ARTICLES

Diagnosis of adult tuberculous meningitis by use of clinical and

laboratory features

G E Thwaites, T T H Chau, K Stepniewska, N H Phu, L V Chuong, D X Sinh, N J White, C M Parry, J J Farrar

Summary
Background The diagnosis of tuberculous meningitis is
difficult. Discrimination of cases from those of bacterial
meningitis by clinical features alone is often impossible, and
current laboratory methods remain inadequate or inaccessible
in developing countries. We aimed to create a simple
diagnostic aid for tuberculous meningitis in adults on the basis
of clinical and basic laboratory features.
Methods We compared the clinical and laboratory features on
admission of 251 adults at an infectious disease hospital in
Vietnam who satisfied diagnostic criteria for tuberculous
(n=143) or bacterial (n=108) meningitis. Features
independently predictive of tuberculous meningitis were
modelled by multivariate logistic regression to create a
diagnostic rule, and by a classification-tree method. The
performance of both diagnostic aids was assessed by
resubstitution and prospective test data methods.
Findings Five features were predictive of a diagnosis of
tuberculous meningitis: age, length of history, white-blood-cell
count, total cerebrospinal fluid white-cell count, and
cerebrospinal fluid neutrophil proportion. A diagnostic rule
developed from these features was 97% sensitive and 91%
specific by resubstitution, and 86% sensitive and 79% specific
when applied prospectively to a further 42 adults with
tuberculous meningitis, and 33 with bacterial meningitis. The
corresponding values for the classification tree were 99% and
93% by resubstitution, and 88% and 70% with prospective test
data.
Interpretation This study suggests that simple clinical and
laboratory data can help in the diagnosis of adults with
tuberculous meningitis. Although the usefulness of the
diagnostic rule will vary depending on the prevalence of
tuberculosis and HIV-1 infection, we suggest it be applied to
adults with meningitis and a low cerebrospinal fluid glucose,
particularly in settings with limited microbiological resources.
Lancet 2002; 360: 128792

University of Oxford-Wellcome Trust Clinical Research Unit, Centre

for Tropical Diseases, 190 Ben Ham Tu, Quan 5, Ho Chi Minh City,
Vietnam (G E Thwaites MRCP, K Stepniewska PhD, N J White FRCP,
C M Parry FRCPath, J J Farrar FRCP); Centre for Tropical Diseases,
Nuffield Department of Clinical Medicine, John Radcliffe Hospital,
University of Oxford, Oxford, UK (G E Thwaites, K Stepniewska,
N J White, C M Parry, J J Farrar); Centre for Tropical Diseases, Ho
Chi Minh City, Vietnam (T T H Chau MD, N H Phu MD, L V Chuong MD,
D X Sinh MD)
Correspondence to: Dr G E Thwaites
(e-mail: gthwaites@hcm.vnn.vn)

THE LANCET Vol 360 October 26, 2002 www.thelancet.com

Introduction
The prompt diagnosis and treatment of tuberculous
meningitis saves lives.15 However, about 30% of patients
with tuberculous meningitis die despite anti-tuberculosis
chemotherapy.6 Delays in diagnosis and treatment are
regarded as major contributing factors in the high
mortality reported in many recent series.710 The diagnosis
of tuberculous meningitis relies on isolation of
Mycobacterium tuberculosis from the cerebrospinal fluid.
Unfortunately, culture is too slow and insensitive to aid
clinical decision-making. Direct Ziehl-Neelsen staining of
the cerebrospinal fluid for acid-fast bacilli remains the
cornerstone of rapid diagnosis, but this technique lacks
sensitivity.6 Newer diagnostic techniques, such as those
that use PCR, have not been assessed completely,11 and are
not possible in most settings in the developing world where
most cases of tuberculous meningitis are seen.12,13
Consequently, the decision to treat a patient for
tuberculous meningitis is frequently empirical, irrespective
of the diagnostic facilities available to clinicians. Until new,
affordable, sensitive, and specific diagnostic assays become
available, clinicians must depend on the discriminative
clinical and laboratory features of the disease for successful
diagnosis and treatment.
The presenting clinical features of tuberculous
meningitis in adults are similar to those of many meningoencephalitides, which result in frequent diagnostic
confusion. Delays in starting appropriate antibiotics for
tuberculous meningitis or pyogenic meningitis worsens
prognosis, yet physicians are often reluctant to start
months of antituberculosis treatment without firm
evidence. Diagnostic uncertainty arises commonly in
patients who present with a few days of headache, fever,
and neck-stiffness; undefined treatment in the community;
a low concentration of glucose in cerebrospinal fluid
(<50% of that in blood), and neutrophils and lymphocytes
in the cerebrospinal fluid.
Multivariate logistic regression has been used to model
the clinical predictors of tuberculous meningitis in
232 children.14 Five presenting clinical features were found
to be independently predictive of tuberculous meningitis:
prodromal stage 7 days or longer, optical atrophy on
fundal examination, focal deficit, abnormal movements,
and cerebrospinal fluid leucocytes comprising less than
50% polymorphs. The researchers developed a simple
diagnostic rule. Sensitivity was 984% and specificity
435% when one or more predictor variables were present,
and specificity was 983% and sensitivity 545% if three or
more were present. However, there are several problems
with this rule. First, the performance of the rule varies
according to the prevalence of tuberculosis. Second, the
rule can be difficult to apply: if antituberculosis
chemotherapy is started on the basis of one diagnostic
variable, more than 50% will be wrongly treated; if
treatment is only begun in the presence of three variables,
nearly half will not receive appropriate treatment. In the

1287

For personal use. Only reproduce with permission from The Lancet Publishing Group.

ARTICLES

Methods

Diagnostic criteria for tuberculous and bacterial

meningitis
Tuberculous meningitis
Mycobacterium tuberculosis
isolated from cerebrospinal
fluid

Bacterial meningitis
Pathogenic bacteria isolated
from cerebrospinal fluid

Or

Or

Clinical meningitis with

Clinical meningitis with all of
negative gram and India
the following:
Lymphocytes and neutrophils
ink stains, plus sterile
bacterial and fungal cultures,
in cerebrospinal fluid
Low concentration of glucose
plus one or more of the
following:
in cerebrospinal fluid (<50%
Cranial CT scan consistent
of that in blood)
with tuberculous meningitis Sterile blood and
(hydrocephalus, oedema,
cerebrospinal fluid cultures
Full recovery (without
basal meningeal
enhancement)
antituberculosis
Chest radiograph consistent
chemotherapy) 3 months
with active pulmonary
after admission
tuberculosis
Good response to
antituberculosis
chemotherapy

first scenario, patients are unnecessarily exposed to the

risks of drug toxicity. In the second, many patients could
die. Finally, a rule developed in children might not be
applicable to adults.
Nevertheless, there are strong arguments for developing
a simple diagnostic algorithm for tuberculous meningitis in
areas of high tuberculosis prevalence. First, tuberculous
meningitis tends to be commonest in areas with the least
clinical and laboratory resources. Second, a diagnostic rule
developed and used in high tuberculosis prevalence areas is
likely to perform consistently. And third, early diagnosis
and treatment improves outcome.15 We aimed to develop
such a diagnostic algorithm on the basis of clinical and
laboratory findings of adult patients in Vietnam.

Patients
The adults described in this study were admitted to the
Clinical Research Unit at the Centre for Tropical
Diseases, Ho Chi Minh City, Vietnam. The Centre for
Tropical Diseases is a 500-bed infectious diseases hospital
that serves the local community, and acts as the tertiary
referral centre for infectious disease for the whole of
southern Vietnam. The hospital treats about 30 000
inpatients and 52 000 outpatients each year. The Clinical
Research Unit is a 15-bed ward for patients with centralnervous-system infection, severe malaria, sepsis, and renal
failure. Clinical data were recorded prospectively from all
adults (>15 years old) admitted with suspected centralnervous-system infection to the ward between 1997 and
2000.
Procedures
After providing informed consent, each patient underwent
standard history taking, examination, and baseline
investigations (including lumbar puncture, chest
radiography, and computed tomography of the head when
possible). Patients judged to be at risk of HIV-1 infection
were tested for antibodies to this virus. Each sample of
cerebrospinal fluid was centrifuged, and a portion of
deposit was examined by microscopy with gram, ZiehlNeelsen, and India ink stains. The remaining deposit was
cultured on blood and chocolate agar, and LwensteinJensen media.
After 48 h, a second lumbar puncture was done as part
of routine hospital management in most patients to assess
response to treatment. Those with culture-proven or
suspected bacterial meningitis received 10 days of
intravenous ceftriaxone (Roche, Basel, Switzerland; 2 g
twice per day). Those with suspected tuberculous
meningitis received four drugs (streptomycin, isoniazid,
rifampicin, and pyrazinamide) for 3 months, followed by
three drugs (isoniazid, rifampicin, and pyrazinamide) for
6 months. Clinical progress and response to treatment
were recorded in individual study notes. The Hospital
Scientific and Ethical Committee approved the study.

Tuberculous meningitis

Age (years)
Male sex
Duration of illness (days)
Duration of fever (days)
Duration of headache (days)
Neck stiffness
Presence of coma before admission
Glasgow coma score (/15)
Pulse (/min)
Systolic BP (mm Hg)
Diastolic BP (mm Hg)
Hemiplegia
Cranial nerve palsies
Packed-cell volume (%)
WBC count (103/mL)
% neutrophils
Blood sodium (mmol/L)
CSF opening pressure (cm H20)
Clear CSF appearance
CSF total WCC (103/mL)
CSF % neutrophils
CSF % lymphocytes
CSF/blood glucose
CSF chloride (mmol/L)
CSF protein (g/dL)
CSF lactate (mmol/L)

Bacterial meningitis

Median (90% range)

Number of patients

Median (90% range)

Number of patients

34 (1664)

12 (434)
10 (230)
10 (130)

13 (715)
90 (60120)
120 (90150)
70 (5090)

40 (3048)
9800 (500016 200)
80 (6089)
135 (122143)
23 (944)

300 (701090)
37 (184)
64 (1699)
028 (011052)
85120
191 (80490)
54 (1598)

143
91 (64%)
142
139
136
120 (91%)
49 (36%)
143
143
143
143
11 (8%)
32/143 (22%)
139
137
130
98
133
81/141 (57%)
143
142
142
139
136
141
102

41 (1770)

3 (111)
3 (111)
3 (111)

14 (615)
92 (72120)
120 (90150)
70 (5090)

42 (3048)
15 250 (747031 500)
86 (7095)
138 (125148)
24 (7540)

2583 (38220 000)

90 (6099)
10 (140)
020 (003046)
97121
270 (89730)
94 (21197)

108
84 (78%)
107
106
106
81 (84%)
53 (50%)
107
108
108
108
4 (4%)
9/107 (8%)
104
107
107
102
60
2/107 (2%)
108
108
108
101
67
107
92

00076
00160
00001
00001
00001
00910
00271
00807
00023
07588
02267
02820
00031
01255
00001
00001
00009
09290
00001
00001
00001
00001
00001
00120
00001
00001

BP=blood pressure; WBC=white blood cell; CSF=cerebrospinal fluid; WCC=white-cell count.

Table 1: Univariate analysis comparing admission variables between patients with tuberculous and bacterial meningitis

1288

THE LANCET Vol 360 October 26, 2002 www.thelancet.com

For personal use. Only reproduce with permission from The Lancet Publishing Group.

ARTICLES

Tuberculous meningitis

CSF opening pressure (cm H2O)

Clear CSF appearance
CSF total WCC (103/mL)
CSF % neutrophils
CSF % lymphocytes
CSF/blood glucose
CSF chloride (mmol/L)
CSF protein (g/dL)
CSF lactate (mmol/L)

Bacterial meningitis

Median (90% range)

Number of patients

Median (90% range)

Number of patients

27 (844)

470 (1732575)
47 (487)
53 (1396)
027 (01305)
102 (81116)
190 (98680)
64 (2911)

43
21 (49%)
51
51
51
47
48
49
36

14 (427)

760 (8012 000)

74 (897)
26 (390)
043 (011069)
115 (105126)
110 (37580)
48 (18104)

58
23 (23%)
106
106
106
106
64
104
77

00027
00001
00484
00001
00001
00001
00001
00001
00105

CSF=cerebrospinal fluid; WCC=white-cell count.

Table 2: Second cerebrospinal fluid analysis taken after 4872 h of parenteral ceftriaxone

Diagnostic criteria
Diagnostic criteria for tuberculous meningitis and bacterial
meningitis were applied to all patients admitted to the
Clinical Research Unit with meningitis. The criteria are
presented in the panel. There is no acceptable gold
standard diagnostic test for tuberculous meningitis.
Culture and Ziehl-Neelsen staining of the cerebrospinal
fluid are specific, but too insensitive to be used as the sole
criteria for diagnosis. We therefore developed clinical
criteria similar to those used by other researchers.14,15 The
criteria for diagnosis of bacterial meningitis included those
for partly treated bacterial meningitis (ie, sterile
cerebrospinal fluid cultures), since this group of patients is
commonly misdiagnosed with tuberculous meningitis.
Tuberculous meningitis was excluded in these patients if
they fully recovered, without antituberculosis chemotherapy, 3 months after admission. Untreated tuberculous
meningitis would almost always be fatal within this time.
Statistical analysis
The clinical and laboratory features of those fulfilling the
diagnostic criteria for tuberculous meningitis and bacterial
meningitis were compared. These features comprised
26 clinical and laboratory variables at admission, including
characteristics of the cerebrospinal fluid sample.
Additionally, data from a second cerebrospinal fluid
sample were collected from all patients who received 48 h
of intravenous ceftriaxone. Diagnostic uncertainty on
admission often leads clinicians in our hospital to use a
trial of ceftriaxone, and reconsider the diagnosis after a
second cerebrospinal fluid examination. The purpose of
collecting these additional data was to compare the
cerebrospinal fluid variables in both groups, and to
develop a second diagnostic rule for patients managed
by this approach. The relative change in each
cerebrospinal fluid variable was calculated by (a[CSF
2]a[CSF 1])/a(CSF 1), where a is the specified variable,
and CSF 1 and CSF 2 are the first and second
cerebrospinal fluid samples, respectively. All those given
immediate antituberculosis chemotherapy were excluded
from this analysis. The Kruskal-Wallis test was used to

compare continuous variables, and the 2 test (or Fishers

exact test for small proportions) was used for categorical
variables.
Three diagnostic aids were developed by means of two
statistical approaches: classification trees (classification
and regression tree, CART) and logistic regression. The
classification trees were developed by consideration of all
the variables separately. The range of each variable was
divided into two groups to obtain the best separation
between patients with bacterial meningitis and those with
tuberculous meningitis. The division corresponding to the
best separation was selected. The resulting subsets of cases
were then partitioned independently in turn. The process
was done recursively, until a stopping condition was
satisfied. Node deviance, which measures node
heterogeneity, was set to 01 to stop the tree-growing
process. Subsets smaller than 10 were not partitioned
further.
Logistic regression was used to model the probability of
having tuberculous meningitis. A stepwise forward variable
selection procedure was used to find independent
predictors of tuberculous meningitis with p-to-enter of
005 or less, and p-to-remove of 0055 or more. Once the
final model was constructed, the continuous variables in
the model were dichotomised by use of cutoffs from the
univariate classification trees, and the model was refitted.
Rounded -coefficients from the model with dichotomised
variables were used to define a diagnostic index for each of
the clinical variables. A receiver operator characteristic
(ROC) curve analysis was selected to find an optimum
cutoff for the combined diagnostic indices. ROC analysis
was done on the original dataset, and the completed rule
(diagnostic index with cutoff) was applied to the test data.
The three diagnostic aids were assessed by
resubstitution and prospective test data methods. The
sensitivity and specificity were calculated and compared
with the study diagnostic criteria. The resubstitution
method used the original data set. The test data method
used data recorded from a further 75 patients enrolled in
the same manner and subject to the same diagnostic
criteria. All analyses were done with STATA and Splus.

Tuberculous meningitis

CSF opening pressure

CSF WCC
CSF % neutrophils
CSF % lymphocytes
CSF glucose/
blood ratio
CSF lactate
CSF protein
CSF chloride

Bacterial meningitis

Median (90% range)

Number of patients

Median (90% range)

Number of patients

8% (75 to 267)
13% (68 to 964)
1% (45 to 500)
5% (54 to 145)

36
51
50
50

41% (83 to 56)

75% (97 to 918)
15% (88 to 23)
144% (88 to 155)

51
106
106
106

00001
00001
00001
00001

6% (66 to 92)
0% (55 to 314)
2% (37 to 275)
4% (14 to 8)

45
27
49
48

100% (61 to 833)

48% (81 to 114)
54% (90 to 51)
4% (8 to 21)

99
70
103
63

00001
00001
00001
00001

CSF=cerebrospinal fluid; WCC=white-cell count.

Table 3: Change in cerebrospinal fluid variables after 4872 h of parenteral ceftriaxone

THE LANCET Vol 360 October 26, 2002 www.thelancet.com

1289

For personal use. Only reproduce with permission from The Lancet Publishing Group.

ARTICLES

Age
History of illness
Blood WCC
CSF WCC
CSF % neutrophils

0069
0243
00003
0002
0075

Odds ratio (95% CI)

0933 (08850984)
1275 (11321437)
09997 (0999509999)
0998 (09960999)
0928 (08840973)

00098
00001
00007
00002
00020

WCC=white-cell count; CSF=cerebrospinal fluid. Values of five variables found to

be associated independently with tuberculous meningitis on logistic regression
analysis. All variables are continuous.

Table 4: Multivariate logistic regression analysis of admission

data

Role of the funding source

The study sponsor had no role in the study design; data
collection, analysis, or interpretation; or in the writing of the
report and the decision to submit the paper for publication.

Results
357 adults were admitted to the Centre for Tropical
Diseases with meningitis between 1997 and 2000. 251
satisfied the diagnostic criteria for inclusion in this study:
143 with tuberculous meningitis, and 108 with bacterial
meningitis. 66 of the 251 adults were tested for antibodies
to HIV-1: eight were positive (seven with tuberculous
meningitis, one with bacterial meningitis).
163 of 357 adults received antituberculosis chemotherapy
for suspected tuberculous meningitis: M tuberculosis was
isolated from the cerebrospinal fluid of 37 patients, and 106
were defined as having clinical tuberculous meningitis.
Supportive radiological evidence of tuberculous meningitis
was present in 85 of 106 adults with clinical tuberculous
meningitis. 20 patients treated for tuberculous meningitis
were excluded because they did not meet the study criteria
for the diagnosis of tuberculous meningitis, and all died
shortly after the start of antituberculosis chemotherapy. No
significant differences (p<005) were found between the
admission, clinical, and laboratory parameters of those with
culture-confirmed and clinical tuberculous meningitis.
194 of the 357 adults admitted over the same period were
not treated for tuberculous meningitis: 108 satisfied the
study criteria for a diagnosis of bacterial meningitis.
A bacterial pathogen was isolated from the cerebrospinal
fluid of 68 adults: 24 had Streptococcus pneumoniae,
31 Streptococcus suis, four Neisseria meningitidis, four
Haemophilus influenzae, four Klebsiella spp, and one
Escherichia coli. A further 40 adults satisfied the diagnostic
criteria for bacterial meningitis (presumed partly treated).
86 adults were excluded from the study for the following
reasons: 61 had a normal cerebrospinal fluid glucose
Diagnostic index
Age (years)
36
<36

2
0

Blood WCC (103/mL)

15 000
<15 000

4
0

History of illness (days)

6
<6

5
0

CSF total WCC (103/mL)

900
<900

3
0

CSF % neutrophils
75
<75

4
0

WCC=white-cell count; CSF=cerebrospinal fluid.

Table 5: Weighted diagnostic index scores for dichotomised

clinical variables used for admission diagnostic rule

1290

Area under ROC curve=09901

100

075
Sensitivity

-coefficient

050

025

0
0

025

050
1specificity

075

100

Figure 1: Receiver-operator characteristic (ROC) curve for

prognostic index derived from logistic regression model

concentration (>50% of that in blood), 15 died within

3 months, and 11 were lost to follow-up.
Admission variables are shown in table 1. Excluding the
92 patients who received immediate antituberculosis
chemotherapy, the results from a second lumbar puncture
after 48 h of ceftriaxone were available in 157 of 251
patients: 51 with tuberculous meningitis and 106 with
bacterial meningitis (table 2). Two of 251 patients with
culture-confirmed bacterial meningitis died before a second
sample could be taken. After 48 h of ceftriaxone, significant
differences were apparent between the two groups (table 3).
The cerebrospinal fluid variables changed little in those with
tuberculous meningitis. However, in those with bacterial
meningitis, cerebrospinal fluid pressure, white-cell count,
protein, and lactate fell, whereas the cerebrospinal
fluid:blood glucose ratio rose.
Multivariate analyses were done to construct a diagnostic
rule. Glasgow coma score, diastolic and systolic blood
pressure, presence of hemiplegia, and packed-cell volume at
admission were excluded due to the non-significant results
by univariate analysis (table 1). Blood sodium,
cerebrospinal fluid opening pressure, cerebrospinal fluid
chloride, and cerebrospinal fluid lactate were excluded
because of a large number of missing values. Only the total
duration of combined symptoms before admission was
included for analysis: length of fever, and headache, were
excluded.
Stepwise logistic regression analysis found five variables
independently associated with a diagnosis of tuberculous
meningitis on admission. The final logistic model with these
variables is described in table 4. The formula for diagnostic
index was derived from the final model by dichotomising
the variables and rounding the coefficients in the model.
The index was adjusted to a positive scale for ease of use.
The total diagnostic index (DI) was calculated for each
patient according to the formula:
DI (age)+DI (blood white-cell count)+DI (history of
illness)+DI (cerebrospinal fluid white-cell count)+DI
(cerebrospinal fluid % neutrophils)

The diagnostic index for each of the five variables is given

in table 5. The optimum cutoff for the total diagnostic index
(by which to classify a patient as having tuberculous
meningitis) was found by use of an ROC curve (figure 1).
The three points close to the top left-hand corner of the
curve correspond to cutoffs of 4, 3, or 2. The respective
diagnostic sensitivities and specificities for each point are
97% and 91%, 96% and 91%, and 91% and 97%. Our

THE LANCET Vol 360 October 26, 2002 www.thelancet.com

For personal use. Only reproduce with permission from The Lancet Publishing Group.

ARTICLES

At admission

CSF white-cell count

760103/mL
Yes

History of
illness
7 days

Yes

Yes

No

Test data method

Sensitivity

Sensitivity

Specificity

Specificity

At admission
Classification 99% (134/136) 93% (99/107) 88% (37/42) 70% (23/33)
tree
Diagnostic
91% (123/135) 97% (104/107) 86% (36/42) 79% (26/33)
rule

No

History of
illness
6 days

Resubstitution method

After second lumbar puncture

Classification 93% (39/42)
95% (92/97)
tree

No

57% (8/14)

76% (25/33)

Table 6: Sensitivity and specificity of diagnostic aids

Age 42
years

Blood WCC
10 200103/mL
No
Bacterial
meningitis

Yes

No

Tuberculous
meningitis

Bacterial
meningitis

Yes
Tuberculous
meningitis

After 48 h of broad-spectrum antibiotics

History of illness
8 days
Yes
Blood
neutrophils
80%
Yes

No

No
Rise in CSF/
blood glucose
100%
No

Yes

CSF neutrophils
81%
Yes
Tuberculous
meningitis

No
Bacterial
meningitis

Tuberculous
meningitis

Figure 2: Diagnostic classification trees for use at admission

and after 48 h of broad-spectrum antibiotics
WCC=white-cell count; CSF=cerebrospinal fluid.

suggested diagnostic rule was therefore: if the patient has a

total diagnostic index score of 4 or less, he or she has
tuberculous meningitis, and if the patient has a score of
more than 4, he or she has bacterial meningitis. No
improvement in the diagnostic rule was obtained by use of
data from both admission and the second cerebrospinal
fluid sample. Missing values, which reduced the sample size
to 121 observations, might have accounted for this failure.
Figure 2 shows the diagnostic trees generated from the
CART analysis. The performance of these diagnostic aids,
and of that generated by logistic regression, are summarised
in table 6. Resubstitution of the original data set into the
admission diagnostic tree misclassified ten patients (two
with tuberculous meningitis and eight with bacterial
meningitis), giving 99% sensitivity and 93% specificity. The
second tree misclassified eight patients (three with
tuberculous meningitis and five with bacterial meningitis),
giving 93% sensitivity and 95% specificity.
The test data were recorded from a further 75 patients
who satisfied the study diagnostic criteria: 20 adults had
culture-confirmed tuberculous meningitis, 22 had clinical
tuberculous meningitis, 21 had culture-confirmed bacterial
meningitis, and 12 had clinical bacterial meningitis. Clinical
data from this group were applied to each diagnostic aid.
The sensitivities and specificities are presented in table 6.

THE LANCET Vol 360 October 26, 2002 www.thelancet.com

Discussion
The diagnosis of tuberculous meningitis in adults is difficult
whatever the resources available to the physician. This study
used diagnostic criteria derived from cerebrospinal fluid
culture results and observed response to specific treatment.
Such criteria suggest that even in the best settings
cerebrospinal fluid culture is insensitive, and some cases of
tuberculous meningitis or bacterial meningitis will never be
proven microbiologically.
Because untreated tuberculous meningitis is always fatal,
a clinical diagnostic aid or laboratory assay for tuberculous
meningitis must be sensitive. The potential toxic effects and
duration of antituberculosis chemotherapy, and the limited
resources of many tuberculosis programmes, also mandates
diagnostic specificity. At present, no rapid laboratory
method for the diagnosis of tuberculous meningitis satisfies
these requirements. A clinical diagnostic rule or
classification tree for tuberculous meningitis might improve
the sensitivity of current laboratory methods, and could be
used in settings with limited microbiological diagnostic
supportie, where tuberculous meningitis is most
common.
We compared patients with tuberculous meningitis with
those having confirmed or probable bacterial meningitis for
two reasons. First, both groups of patients require
immediate decisions about chemotherapy, and second,
those with bacterial meningitis, particularly partly treated
bacterial meningitis, are difficult to distinguish from those
with tuberculous meningitis. Low cerebrospinal fluid
glucose is usually present in both disorders, and forms an
important discriminating feature from other meningoencephalitides. A low cerebrospinal fluid glucose (<50% of
that in blood) was present on admission in all but one
patient entering this study.
The most common agent isolated from patients with
bacterial meningitis was S suis, which has been well
described in southeast Asia.16 The infection is most
commonly seen in men working in frequent contact with
pigs, which explains the 78% male preponderance in the
bacterial meningitis group. The organisms isolated were all
sensitive to ceftriaxone, as were all the S pneumoniae. The
clinical features of S suis meningitis are similar to those of
other bacterial meningitides, and their inclusion in this
study is unlikely to affect the application of the diagnostic
aids produced.
The prevalence of HIV-1 infection in the study
population was low, although only 66 of 251 patients were
tested. At the time of the study, the prevalence of HIV-1
infection was high in two groups: prostitutes and
intravenous drug users.17 The prevalence of HIV-1 infection
among women attending maternity clinics in Vietnam
during 199498 was less than 015%.17 Although HIV-1
testing was limited to at-risk groups, we suggest the study is
unlikely to have included substantial numbers of patients
with unrecognised HIV-1 infection.
Although HIV-1 seems to increase the risk of developing
tuberculous meningitis,18 it does not alter the clinical and

1291

For personal use. Only reproduce with permission from The Lancet Publishing Group.

ARTICLES

laboratory features of the disease.19 However, HIV-1

infection alters the differential diagnosis in meningitic
adults: opportunistic infection with unusual pathogens must
be considered, in particular Cryptococcus neoformans, which
can present subacutely in a similar way to tuberculous
meningitis. The results of this study should therefore be
applied with caution in areas with a high prevalence of
HIV-1 infection.
The univariate analysis of admission variables suggests a
set of potentially discriminative clinical features. Patients
with tuberculous meningitis present with a longer history;
they are more likely to have cranial nerve palsies; they will
not usually have a blood leucocytosis; and their
cerebrospinal fluid will frequently be clear, with moderate
numbers of lymphocytes and neutrophils, in combination
with an increased protein concentration and a low ratio of
cerebrospinal fluid:blood glucose. However, diagnostic
uncertainty frequently persists despite the first cerebrospinal
fluid analysis. In such cases, physicians in our hospital can
assess response to treatment with a broad-spectrum
antibiotic after 48 h. As expected, the cerebrospinal fluid
variables change little over these 48 h in those with
tuberculous meningitis. The changes seen in patients with
bacterial meningitis presumably reflect successful
antimicrobial
effect.
The
dangers
of
delayed
antituberculosis chemotherapy focus attention on a
diagnostic aid that makes use of clinical features on
admission. Multivariate logistic regression analysis defined
five characteristics independently predictive of distinction
between tuberculous meningitis and bacterial meningitis:
age, history of illness, white-cell count in blood, white-cell
count in cerebrospinal fluid, and percentage of neutrophils
in cerebrospinal fluid (table 4). We elected to use a cutoff of
4 in the ROC analysis because it provides the greatest
sensitivity (97%) with acceptable specificity (91%).
Application of the test data revealed 86% sensitivity and
79% specificity. These figures are similar to those of the best
available laboratory assays.6
Two diagnostic classification trees were developed: one
for admission, and a second to be used after a trial of broadspectrum antibiotics (figure 2). When the test data were
applied to the admission tree, sensitivity was similar to that
of the diagnostic rule (88%), but specificity was reduced
(70%). The second tree performed less well, although the
data available to test this tree were small
(14 patients with tuberculous meningitis, 33 with bacterial
meningitis). Nevertheless, a diagnostic tree incorporating
the second set of cerebrospinal fluid variables is attractive
given the striking differences documented between those
with tuberculous meningitis and those with treated bacterial
meningitis, but such an aid requires further development
and prospective assessment.
There are some important limitations to this study. First,
the prevalence of both tuberculosis and HIV-1 infection will
affect the performance of the rule. We therefore suggest that
this rule should not be used in areas of substantially
different tuberculosis and HIV-1 prevalence to those of
southern Vietnam without prospective evaluation in those
settings. Because this difference in prevalence will remain a
fundamental problem for all similar diagnostic methods,
future research should also be directed at investigating the
effect of diagnostic algorithms on outcome.
This study suggests that simple clinical and laboratory
data can be used to help diagnose tuberculous meningitis in
adults, and we propose an admission diagnostic rule with
86% sensitivity and 79% specificity. We suggest that the
rule should be applied to adults in high tuberculosis
prevalence settings with meningitis and a cerebrospinal
fluid:blood glucose ratio of less than 50%. Since the

1292

diagnosis and management of meningitis rest on clinical and

cerebrospinal fluid assessment, efforts are necessary to
support clinical and appropriate laboratory diagnostic
services in low-income countries. The search for acid-fast
bacilli in the cerebrospinal fluid still represents the best
rapid laboratory diagnostic technique, but requires large
volumes of cerebrospinal fluid, and meticulous microscopy
to achieve the best results.20 A clinical rule might focus
attention on patients in whom such an approach is
warranted, and thereby optimise the use of often-limited
laboratory resources.
Contributors
The study was designed by G E Thwaites, C M Parry, and J J Farrar. Data
were collected by T T H Chau, N H Phu, L V Chuong, and
D X Sinh, and analysed by G E Thwaites and K Stepniewska.
G E Thwaites, C M Parry, J J Farrar, and N J White wrote the paper.

Conflict of interest statement

None declared.

Acknowledgments
We thank the doctors and nurses of the Clinical Research Unit, and the staff
of the Microbiology Department, Centre for Tropical Diseases.
The study was funded by the Wellcome Trust, UK.

References
1
2

3
4

6
7

10
11
12

13

14
15

16
17

18

19

20

Fallon RJ, Kennedy DH. Treatment and prognosis in tuberculous

meningitis. J Infect 1981; 3: 3944.
Humphries MJ, Teoh R, Lau J, Gabriel M. Factors of prognostic
significance in Chinese children with tuberculous meningitis.
Tubercle Lung Dis 1990; 71: 16168.
Misra UK, Kalita J, Srivastava M, Mandal SK. Prognosis of tuberculous
meningitis: a multivariate analysis. J Neurol Sci 1996; 137: 5761.
Kalita J, Misra UK. Outcome of tuberculous meningitis at 6 and 12
months: a multiple regression analysis. Int J Tuberc Lung Dis 1999; 3:
26165.
Lincoln EM, Sordillo SVR, Davies PA. Tuberculous meningitis in
children: a review of 167 untreated and 74 treated patients with special
reference to early diagnosis. J Paediatr 1960; 57: 80723.
Thwaites GE, Chau TT, Mai NT, et al. Tuberculous meningitis.
J Neuro Neurosurg Psychiatry 2000; 68: 28999.
Girgis NI, Sultan Y, Farid Z, et al. Tuberculous meningitis, Abbassia
Fever HospitalNaval Medical Research Unit. Am J Trop Med Hyg
1998; 58: 2834.
Karstaedt AS, Valtchanova S, Barriere R, Crewe-Brown HH.
Tuberculous meningitis in South African urban adults. Q J Med 1998;
91: 74347.
Hosoglu S, Ayaz C, Geyik MF, Kokoglu OF, Ceviz A. Tuberculous
meningitis in adults: an eleven-year review. Int J Tuberc Lung Dis 1998;
2: 55357.
Verdon R, Chevret S, Laissy JP, Wolff M. Tuberculous meningitis in
adults: review of 48 cases. Clin Infect Dis 1996; 22: 98288.
Pfyffer GE. Nucleic acid amplification for Mycobacterial diagnosis.
J Infect 1999; 39: 2126.
Wilson SM. Application of molecular methods to the study of diseases
prevalent in low income countries. Trans R Soc Trop Med 1998; 92:
24144.
Foulds J, OBrien R. New tools for the diagnosis of tuberculosis: the
perspective of developing countries. Int J Tuberc Lung Dis 1998; 2:
77883.
Kumar R, Singh SN, Kohili N. A diagnostic rule for tuberculous
meningitis. Arch Dis Child 1999; 81: 22124.
Ahuja GK, Mohan KK, Prasad K, Behari M. Diagnostic criteria for
tuberculous meningitis and their validation. Tubercle Lung Dis 1994; 75:
14952.
Kay R, Cheng AF, Tse CY. Streptococcus suis infection in Hong Kong.
Q J Med 1995; 88: 3947.
Nguyen TH, Hoang TL, Pham KC, et al. HIV monitoring in Vietnam:
system, methodology, and results of sentinel surveillance.
J Acquir Immune Defic Syndr 1999; 1: 33846.
Berenguer J, Moreno S, Laguna F, et al. Tuberculous meningitis in
patients with human immunodeficiency virus. N Engl J Med 1992; 326:
66872.
Yechoor VK, Shandera WX, Rodriguez P, Cate TR. Tuberculous
meningitis among adults with and without HIV infection.
Arch Intern Med 1996; 156: 171016.
Kennedy DH, Fallon RJ. Tuberculous meningitis. JAMA 1979; 241:
26468.

THE LANCET Vol 360 October 26, 2002 www.thelancet.com

For personal use. Only reproduce with permission from The Lancet Publishing Group.