Tuberculosis in the intensive care unit: A retrospective descriptive cohort study with determination of a predictive fatality score

S Valade, L Raskine, M Aout, al. Tuberculosis in the intensive care unit: A retrospective descriptive cohort study with determination of a predictive fatality score. Can J Infect Dis Med Microbiol 2012;23(4):173-178. bACkgRounD: Despite effective treatments, tuberculosis-related mortality remains high among patients requiring admission to the intensive care unit (ICU). obJeCTIVe: To determine prognostic factors of death in tuberculosis patients admitted to the ICU, and to develop a simple predictive scoring system. MeThoDS: A 10-year, retrospective study of 53 patients admitted consecutively to the Hôpitaux de Paris, Hôpital Lariboisière (Paris, France) ICU with confirmed tuberculosis, was conducted. A multivariate analysis was performed to identify risk factors for death. A predictive fatality score was determined. ReSuLTS: Diagnoses included pulmonary tuberculosis (96%) and tuberculous encephalomeningitis (26%). Patients required mechanical ventilation (45%) and vasopressor infusion (28%) on admission. Twenty patients (38%) died, related to direct tuberculosis-induced organ failure (n=5), pulmonary bacterial coinfections (n=14) and pulmonary embolism (n=1). Using a multivariate analysis, three independent factors on ICU admission were predictive of fatality: miliary pulmonary tuberculosis (OR 9.04 [95% CI 1.25 to 65.30]), mechanical ventilation (OR 11.36 [95% CI 1.55 to 83.48]) and vasopressor requirement (OR 8.45 [95% CI 1.29 to 55.18]). A score generated by summing these three independent variables was effective at predicting fatality with an area under the ROC curve of 0.92 (95% CI 0.85 to 0.98). ConCLuSIonS: Fatalities remain high in patients admitted to the ICU with tuberculosis. Miliary pulmonary tuberculosis, mechanical ventilation and vasopressor requirement on admission were predictive of death. ventilation mécanique (RRR 11,36 [95 % IC 1,55 à 83,48]) et des besoins vasopressifs (RRR 8,45 [95 % IC 1,29 à 55,18]). Un indice conforme à la somme de ces trois variables indépendantes était efficace pour prévenir la fatalité, avec une zone sous la courbe ROC de 0,92 (95 % IC 0,85 à 0,98). ConCLuSIonS : Les décès demeurent élevés chez les patients tuberculeux admis à l’USI. La tuberculose miliaire, la ventilation mécanique et les besoins vasopressifs à l’admission sont prédictifs d’un décès. the ICU or in the medical wards. Two patients (4%) presented with extrapulmonary tuberculosis Chest x-rays showed alveolar pneumonia nodules miliary lesions cavitary lesions pleural effusion (29%) and mediastinal lymphadenopathy (27%). More two lobes were involved in 28 of 53 patients (53%).

). Un indice conforme à la somme de ces trois variables indépendantes était efficace pour prévenir la fatalité, avec une zone sous la courbe ROC de 0,92 (95 % IC 0,85 à 0,98). ConCLuSIonS : Les décès demeurent élevés chez les patients tuberculeux admis à l'USI. La tuberculose miliaire, la ventilation mécanique et les besoins vasopressifs à l'admission sont prédictifs d'un décès. T uberculosis remains a major public health issue despite available effective therapies. In 2007, an estimated 1.78 million people died from tuberculosis, which made it the leading infectious cause of fatalities worldwide (1). However, diagnosis may be difficult due to atypical presentations, resulting in delay of specific antibiotic therapy. Rarely, organ failure occurs. Patients with tuberculosis requiring intensive care unit (ICU) admission represent approximately 1% to 3% of all patients with tuberculosis (2). Data regarding tuberculosis in the ICU are sparse and generally refer to respiratory failure (2)(3)(4)(5)(6)(7)(8). The prognosis for tuberculosis patients appears significantly worse than for individuals experiencing nontuberculous pneumonia requiring mechanical ventilation (7). Our aims were to investigate circumstances surrounding ICU admission leading to the diagnosis of tuberculosis and to determine prognostic factors of death to develop a simple scoring system.

Patient selection and definitions
The medical charts of all patients ultimately confirmed to have tuberculosis who were hospitalized in the Hôpitaux de Paris, Hôpital Lariboisière (Paris, France) ICU over a 10-year period (March 2000 to July 2009) were retrospectively reviewed. A patient list of confirmed cases of tuberculosis was obtained from the microbiology laboratory database. Usual clinical and laboratory parameters were collected, and final outcomes of survival or death were recorded. A diagnosis of tuberculosis was based on a positive Mycobacterium tuberculosis culture in one or more biological fluids or biopsied tissue obtained during the hospital stay. Miliary tuberculosis was defined as the presence of disseminated micronodules on chest radiograph or tomography. Organ system dysfunction and acute respiratory distress syndrome (ARDS) were assessed using standard definitions. Renal injury was defined according to the Risk, Injury, Failure, Loss and ESRD (RIFLE) criteria (9). Physiological variables measured on admission were used to calculate the Simplified Acute Physiology Score (SAPS) II (10).

Statistical analysis
Continuous variables were expressed as means (± SD) or median (interquartile range [IQR]) depending on their distribution and compared using Student t tests or, if not applicable, Mann-Whitney U tests. Qualitative variables were reported as frequencies and percentages and compared using χ 2 or Fisher's exact tests when appropriate. Logistic regression with Firth's correction was used to identify independent predictors of death. A univariate logistic regression was performed first. Statistically significant variables at a 20% threshold in the univariate analysis were introduced into the stepwise multivariate logistic regression model to select independent predictive factors for the end point. The ORs were reported with their 95% CIs. The final model was evaluated for its predictive performance using the area under the ROC curve and was reported with its corresponding 95% CIs. To test the robustness of the model, internal validation was performed using bootstrap procedures (1000 bootstrap samples) to estimate over optimism associated with the area under the ROC curve and Brier scores (model scores range from 0 [perfect] to 0.25 [worthless]). The two-sided significance level was set at 5%. Analyses were conducted using SAS version 9.2 software (SAS Institute Inc, USA) and R2.11 software (www.R-project.org); P<0.05 was considered to be statistically significant.

ICu management
The median duration of ICU stay was six days (IQR three to 16 days). Management included mechanical ventilation (51%; median duration six days [IQR three to 17 days]), vasopressors (42%), corticosteroids (50%), hemodialysis (17%) and pulmonary embolization (4%). Among ventilated patients, eight were initially treated with noninvasive ventilation; however, all eventually required intubation. Quadruple antituberculosis therapy was administered in the ICU (75%) with a median three-day (IQR zero to 21 days) delay after admission. Diagnosis was made after ICU transfer to the medical wards (25%) and antituberculosis therapy was continued. In these cases, the diagnosis on ICU discharge was bacterial pneumonia (n=8), pleural lymphocytic effusion (n=3), pneumocystosis in an HIV-infected patient (n=1) and cardiogenic pulmonary edema (n=1).

outcome and prognostic factors
Eleven patients developed mechanical ventilation-acquired pneumonia (Pseudomonas aeruginosa, 33%). Twenty patients (38%) died in the ICU. Death was related to tuberculosis-induced ARDS or cardiovascular failure (n=5), pulmonary coinfections (n=4), hospital-acquired pneumonia (n=10) and massive pulmonary embolism (n=1). The univariate analysis is shown in Table 1. Comparisons between patients who received antituberculosis therapy in the ICU and those who were treated after transfer to the medical ward are shown in Table 2. Using a logistic regression model, three independent factors on ICU admission were associated with mortality: miliary pulmonary tuberculosis, mechanical ventilation and vasopressor requirement ( Table 3). The area under the ROC curve of our model was 0.92 (95% CI 0.84 to 0.99). The area under the ROC curve-associated optimism, evaluated using a bootstrap procedure, was 3% and the bias-corrected Brier score was moderate and equal to 0.13. The output from the multivariate logistic regression was used to build a simple score by summing the three independent variables (Table 4). This score was effective for predicting fatality rate with an area under the ROC curve of 0.92 (95% CI 0.85 to 0.98) ( Table 5).

Initial suspicion of tuberculosis on ICU admission N = 38
No suspicion of tuberculosis on ICU admission N = 15 Anti-tuberculosis treatment in the ICU N = 35 Anti-tuberculosis treatment after discharge from ICU N = 3 Before microbiological assessment N = 13

DISCuSSIon
In a series of tuberculosis patients admitted to the ICU in France, we found an elevated death rate (38%). We demonstrated that miliary tuberculosis, mechanical ventilation and vasopressor requirement on ICU admission were predictive of a fatal outcome. We determined a simple scoring system that was predictive of fatality based on these three independent factors. The major features of patients with tuberculosis requiring ICU admission included acute respiratory distress, altered consciousness and cardiovascular failure. A recent study (3) reported the following primary causes for ICU admission in 67 patients with tuberculosis in Brazil: respiratory failure (63%), cardiopulmonary arrest (10%), septic shock (7%), sepsis (6%) and altered sensorium (6%). Another study involving 99 patients (7) assessed the presence of shock, neurological disorders and renal failure in addition to acute respiratory failure, in 20%, 18% and 10% of the patients with active pulmonary tuberculosis admitted to the ICU, respectively. In our series, respiratory failure was the primary cause for patient admission (57%). Tuberculosis is a rare cause of ARDS, representing approximately 4.9% of the cases in one series (11). Among patients with pulmonary tuberculosis, those with miliary and disseminated forms are more likely to require mechanical ventilation and develop ARDS (7). Altered consciousness was present in 55% of our patients. Of these patients, 14 had tuberculous meningitis with abnormal neuroradiological findings including hydrocephalus, meningeal hyperfixation and abscesses in one-half of the cases. In the remaining 15 patients, neurological disorders were related to severe sepsis and hypoxemia in the absence of meningitis and abnormal computed tomography scans. Cardiovascular failure (42%) was related to sepsis, mainly due to bacterial coinfections and only rarely to M tuberculosis, suggesting a major role for associated pulmonary infections in the acute presentation of tuberculosis requiring ICU admission.
We found a mortality rate of 38%, which appears to be consistent with what has been reported previously in the literature (22% to 69%) (2)(3)(4)(5)(6)(7)(8)12). Mortality differed among these series according to the proportion of mechanically ventilated patients and extrapulmonary involvements. Consistently, Erbes et al (5) found a mortality rate of 66%, while greater than 90% of their patients required mechanical ventilation. Similarly, 68% of the 59 mechanically ventilated tuberculosis patients in a Taiwanese study (5) died. In contrast, when the rate of ventilated patients was lower (37.9%), mortality rate among tuberculosis patients was reduced (22.4%). In ventilated patients, the mortality rate with tuberculosis is nearly twice that for usual bacterial pneumonia (7). Of the patients who died in our series, 25% died directly from M tuberculosis-associated ARDS or septic shock without evidence of any other bacterial infection, as has been previously described among immunocompetent patients (13,14). However, death generally resulted from community-acquired bacterial coinfections or ventilator-associated pneumonia. Interestingly, all cases of hospital-acquired pneumonia, except one, led to the patients' death, due to the severity of the underlying tuberculosis damage affecting the lungs. Contributions from hospital-acquired infection, particularly with Gram-negative bacteria, in tuberculosis death has been suggested (4). Lin et al (12) found that nosocomial pneumonia was  Table 4 The distribution of a simple intensive care unit (ICU) tuberculosis score to predict death in tuberculosis-infected patients admitted to the ICU ICU tuberculosis score Survivors (n=33) Nonsurvivors (n=20)   independently associated with in-hospital mortality. Surprisingly, in one report, ventilation-associated pneumonia was found to be protective for mortality; however, its contribution to death was noted to be a confounding factor due to the long duration of mechanical ventilation in survivors (2). In our study, prognosticators of fatality on ICU admission were miliary pulmonary tuberculosis, mechanical ventilation (mainly for acute respiratory distress) and cardiovascular failure requiring vasopressors within the first 24 h. We were able to determine a simple predictive fatality score in patients with active tuberculosis admitted to the ICU. To the best of our knowledge, no scoring system has been published. In the literature, other clinical parameters have been significantly associated with death in the ICU, including high Acute Physiology and Chronic Health Evaluation (APACHE) II or SAPS II scores, sepsis, organ failure (ARDS and acute renal failure), hypoalbuminemia (<20 g/L) and multilobar involvement (2)(3)(4)(5)(6)(7)(8). Predicted mortality based on our score was significantly higher than what was predicted using SAPS II, except for class 0. This observation supports the notion that differences exist between both models regarding tuberculosis patients and, as previously shown, that SAPS II can underestimate mortality (7,15). Interestingly, a Sequential Organ Failure Assessment (SOFA) score was evaluated as being a valuable prognostic indicator in ICU patients with ARDS caused by miliary tuberculosis (16). Thus, we believe that our score could be complementary to SOFA and SAPS II scores, allowing for better assessment of the final prognosis for such patients. However, in contrast to our findings showing that miliary tuberculosis was predictive of death, Ki et al (17) did not find any differences regarding mortality among 66 patients with tuberculous pneumonia and 24 with miliary tuberculosis, while shock and advanced age were associated with poor outcome. Late admission to the ICU, as well as delayed treatment (>1 month after symptom onset), have also been recognized as risk factors for death (2,5,6). Similarly, in another study including 99 ICU patients in France (7), delayed treatment was assessed as an independent predictor of threeday mortality with an OR of 3.73. In contrast, similar to our study, comorbidities including chronic obstructive pulmonary disease, alcohol abuse, cancer and HIV infection did not impact tuberculosis mortality rate in the ICU (2)(3)(4)(5)(6)(7)(8). HIV-infected patients have been reported to be more likely to have extrapulmonary, disseminated and miliary tuberculosis (18). In our study, HIV-infected patients (23%) were severely immunocompromised (CD4 T lymphocyte count <200×10 9 /L) with a high incidence of extrapulmonary forms (50%); their mortality rate was comparable with non-HIV infected patients.
Tuberculosis patients are generally admitted to the ICU because of life-threatening symptoms requiring immediate supportive management including mechanical ventilation and vasopressors. Noninvasive ventilation has been shown to be effective in several cases of M tuberculosisrelated ARDS (19). In the present study, noninvasive pressure support ventilation was initiated in eight patients; however, all of them ultimately required intubation. Delayed diagnosis of tuberculosis is also common in the ICU. In one series, diagnosis was obtained on postmortem findings in three of 61 tuberculosis patients admitted to the ICU (8). In our series, 25% of patients did not receive antituberculosis antibiotics until after transfer from the ICU to the medical wards. These patients presented with acute respiratory distress or features that were related to either nontuberculosis bacterial pneumonia or noninfectious etiologies. Direct M tuberculosis staining and polymerase chain reaction tests were negative. They received general antibiotics and supportive care, which were effective treatments leading to improvement and transfer out of the ICU, despite no prophylactic antituberculosis treatment. Interestingly, the median delay for positive cultures was 19 days, while the median length of ICU stay was six days. Atypical presentations, including consolidations or absence of any abnormalities on chest radiographs, are known to be underdiagnosed (5,6). A long duration of respiratory symptoms before ICU admission in severe pneumonia should heighten suspicion for possible undiagnosed pulmonary tuberculosis (14). Fortunately, none of our patients with undiagnosed pulmonary tuberculosis died after transfer to the medical ward, despite treatment delay, which has been shown to worsen outcomes (3,6,7,20). Surprisingly, delayed diagnosis and initiation of quadruple antituberculosis therapy were not associated with a worsening prognosis in our study. These patients appeared to be less critically ill on ICU admission, with a significantly lower need for vasopressors and mechanical ventilation and a subsequently shorter length of ICU stay ( Table 2). We acknowledge that this finding may represent a bias -patients with atypical presentations, negative direct microbiological tests and no life-threatening symptoms received specific antibiotics only after microbiological evidence of tuberculosis. In contrast, tuberculosis associated with lung damage and miliary forms of tuberculosis were more rapidly treated with antituberculosis therapy.
Our study has significant limitations related to its retrospective nature and single-centre methodology. Our study may have been underpowered due to the small number of patients. Patients with massive hemoptysis were not admitted to our hospital because embolization is not available in our hospital. Due to our inclusion criteria and to the retrospective nature of our study, the definitive contribution of tuberculosis to critical illness requiring ICU admission may be difficult to assess. The partial reversibility of ICU-precipitating events in 13 patients who did not receive specific antituberculosis treatment may argue against a direct role for tuberculosis in determining the final ICU outcome. However, based on a case-by-case analysis, we estimated that tuberculosis significantly contributed to critical illness of all our patients. Finally, we were not able to analyze the consequences of resistance to antituberculosis agents because only two patients had rifampicin-resistant strains.

ConCLuSIonS
Despite available effective treatments, tuberculosis may still require ICU admission and result in high fatality rates in the 21st century, even in a developed country such as France. Miliary pulmonary tuberculosis, mechanical ventilation and vasopressor requirements on admission were associated with increased mortality, with hospitalacquired infections worsening prognosis. Physicians should be aware of atypical tuberculosis presentations on ICU admission and maintain high clinical suspicion, even despite initially negative microbiological tests.

DISCLoSuReS:
The authors have no financial disclosures or conflicts of interest to declare.