrossRef][Medline]
24. Leib SL, Kim YS, Black SM, Tureen JH, Tauber MG Inducable nitric oxide synthase and the effect of aminoguadine in experimental neonatal meningitis. J Infect Dis 1998; 177:692-700 [Medline]
25. Leib SL, Kim YS, Ferriero DM, Tauber MG Neuroprotective effect of excitatory amino acid antagonist kynurenic acid in experimental bacterial meningitis. J Infect Dis 1996; 173:166-171 [Medline]
26. Pfister HW, Scheld WM Brain injury in bacterial meningitis: therapeutic implications. Curr Opin Neurol 1997; 10:254-259 [Medline]