Functional outcomes in adults with tuberculous meningitis admitted to the ICU: a multicenter cohort study

We conducted a retrospective cohort study on consecutive adult TBM cases admitted to the medical ICUs of 12 hospitals, located in the Paris area, France, from January 1, 2004 to June 15, 2016. Patients were identified using the national information system (PMSI) with the following ICD-9 codes: lymphocytic meningitis, meningoencephalitis (G049), tuberculous meningoencephalitis (G050, A178), and tuberculous abscess (G07). The ethics committee of the French Society of Intensive Care Medicine (SRLF) approved the study protocol. In accordance with French law, informed consent was not required for this observational study.

Patients were included if they fulfilled the diagnosis criteria for TBM, established by the expert consensus score of 2010 [8]. This score is based on clinical information, biological criteria (including CSF analysis), brain imaging, and evidence of tuberculosis elsewhere to classify cases into three TBM categories, based on their total diagnostic score: definite TBM (i.e., microbiological identification or evidence from commercial nucleic acid amplification tests of central nervous system M. tuberculosis infection); probable TBM (diagnostic score of 12 or above when imaging is available); and possible TBM (diagnostic score of 6–11 when imaging is available).

Patients were excluded if data on the primary outcome were missing, if an alternative diagnosis was established at hospital discharge, or if a favorable outcome was observed in the absence of anti-tuberculosis therapy.

Standardized methods for enhanced quality and comparability of TBM study guidelines were followed for data collection [18]. Prior health status was assessed by the Knaus score [19]. Patients were considered immunocompromised in the case of HIV infection, solid organ transplantation, solid cancer, hematological malignancy, steroid therapy, chemotherapy, chronic alcoholism, and/or intravenous drug use. A major neurological deficit was defined as monoparesis, hemiparesis, paraparesis, or tetraparesis. Neurological status at admission was staged based on the modified British Medical Research Council (MRC) criteria: stage 1 was defined as a score on the Glasgow Coma Scale (GCS) of 15 and the absence of neurological deficit; stage 2 as a score on the GCS of 11–14, or a score on the GCS of 15 associated with focal neurological sign; and stage 3 as a score on the GCS ≤ 10 [18]. The use of invasive mechanical ventilation, vasopressors, and/or neurosurgical interventions (external ventricular drainage and/or brain biopsy) during the ICU stay was recorded. Bacterial findings were recorded, including types of samples analyzed, microbiological methods used for detection of M. tuberculosis, and anti-tuberculosis drug resistance. Initial anti-tuberculosis therapy consisted of a standard regimen with four drugs: isoniazid, rifampicin, ethambutol, and pyrazinamide. In the case of resistance, minor drugs were used [10]. Data on adjunctive steroids at admission (dose, molecule) were collected.

Functional outcome was graded 90 days after ICU admission according to the modified Rankin Scale (mRS). The primary endpoint was poor functional outcome, defined by a score of 3–6 (i.e., functional dependence, severe disability, or death) on the mRS 90 days after ICU admission. The secondary endpoint was mortality 1 year after ICU admission.

Data are reported as median (interquartile range) or number (%). Patients’ characteristics were compared according to functional outcome at 90 days, using Fisher’s exact tests for categorical variables and Mann–Whitney tests for continuous variables. Durations were calculated from the time of ICU admission. Univariate logistic regression analysis on nonimputed data was performed to evaluate associations between variables and functional outcome. Clinically relevant variables, including use of steroids, and other variables associated with poor outcome in univariate analysis (p <  0.1) were included in the multivariate model on imputed data. A stepwise selection was used to select variables to construct the final model. The adjusted odds ratio (aOR) values with 95% confidence intervals (95% CIs) were computed. Univariate Cox proportional hazard modeling on nonimputed data was performed to evaluate associations between variables and 1-year mortality. Clinically relevant variables, including use of steroids, and those associated with mortality in univariate analysis (p <  0.1) were entered into the multivariate model on imputed data. Hazard ratios (HRs) and their 95% CIs were computed. Survival rates between patients who received adjunctive steroids and patients who did not were compared with the log-rank test and Kaplan–Meier survival curves were computed. Multiple imputations were performed using the multivariate normal distribution (MVN) method. Sensitivity analysis was conducted to assess the impact of no anti-tuberculosis therapy. All analyses were performed using SAS 9.4 (SAS, Inc.) software. p ≤ 0.05 was considered statistically significant.

Among the 112 eligible patients, 22 were excluded (Additional file 1). Baseline characteristics of the 90 included patients (age 43 (29–58) years, 56 (62%) males) are presented in Table 1. Overall, 41 (46%) patients were immunocompromised, including 20 (22%) with HIV infection, and 61 (68%) patients had MRC grade 3 illness at admission. The temperature was 38 (37–39) °C, and fever (T° > 38 °C) was documented in 43 (48%) cases. A major neurological deficit was observed in 41 (46%) cases and seizures were noted in 31 (34%) patients (including 12 (13%) with convulsive status epilepticus). Extraneurological findings were reported in 62 (69%) patients.

Hyponatremia was common, with blood sodium levels of 131 (127–135) mmol/L at admission.

CSF analysis revealed a typical pleocytosis of 130 (27–300) cells/μl with a lymphocyte proportion of 73% (37–94%), with elevated CSF protein levels (1.9 (1.0–3.0) g/L) and low glucose levels (2 (1.1–3.0) mmol/L). Bacteriological findings are reported in Additional file 2.

Brain CT and MRI findings are presented in Table 1. MRI appeared superior to CT for detection of infarction, arachnoiditis, and tuberculoma at ICU admission (Fig. 1).

Based on the consensus diagnostic score, 41 (46%) patients were diagnosed with definite TBM, 28 (31%) patients with probable TBM, and 21 (23%) patients with possible TBM. Data on organ support, specific therapy, and adjunctive steroids are presented in Additional file 3.

First-line anti-tuberculosis therapy was initiated in 85 (96%) cases (five patients died before treatment could be started). Anti-tuberculosis drugs for suspected drug-sensitive TBM consisted of a combination of isoniazid (5 mg/kg, maximum 300 mg, 12 months), rifampicin (10 mg/kg, maximum 600 mg, 12 months), ethambutol (15 mg/kg, first 2 months), and pyrazinamide (25 mg/kg, first 2 months). Ten patients were secondarily detected with resistance to standard treatment and switched to other drugs during their ICU stay.

Adjunctive steroids were initiated in 72 (80%) patients. The steroid dose was recorded for 61/72 (93%) patients and was > 0.4 mg/kg/day of dexamethasone equivalent in 15/61 (25%) patients. Overall, a total of 63 (70%) patients required invasive mechanical ventilation, and 36 (40%) patients required vasopressors. Neurosurgery was performed in 18 (20%) patients.

At 90 days, 55/90 (61%) patients had a poor outcome, including 39/90 (43%) deaths (Fig. 2). Most deaths occurred in the ICU (n = 36/90, 40%), including 14/36 (39%) because of severe neurological injury. The univariate logistic regression analysis is presented in Additional file 4. Multivariate analysis identified three independent indicators of poor outcome (Table 2): age (aOR 1.03, 95% CI 1.0–1.07), CSF protein level ≥ 2 g/L (aOR 5.31, 95% CI 1.67–16.85), and hydrocephalus on MRI (aOR 17.2, 95% CI 2.57–115.14). By contrast, adjunctive steroids were protective (aOR 0.13, 95% CI 0.03–0.56). Sensitivity analysis conducted after exclusion of the five patients who did not receive anti-tuberculosis therapy displayed no major change in the adjusted odds ratio (aOR 0.22, 95% CI 0.05–1.04). Associations between neurologic complications on brain MRI and outcomes at 90 days are shown in Fig. 3.

One-year outcomes were available for 80 patients. Kaplan–Meier analysis estimated the 1-year overall mortality at 47% (95% CI 37–59%) (Fig. 2). Among 1-year survivors, functional independence (mRS of 0–2) was observed in 27/37 (73%, 95% CI 59–87%) cases.

The univariate Cox analysis is presented in Additional file 5. Multivariate Cox regression analysis identified one parameter positively associated with 1-year mortality (Additional file 6): CSF protein level ≥ 2 g/L (aHR 2.22, 95% CI 1.1–4.49). By contrast, adjunctive steroids had an independent protective effect on mortality (aHR 0.23, 95% CI 0.11–0.44). Sensitivity analysis conducted after exclusion of the five patients who did not receive anti-tuberculosis therapy displayed no major change in the adjusted hazard ratio (aHR 0.32, 95% CI 0.14–0.70). Kaplan–Meier survival analysis according to the use of adjunctive steroids is reported in Additional file 7.

An exploratory analysis comparing baseline characteristics and outcomes according to two arbitrarily defined periods is presented in Additional file 8. We observed no major difference in presentation, imaging findings, and outcomes between the two periods.