Clinical predictors of bacterial meningitis in infants and young children in The Gambia


Dr Martin Weber, WHO/CAH, 1211 Geneva 27, Switzerland. Fax: +41 22 791 4853; E-mail:


Background  Bacterial meningitis is an important cause of childhood morbidity and mortality world-wide. In the developing world, where the burden of acute meningitis and its long-term sequelae are especially high, staff with limited training at primary health care facilities must be able to recognize the symptoms and signs of meningitis, so that suspected cases can be referred urgently to hospitals.

Methods  Children who presented with possible invasive bacterial infection to health facilities in The Gambia, West Africa, between 1993 and 1995 were investigated in a standardized manner and clinical findings were documented. Bacterial meningitis was defined as the growth of bacteria from the cerebrospinal fluid. Clinical findings were compared between cases of meningitis and other children.

Results  Of 2097 children between 2 months and 3 years of age investigated, 51 had a confirmed diagnosis of bacterial meningitis. In multivariate analysis using a model adjusting for age but not including respiratory signs, the variables associated independently with meningitis were appearance of being very sick (odds ratio for meningitis vs. no meningitis or no lumbar puncture performed (OR) 4.1, 95% CI 1.5–11.1), being lethargic or unconscious (OR 5.2, 95% CI 2.1–13), a stiff neck (OR 29.3, 95% CI 12.2–70.3), a bulging fontanel (OR 3.2, 95% CI 1.2–8.5) and reduced feeding as a prompted complaint (OR 2.9, 95% CI 1.3–6.7). A combination model of a history of convulsions, or being lethargic or unconscious, or having a stiff neck, as used in the WHO–Integrated Management of Childhood Illness (IMCI) guidelines, had a sensitivity of 98% and a specificity of 72% to predict meningitis.

Conclusions  A combination of a limited number of signs is sufficient to predict meningitis with high sensitivity, without a large number of children who do not have meningitis being unnecessarily referred.


Acute bacterial meningitis is one of the most severe infectious diseases in childhood. The global burden of the disease is high. Apart from epidemics, at least 1.2 million cases of meningitis are estimated to occur every year with 135 000 deaths. (Tikhomirov et al. 1997) About half of the cases and deaths occur in Africa (Tikhomirov et al. 1997). Endemic acute bacterial meningitis is primarily a disease of infants and younger children. It is caused by a variety of microorganisms but, beyond the neonatal period, the most important ones are Streptococcus pneumoniae and Haemophilus influenzae (Molyneux et al. 1998; Lehmann et al. 1999; Palmer et al. 1999).

Rapid diagnosis of bacterial meningitis is essential to avoid a poor outcome (Goetghebuer et al. 2000). This is difficult to achieve in industrialized countries with many physicians, and the problem is even larger in most of the developing world where primary health care (PHC) staff with limited training are the first contact points for patients with meningitis and must decide which patients are referred on to district hospitals. To improve case management in such settings, WHO has developed treatment guidelines which identify sick children in need of referral. These algorithms form part of the Integrated Management of Childhood Illness (IMCI) (Tulloch 1999). The algorithms have to be both sensitive and specific. If the sensitivity of the referral criteria is too low, a potentially life-threatening illness, such as meningitis, will be missed. However, if the specificity is too low, the system will be overburdened with unnecessary referrals, and parents will have to bear the time and expense of referral to a hospital that is frequently far from home. Thus, a careful balance between sensitivity and specificity needs to be found.

We took the opportunity of a vaccination trial against H.influenzae type b in The Gambia (Mulholland et al. 1997) which examined children with possible bacterial infection, to identify potential clinical predictors of meningitis, and to validate the usefulness of existing guidelines.


Study population

The study was conducted in The Gambia, West Africa, between March 1993 and December 1995. Patients eligible for the study were those enrolled in a trial of a vaccine against H. influenzae type b (Mulholland et al. 1997). Any children recruited to the trial who presented with suspected invasive bacterial disease to either the Royal Victoria Hospital (RVH), Banjul, the only tertiary paediatric referral hospital in the Gambia, or to the hospital of the Medical Research Council (MRC) in Fajara, were evaluated by the study paediatricians. Both hospitals provide ambulatory outpatient facilities with free access for patients. In addition, neighbouring health centres and private clinics were encouraged to refer sick children to these facilities. This was intended to make it possible to investigate as many patients as possible for the purpose of the vaccine trial. It is estimated that about 83% of the children born in the study area were recruited into the trial (Adegbola et al. 1999). Because children received their first dose of vaccination with the study vaccine at or after 2 months of age, only children from this age onwards were investigated. At the end of the study the oldest child presenting as a possible case of invasive bacterial disease was 33 months of age.

Documentation and case definition

A standardized questionnaire, which documented symptoms and signs and the results of laboratory investigations, was completed prospectively for each child. The history of illness was initially obtained from complaints volunteered by the mother. This was then complemented by specific questions from the investigator. The physical examination of the child documented the general status of the child, signs of meningeal irritation, respiratory signs, anthropometric measurements and other vital measurements. Standardization of clinical measurements was carried out by the study paediatricians at the beginning of the study, and on several subsequent occasions until its completion. After the physical signs had been recorded, further investigations were undertaken as indicated by the attending physicians. Lumbar puncture was performed if there was a suspicion of meningitis. Children were admitted to hospital for inpatient care as indicated. Treatment was given according to the best medical care available in The Gambia. Children with meningitis were usually treated with a combination therapy of chloramphenicol with either penicillin or ampicillin. The outcome of the admission (survival or death) was documented.

Weight for age and other anthropometric measurements were calculated using the United States National Center for Health Statistics standard as a comparison.

All CSF specimens were analysed at the MRC laboratories, Fajara. Cell count, antigen detection and Gram stain were performed, an aliquot was incubated on culture media and any organism isolated was identified using standard techniques as described previously (Mulholland et al. 1997).

For the purpose of this analysis, children were classified as follows: (a) Lumbar puncture not performed, no suspicion of meningitis; (b) lumbar puncture performed, no meningitis: CSF culture negative, antigen detection negative, and white cell count in CSF < 5 cells/μl; (c) lumbar puncture performed, possible meningitis: CSF culture negative, but either antigen detection positive or white cell count ≥ 5 cells/μl; (d) lumbar puncture performed, definite bacterial meningitis: CSF culture positive.

Statistical methods

The primary analysis was for predictors of meningitis. This involved comparing children who did not undergo lumbar puncture and those with normal CSF findings, to those with definite meningitis (a + b vs. d). A secondary analysis was performed which compared children without meningitis with those with definite meningitis (b vs. d), but this was restricted to the univariate presentation of findings.

For the univariate analysis, most clinical signs and basic laboratory measurements that were continuously distributed were categorized into binary variables using clinically accepted cut-offs. These were then analysed using either a chi-squared test or Fisher's exact test as appropriate. Odds ratios were calculated by logistic regression adjusting for age.

Signs and symptoms that were statistically significant predictors at a level of P < 0.05 were then considered for multivariate analysis. This was performed using unconditional multiple logistic regression to assess their independent utility as predictors of meningitis. Models were built up sequentially for each clinical category of variables (i.e. complaints, respiratory signs, general status, etc.). When the final model was reached, each variable was dropped in turn to assess its effect on the model. A model was then built using the significant and independent predictors of meningitis from each of these categories to obtain a final set of predictors. Different models were compared using changes in log likelihood (i.e. the deviance with the corresponding changes in degrees of freedom which follow a chi-squared distribution). Signs included in the WHO guidelines for the IMCI algorithm were tested independently and compared with the best model. Analyses were performed using the Stata software package.

Verbal consent was obtained from the mother or other guardian, and all explanations were given in the appropriate local language. The study was approved by the Gambian Government/Medical Research Council Ethical Committee.


A total of 2097 children presented with possible invasive bacterial infection and were investigated. Of these, 322 (15.4%) underwent lumbar puncture: 51 (16%) had bacterial meningitis, 233 (72%) had a normal CSF, and 38 (12%) had an abnormal CSF but no bacterial growth in it. The remaining 1775 children (84.6% of the total number) had no lumbar puncture performed as there was no clinical suspicion of meningitis. The organisms cultured from thechildren with meningitis were H. influenzae (25), S.pneumoniae (20), Salmonella spp. (2), Escherichia coli (2), Neisseria meningitidis (1) and Achromobacter spp. (1). Baseline characteristics of children in each of the four groups are shown in Table 1.

Table 1.  Characteristics of the study population
  LP doneNo LP done
n = 1775
Definite meningitis
n = 51
No meningitis
n = 233
Possible meningitis
n = 38
SexF21 (42%)90 (39%)18 (47%)812 (46%)
M30 (58%)143 (61%)20 (53%)964 (54%)
MalariaPositive blood film5 (10%)46 (20%)12 (32%)196 (11%)
Duration of illnessMean4.
Interquartile range(2–5)(2–5)(2–4)(3–5)
Number of deaths
  [case fatality (%)]
14 (27%)23 (10%)2 (5.3%)66 (3.7%)

The overall hospital mortality for all children recruited in the study was 5%. Case fatality in hospital in the different groups is shown in Table 1. There was no significant difference in the mean duration of illness between children with and without meningitis, nor was there a statistically significant association between mortality rates and the duration of illness in those with meningitis.

Univariate analysis

Table 2 shows the odds ratios for meningitis for complaints volunteered spontaneously by mothers/guardians of study children, and those for complaints which were mentioned when the guardian was asked directly about the complaint by the physician. The findings on physical examination are grouped into general signs, signs of meningeal irritation, respiratory signs and other signs. Physical findings in the four groups of children, and the odds ratios for a diagnosis of meningitis are shown in Tables 3 and 4.

Table 2.  Complaints. The first part shows volunteered complaints, which are mentioned by the guardian when asked in an open manner, the second part shows prompted complaints, where the guardian affirms that the problem is present when asked directly
ComplaintLP doneOdds ratios
(n = 51)
(n = 233)
(n = 38)
No LP done
(n = 1775)
Definite meningitis
vs. no meningitis
or no LP done
Definite meningitis
vs. no meningitis
LP done
vs. no LP done
  1. Only significant complaints (P < 0.05) in the comparison of ‘Definite meningitis vs. no meningitis or no LP done’ are included in the list. Odds ratios are adjusted for age (95% CI in brackets).

Volunteered complaints
Cough18 (35%)101 (44%)16 (42%)1424 (80%)0.18 (0.1–0.32)0.68 (0.23–0.1.29)0.18 (0.14–0.23)
Vomiting16 (32%) 48 (21%) 3 (8%) 337 (19%)1.85 (1.02–3.4)1.67 (0.85–3.28)1.08 (0.80–1.45)
Chest pain 3 (6%) 23 (10%) 4 (10%) 448 (25%)0.2 (0.06–0.64)0.49 (0.14–1.73)0.3 (0.2–0.44)
Irritability or
 excessive crying
 9 (17%) 44 (19%) 8 (21%)  25 (1%)4.47 (2.06–9.7)0.81 (0.36–1.81)14.0 (8.57–22.8)
Convulsions13 (25%) 58 (25%)16 (42%) 13 (1%)10.3 (5.2–20.6)1.2 (0.58–2.5)75.4 (39.6–145)
Bulging fontanel 7 (14%)  4 (2%) 1 (3%)  2 (0%)42.8 (13.5–135)8.2 (2.28–29.5)29.1 (6.41–132)
Prompted complaints
Cough32 (63%)179 (77%)27 (71%)1673 (94%)0.13 (0.07–0.24)0.43 (0.22–0.84)0.15 (0.11–0.22)
Chest pain19 (37%)127 (55%)20 (53%)1485 (84%)0.14 (0.08–0.26)0.46 (0.24–0.86)0.22 (0.16–0.27)
Difficulty breathing24 (47%)112 (48%)19 (50%)1303 (74%)0.37 (0.21–0.65)0.89 (0.49–1.65)0.33 (0.26–0.42)
Reduced feeding34 (68%) 88 (38%)14 (37%) 337 (19%)8.0 (4.4–14.6)3.54 (1.85–6.78)3.34 (2.6–4.3)
Not drinking11 (22%) 28 (12%) 9 (24%)  61 (3%)5.25 (2.6–10.6)2.2 (0.99–4.83)4.8 (3.2–7.2)
Convulsions24 (48%) 78 (33%)20 (53%)  38 (2%)14.9 (8.2–26.8)2.18 (1.14–4.17)34.7 (22.7–53.1)
Irritable42 (82%)154 (66%)24 (63%) 755 (43%)4.8 (2.3–10.0)2.12 (0.96–4.6)2.65 (2.05–3.43)
Abnormally sleepy22 (43%) 85 (37%)11 (29%) 222 (13%)4.31 (2.43–7.63)1.51(0.80–2.86)4.29 (3.27–5.6)
Table 3.  Vital signs as predictors of meningitis. Odds ratios are adjusted for age (95% CI in brackets)
SignCategoriesLP done Odds ratios
n = 51
n = 233
n = 38
No LP done
n = 1775
Definite meningitis
vs. no meningitis
or no LP done
Definite meningitis
vs. no meningitis
LP done vs.
no LP done
≥ 38 °C42 (86%)180 (78%)33 (87%)1451 (82%)1.53 (0.68–3.47)1.83 (0.77–4.38)0.94 (0.69–1.20)
Respiratory rateMean62.657.655.057.6   
≥ 40 for
≥ 1 year,
≥ 50 for
2–11 months
39 (76%)28 (74%)173 (75%)1407 (79%)1.06 (0.55–2.04)1.21 (0.59–02.48)0.89 (0.68–1.18)
Oxygen saturationMedian98%98%97.5%97%   
Interquartile range96, 10096 10094, 99.595, 99   
< 90%3 (7%)12 (6%)1 (3%)69 (5%)1.22 (0.36–4.06)1.12 (0.30–4.18)1.07 (0.60–1.88)
Weight for ageMean z-score−1.28−1.26−0.97−1.32   
< −2 SD13 (28%)66 (30%)6 (17%)507 (31%)1.21 (0.62–2.35)1.02 (0.50–2.08)1.03 (0.78–1.37)
Table 4.  Physical findings as predictors of meningitis
VariableCategory comparedLP doneOdds ratios
n = 51
n = 233
n = 38
No LP done
n = 1775
Definite meningitis
vs. no meningitis
or no LP done
Definite meningitis
vs. no meningitis
LP done vs.
no LP done
  1. Only significant complaints (P < 0.05) in the comparison of ‘Definite meningitis vs. no meningitis or no LP done’ are included in the list. Odds ratios are adjusted for age (95% CI in brackets).

General Status
AppearanceVery sick45 (90%)112 (48%)21 (55%) 529 (30%)17.0 (7.18–40.2)8.71 (3.54–21.4) 3.1 (2.4–3.95)
ArousalUnresponsive11 (22%) 19 (8%)21 (53%)  21 (1%)13.7 (6.4–29.1)3.4 (1.56–8.5)12.7 (7.11–22.5)
 Lethargic or  unresponsive43 (84%)115 (50%)21 (55%) 367 (21%)17.0 (7.93–37)6.6 (2.92–14.9) 4.91 (3.8–6.32)
MovementWith pain or none14 (28%) 26 (11%) 8 ( 21%)  24 (1%)15.9 (7.92–31.9)3.77 (1.73–8.24)15.5 (9.1–26.4)
Quality of cryAbnormal or none43 (84%)111 (48%)21 (55%) 434 (25%)13.6 (6.34–29.2)6.14 (2.75–13.70) 3.61 (2.82–4.63)
RestlessnessIrritable or restless32 (63%) 93 (40%)19 (50%) 147 (8%)11.23 (6.24–20.2)2.49 (1.33–4.67) 8.69 (6.56–11.5)
Ability to feedNot able or not  interested31 (62%) 69 (30%)13 (34% 149 (8%)12.47 (6.95–22.3)3.94 (2.08–7.47) 6.06 (4.54–8.10)
Abnormally sleepyYes30 (60%) 77 (33%)15 (40%) 197 (11%) 9.38 (5.26–16.71)3.33 (1.75–6.33) 5.20 (3.94–6.85)
Meningeal irritation
FontanelBulging26 (51%) 29 (13%) 6 ( 16%)  11 (1%)41.5 (21.6–79.5)7.00 (3.5–14.0)31.6 (16.4–61.2)
tonusHypertonic vs. normal and  hypotonic30 (60%) 43 (19%) 8 ( 21%)  12 (1%)42.8 (22.6–80.8)6.01 (3.1–11.6)43.5 (23.3–81.3)
Stiff neckPresent39 (78%) 57 (25%)14 (37%)  14 (1%)77.3 (38.5–155)9.79 (4.79–20.0)59.3 (33.3–106)
Respiratory signs
Cough heard on  examinationYes18 (35%)144 (62%)21 (55%)1528 (86%) 0.11 (0.06–0.20)0.31 (0.16–0.58) 0.21 (0.16–0.28)
nasal flaringPresent18 (35%)109 (47%)17 (45%) 536 (30%) 0.27 (0.15–0.49)0.60 (0.32–1.13) 0.35 (0.28–0.45)
intercostal indrawingModerate or severe 6 ( 12%) 47 (20%) 6 ( 16%) 635 (36%) 0.27 (0.12–0.64)0.51 (0.21–1.28) 0.42 (0.31–0.56)
lower chest wall  indrawingModerate or severe 7 ( 14%) 66 (28%)10 (26%) 836 (47%) 0.19 (0.09–0.43)0.37 (0.16–0.88) 0.39 (0.30–0.51)
work of breathingModerately or  severely increased 9 ( 18%) 61 (27%)12 (32%) 744 (42%) 0.33 (0.16–0.67)0.56 (0.26–1.23) 0.48 (0.37–0.63)
Shape of thoraxIncreased volume 3 ( 6%) 30 (13%) 3 ( 8%) 296 (17%) 0.28 (0.08–0.91)0.39 (0.11–1.34) 0.57 (0.40–0.83)
CrepitationsPresent 9 ( 18%) 63 (27%)11 (29%) 728 (41%) 0.17 (0.08–0.36)0.56 (0.26–1.22) 0.24 (0.18–0.31)
WheezePresent 2 ( 4%) 25 (11%) 5 ( 13%) 432 (24%) 0.11 (0.03–0.46)0.29 (0.07–1.29) 0.29 (0.20–0.40)
RhonchiPresent 1 ( 2%) 37 (16%) 5 ( 13%)1172 (66%) 0.04 (0.005–0.27)0.096 (0.012–0.72) 0.28 (0.2–0.4)
Wheeze or rhonchiPresent 2 ( 4%) 44 (19%) 7 ( 18%) 677 (38%) 0.06 (0.02–0.76)0.16 (0.04–0.68) 0.29 (0.11–0.40)
Decreased breath  soundsPresent14 (28%) 76 (33%)17 (45%) 860 (48%) 0.44 (0.24–0.83)0.73 (0.37–1.44) 0.53 (0.41–0.68)
Other signs
JaundicePresent 3 ( 6%)  7 ( 3%) 0  13 (1%) 5.26 (1.49–18.54)2.12 (0.53–8.59) 3.91 (1.68–9.10)
pallor conjunctivaModerate or severe 7 ( 14%) 21 (9%) 7 ( 18%)  98 (6%) 2.83 (1.23–6.47)1.81 (0.71–4.59) 2.31 (1.52–3.49)
pallor tongueModerate or severe 8 ( 16%) 21 (9%) 7 ( 18%) 100 (6%) 3.09 (1.41–6.77)2.09 (0.86–5.13) 2.25 (1.49–3.37)
pallor palmsModerate or severe10 (20%) 27 (12%) 8 ( 21%) 129 (7%) 3.17 (1.55–6.50)1.99 (0.89–4.48) 2.25 (1.55–3.25)

Multiple logistic regression

For the primary analysis (predictors of meningitis in children with meningitis compared with those with negative findings on LP and those who did not have a LP done), regression analysis was performed for the groups of variables to identify independent predictors, adjusting forage. For volunteered complaints, the independent predictors of meningitis were vomiting (OR 3.2, 95% CI 1.6–6.2), convulsions (OR 16, 95% CI 7.6–33.7) and a bulging fontanel (OR 84.3, 95% CI 25.2–282). Independent predictors for meningitis in the category of solicited complaints were reduced feeding (OR 4.5, 95% CI 2.4–8.6), not drinking (OR 2.5, 95% CI 1.01–5.93), convulsions (OR 8.2, 95% CI 4.3–15.8) and irritability (OR 5.4, 95% CI 2.4–12). General clinical signs that were independently significant were sick appearance (OR 5.8, 95% CI 2.3–14.5), arousal (lethargic or none, OR 5.8, 95% CI 2.6–13.3), and restlessness (OR 4.3, 95% CI 1.8–6.5). In the category of meningeal irritation, all three variables had an independent effect. These were stiff neck (OR 23, 95% CI 8.8–60.5), increased tonus (OR 3.1, 95% CI 1.3–7.4) and a bulging fontanel (OR 4.0, 95% CI 1.7–9.2). In the category of respiratory signs, there remained two abnormal respiratory signs as significant negative predictors: the presence of wheezes or rhonchi (OR 0.11, 95% CI 0.03–0.48) and cough heard on examination (OR 0.17, 95% CI 0.1–0.30). In the category of other clinical signs, palmar pallor was the only significant predictor (OR 2.8, 95% CI 1.4–5.9), with jaundice being of borderline significance (OR 3.6, 95% CI 0.96–13.2).

These categories were combined in two models, one with and one without respiratory signs, as respiratory signs were mainly important by their absence. In the model without respiratory signs, significant and independent predictors were very sick appearance (OR 4.1, 95% CI 1.5–11.1), being lethargic or unconscious (OR 5.2, 95% CI 2.1–13), a stiff neck (OR 29.3, 95% CI 12.2–70.3), a bulging fontanel (OR 3.2, 95% CI 1.2–8.5) and reduced feeding as a prompted complaint (OR 2.9, 95% CI 1.3–6.7). When respiratory signs were included, the presence of wheezes or rhonchi entered as a negative predictor (OR 0.12, 95% CI 0.023–0.63), the odds ratios of the other variables changed only slightly and remained significant.

Sensitivity and specificity of different variables and models

Table 5 shows the sensitivity and specificity of predicting meningitis of the above significant variables, and of the combination models. In addition, four signs for meningitis in the current WHO–IMCI algorithm – inability to feed, convulsions, being lethargic or unconscious, and having a stiff neck – were analysed. Regression analysis was performed for this algorithm as well; the results are shown in Table 5.

Table 5.  Sensitivity and specificity of individual and grouped signs and symptoms which were independently significant predictors of meningitis (definite meningitis vs. no meningitis or no LP done)
predictive value
predictive value
  1. Variables are categorized as described in Tables 2–4.

History volunteered
  Vomiting3181 4 98
  Convulsions259615 98
  Bulging fontanel1410054 98
  Any of the three volunteered   complaints6777 7 99
History prompted
  Reduced feeding6779 7 99
  Not drinking229611 98
  Convulsions479417 99
  Irritability8255 4 99
  Any of the 4 prompted   complaints9842 4100
General status
  Appearance8868 7100
  Arousal8476 8 99
  Restlessness638812 99
  Any of the 3 general status   variables9657 5100
Meningeal irritation
  Stiff neck769635 99
  Increased tonus599735 99
  Bulging fontanel519839 99
  Any sign of meningeal irritation809426 99
Respiratory signs
  Absence of wheezes or rhonchi9832 4100
  Absence of cough during   examination6583 9 99
  Absence of all 2 respiratory signs658510 99
Other signs
  Palmar pallor2092 6 98
  Jaundice 69913 98
  Any of ‘other signs’2092 6 98
Combination models
  Final model (without respiratory   signs)9652 5100
  Final model (with respiratory   signs)9464 6100
  WHO–IMCI algorithm9870 8100
  Reduced WHO–IMCI algorithm9872 8100
  (history of convulsions,   lethargic, or stiff neck)
  Further reduced WHO–IMCI   algorithm (lethargic, or   stiff neck)9674 9100


As meningitis is a life-threatening disease, most diagnostic approaches have aimed at maximizing sensitivity. Atraditional paediatric approach is to perform a lumbar puncture whenever meningitis is included in the differential diagnosis. Accordingly, most publications have concentrated on signs associated with meningitis, while ignoring the interaction of signs and their specificity. Collected prospectively during a vaccine trial, our data provided an opportunity to examine both sensitivity and specificity. For the main analysis, we compared children with proven bacterial meningitis with those without meningitis combined with those in whom no lumbar puncture was performed, as we believed that this was the comparison of most clinical importance. This reflects better the situation in an outpatient department than to restrict the analysis to the groups who had undergone lumbar puncture and are found to have or not to have meningitis. However, our patients had already undergone a pre-screening: only children who were potentially ‘septic’ were included in the work-up. This initial screening had already eliminated a number of children who presented with mild illnesses such as upper respiratory tract infections, mild diarrhoea or skin problems. As such, children would normally not have shown signs which were associated with meningitis, their inclusion would almost certainly have improved further the specificity and predictive values of the characteristics we established as predictors of meningitis. None of the signs used was complex and all predictive symptoms and signs could be elicited by a PHC worker.

Other studies have shown that physical signs vary with age. The diagnosis of meningitis is most difficult in neonates (Ziai & Haggerty 1958), an age group not represented in this study. A bulging fontanel can be seen only as long as it is open during the first year of life. It has been found previously to be associated more commonly with bacterial meningitis than with aseptic meningitis (Valmari et al. 1987; Rosenberg & Bobowski 1988; Hensel et al. 1992; Walsh-Kelly et al. 1992). Similarly, lethargy is most commonly reported as a finding in young children with bacterial meningitis (Valmari et al. 1987; Kirkpatrick et al. 1994) and is again more commonly recorded in bacterial than in aseptic meningitis (Rosenberg & Bobowski 1988; Walsh-Kelly et al. 1992). Neck stiffness is reported more frequently in children than in infants, and more frequently in bacterial than in aseptic meningitis (Walsh-Kelly et al. 1992; Chotpitayasunondh 1994; Kirkpatrick et al. 1994). Thus, if a combination of signs is used which are each more common in a particular age group but together cover a wider age range, the sensitivity of the algorithm is increased. Convulsions were common in meningitis in our series, and also recorded frequently in children with meningitis in studies from Thailand and Malaysia (Choo et al. 1990; Chotpitayasunondh 1994). This is in contrast to series which included adults (Tugwell et al. 1976; Geiseler et al. 1980; Kirkpatrick et al. 1994), or which are dominated by cases of meningococcal meningitis (Andersen et al. 1997).

The current WHO–IMCI algorithm performed well. The IMCI criteria had a specificity of over 70%. However, in this series, this still meant that more than 10 times more children without meningitis would have been referred than children with meningitis. It should be kept in mind, however, that some of these children suffered from other conditions requiring hospital admission, such as severe pneumonia, and needed referral for that reason. However, over-referral due to a low specificity is of real concern where children need to be taken over considerable distances for a diagnosis or to receive inpatient treatment. Compared with other disease-specific guidelines, such as those for pneumonia (Mulholland et al. 1992) or neonatal sepsis (The WHO Young Infants Study Group 1999), this combination of clinical signs performed quite well. Of particular interest is the negative predictive value of 100% in all the combination models: Children without any of the identified signs were unlikely to have meningitis. Ideally, similar studies should be carried out in other settings as well, to evaluate and validate the prospective performance of our prediction models in different populations.

In summary, this study showed how well different signs predicted meningitis. This information will be useful in clinical teaching as it provides an evidence base for the selection of signs and symptoms warranting further investigation. A limited selection of clinical variables, as included in the IMCI guidelines, will help people with limited training to identify children with possible meningitis. These signs should be widely promoted.


We thank members of the nursing staff of both the Royal Victoria Hospital and Medical Research Council Laboratories ward for providing the inpatient care to our study children. We also thank also Drs Palmer, Jallow, Enwere, Adegunloye (RVH) and Drs Corrah and Juwah (MRC) for their contribution to the management of these patients.

The efficacy trial of a H. influenzae type b vaccine in The Gambia, during which the clinical data were collected, was supported by the United States Agency for International Development–Public Health Interagency for Vaccine Development and Health Research, the World Health Organization Programme for the Control of Acute Respiratory Infections, the United Nations Children's Fund (UNICEF), the Children's Vaccine Initiative and the United Nations Development Program.