Materials and methods

We conducted our study in the children’s wards of King George’s Medical College Hospital, Lucknow over a period of 30 months from August 1994 to February 1997. This is a large teaching hospital in the capital city of India’s most populous state, Uttar Pradesh, and caters for the poor and seriously ill children from the city and surrounding districts. Children between the ages of 1 month and 12 years admitted with a diagnosis of meningoencephalitis on the basis of history, examination, and pleocytosis in CSF, on three selected days of each week were enrolled and investigated according to a predesigned protocol. A complete physical examination was carried out on admission. Investigations included blood counts, CSF examination for cell count, differential count, protein, sugar, Gram stain, acid fast bacilli (AFB) stain, bacterial and mycobacterial culture, a skin test for tuberculosis, and a chest radiograph. A CT scan was also planned and was usually possible within a week of admission to hospital. The patient’s clinical course was carefully recorded. At discharge, the final diagnosis of tuberculous meningitis or non-tuberculous meningitis was made on the following criteria.

Tuberculous meningitis was diagnosed if: (1) mycobacterial culture/AFB stain was positive in the CSF or (2) basal enhancement or tuberculoma was seen on CT scan and there was a clinical response to antituberculous treatment, with or without other antibiotics. Non-tuberculous meningitis was diagnosed if: (1) CSF was positive for bacterial culture/Gram stain or (2) there was a sustained response without antituberculous treatment.

 Antituberculous treatment consisted of four drugs—streptomycin, rifampicin, isoniazid, and pyrazinamide, in appropriate doses. Pyogenic meningitis was treated with either a combination of ampicillin and chloramphenicol or cefotaxime for 10 days. In general, children in whom there was diagnostic confusion between tuberculous and non-tuberculous meningitis received both forms of treatment.

Thirty clinical/laboratory features of tuberculous and non-tuberculous meningitis were compared by univariate analysis (χ² or two sample t test, or Kruksal-Wallis test for data that were not normally distributed) using the Epi-info program. Table 1 lists the variables tested. Length of prodromal stage was defined as period between first symptom and first neurological manifestation. Weight was expressed as percentage of expected weight for age. Fever was graded as low (< 37.5°C), moderate (37.5–38.5°C), or high (more than 38.5°C). Frequency of convulsions was divided into 1–3/day, 4–6/day, and > 6/day. A history of measles included the previous six months only. Coma was graded according to Corey et al.5Presence of meningeal signs was defined as neck stiffness, Kernig’s or Brudzinsky’s sign. Variables with p < 0.25 were entered into a logistic regression analysis using the multlr package to obtain β coefficients, adjusted odds ratios, and “predictors” of the diagnosis of tuberculous meningitis.

For validation, another set of patients was enrolled in the same way using the same diagnostic (gold standard) criteria. The rule was then applied to them and compared with the gold standard diagnosis in a two by two table to calculate its sensitivity, specificity, and the likelihood ratios for various predictors and their combinations. β Coefficients obtained for the five variables in the logistic regression analysis were rounded off and added up to obtain a weighted score for each patient in the validation dataset. Sensitivity and specificity at each score were calculated and these values were used to construct a receiver operator characteristic (ROC) curve and determine the best cut off score. The best cut off point is taken as that closest to the left upper corner of the ROC curve.

Results

A total of 232 out of the 248 children who satisfied the entry criteria were enrolled. Out of these, 110 had tuberculous meningitis and 94 had non-tuberculous meningitis. A positive CSF culture forMycobacterium tuberculosis and/or smear for AFB was obtained in a total of 18 patients, whereas the remaining patients in the tuberculous meningitis group were diagnosed on the basis of basal enhancement on CT scan, along with a response to antituberculous treatment.

In 28 patients, the final diagnosis could not be reached on the basis of the diagnostic (gold standard) criteria either because the patient died early or because a CT scan could not be done or was non-confirmatory. These patients were excluded from analysis.

Table 1 shows the clinical and laboratory features in the two groups. Eighteen variables were entered into the logistic regression analysis, of which five were found to have a significant independent association with the diagnosis of tuberculous meningitis. These features are shown in table 2.

The 28 patients who were excluded from analysis included 23 boys and five girls. The mean (SD) age of these 28 patients was 66 (39) months. The median prodromal stage of illness was 13 (range, 2–75) days and 19 had a prodromal stage of seven days or more. Fundoscopy was abnormal in three (papilloedema in two and optic atrophy in one). Focal neurological deficits were found in six patients, and extrapyramidal movements were seen in nine. Median CSF polymorphs was 50% (range, 0–95%), being less than 50% in 12 patients. Six of these patients had none of the predictors present, 22 had one or more, 13 had two or more, and seven had three or more predictors present. None of these patients had a positive CSF smear or culture for mycobacteria or other bacterial pathogens. Fifteen died in hospital and two left the hospital against medical advice. A CT scan was performed in 16 patients. It was normal in five, and 11 patients had an abnormality that was not supportive of either diagnosis according to our criteria.

The five predictors of the diagnosis of tuberculous meningitis were then validated in the second set of 143 patients, in 128 of whom the diagnosis of tuberculous meningitis (66) or non-tuberculous meningitis (62) could be confirmed using the same gold standards; the remaining 15 were indeterminate. Nine of the 66 TBM patients had a positive CSF mycobacterial culture, but AFB stain and microscopy of CSF was negative in all. Table 3 shows the sensitivity, specificity, and likelihood ratios at various levels of these predictor variables. The β coefficients shown in table 2 were rounded off as 23 for prodromal stage ⩾ 7 days, two for fundal optic atrophy, 18 for focal deficits, 29 for extrapyramidal movements, and 34 for CSF leucocytes < 50% polymorphs. Figure 1 shows a ROC curve from the sensitivity and specificity calculated for each score from the validation dataset. The best cut off score was 34.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Receiver operator characteristic curve for different scores calculated for each patient in the validation dataset by rounding off and adding the β coefficients shown in table 2 (23 for prodromal stage ⩾ 7 days, two for fundal optic atrophy, 18 for focal deficits, 29 for extrapyramidal movements, and 34 for cerebrospinal fluid leucocytes < 50% polymorphs).

Discussion

Tuberculous meningitis continues to be an important cause of childhood hospital admissions in India. However, at initial presentation it can be confused with other meningoencephalitides, especially partially treated pyogenic meningitis.6 Oral antibiotic treatment may prolong the course of the latter, render culture sterile, and even result in lymphocytic predominance in the CSF. On the other hand, in our experience, tuberculous meningitis often presents with a polymorphonuclear reaction in the CSF and blood. In our hospital, in about 30% of all cases of meningitis, there is diagnostic confusion between tuberculous meningitis and pyogenic meningitis. Yet both are serious life threatening infections requiring early, aggressive treatment to prevent death and disability. Other important differential diagnoses are viral meningitis, encephalitis, and fungal meningitis.

The gold standard for diagnosis used in our study included both investigations and response to treatment and was effective in separating tuberculous meningitis from non-tuberculous meningitis. Response to treatment and follow up often have to substitute for investigations as a gold standard, and are valuable tools for retrospective diagnosis. Bacteriological proof of either tuberculous or non-tuberculous meningitis is available in only a small proportion of patients, because most patients have received antibiotics before hospitalisation, and the yield of CSF culture for mycobacteria/AFB stain is generally low. We find that a CT scan helps in differentiating tuberculous meningitis from other meningoencephalitides. Even in patients in whom polytherapy was started initially, diagnosis was usually clear after a CT scan. Some patients with pyogenic meningitis might respond initially to antituberculous treatment but would not show basal enhancement on CT. Basal enhancement has not been documented in numerous studies on pyogenic meningitis in the West,7-10and this fact has also been pointed out by other authors.11 12 The other conditions that could produce such a CT picture are fungal meningitis, torulosis, neurosarcoidosis, and carcinomatous infiltration.11 12 These conditions would not improve on antituberculous treatment and thus would not be misallocated to the tuberculous meningitis group. Viral encephalitis or meningitis would also not be misdiagnosed as tuberculous meningitis because they are not known to produce basal enhancement. This picture of basal enhancement with or without tuberculoma is seen in about 90% of children with tuberculous meningitis.13 14

Twenty eight patients did not have a firm diagnosis, either because they died or their CT scan could not be done or did not show the characteristic features of basal enhancement/tuberculoma. Such patients had to be treated as missing data and excluded from analysis because they did not satisfy the criteria for the diagnosis of either tuberculous or non-tuberculous meningitis. This number constitutes 12% of the total patients and this proportion of incomplete data is inevitable in a study of this kind.

We found that 36% of patients with tuberculous meningitis had a predominant (> 50%) polymorphonuclear type of CSF pleocytosis at initial diagnosis. This is much higher than reported previously and this type of presentation is responsible for much of the initial diagnostic confusion. We find that it is not unusual to see patients with tuberculous meningitis with a polymorphonuclear predominant CSF pleocytosis, low CSF sugar, and even polymorphonuclear leucocytosis in the peripheral blood film. Patients with pyogenic meningitis, even if treated with oral or intramuscular antibiotics, usually have a polymorphonuclear CSF response. In our study, of the 94 patients with pyogenic meningitis, only five had a predominant lymphocytic response in CSF, and four of these had received intravenous antibiotics just before presentation.

On univariate analysis, 18 variables were significantly associated with the diagnosis of tuberculous meningitis. However, on multivariate analysis only five clinical/laboratory features showed significant independent association with the diagnosis (table 2). Presence of one or more of these features was seen in almost all patients with tuberculous meningitis. Thus, if none of these signs is present, the diagnosis of tuberculous meningitis becomes highly unlikely. When three or more of these signs were present, the sensitivity was 54.5% and specificity was 98.4%, meaning that tuberculous meningitis was “ruled in” with a high degree of certainty. These values and the intermediate sensitivity and specificity levels achieved by the presence of two of these five scores (table 3) could be substantially improved by using the scoring system and ROC plot shown in fig 3.

Focal deficits were significantly more common in patients with tuberculous meningitis than other meningoencephalitides. Tuberculous meningitis produces an endarteritis and more frequent infarcts than other forms of central nervous system inflammation.15 16Extrapyramidal rigidity and movement disorders, decerebrate posturing, and optic neuritis were also seen more in tuberculous meningitis cases, perhaps because tuberculous meningitis produces more basal exudates, therefore affecting basal structures such as basal ganglia, brain stem, and cranial nerves.3 16

Our study suggests that similar levels of accuracy can be obtained using clinical features as can be achieved with the use of complex laboratory investigations. This approach is certainly more cost effective where resources are limited. However, when differentiating dangerous but treatable disorders like tuberculous meningitis and pyogenic meningitis, a test with very high accuracy is needed, or the physician may be tempted to use both forms of treatment (conventional antibiotics and antituberculous drugs). Despite this consideration, this rule will at least guide physicians towards the diagnosis of tuberculous meningitis when one or more of these predictors are present and away from such a diagnosis when all are absent.