Clinical review: Tuberculosis on the intensive care unit

Rates of tuberculosis (TB) are increasing in most west European nations. Patients with TB can be admitted to an ICU for a variety of reasons, including respiratory failure, multiorgan failure and decreased consciousness associated with central nervous system disease. TB is a treatable disease but the mortality for patients admitted with TB to an ICU remains high. Management challenges exist in establishing a prompt diagnosis and administering effective treatment on the ICU with potentially poor gastric absorption and high rates of organ dysfunction and drug toxicity. In this review reasons for ICU admission, methods of achieving a confident diagnosis through direct and inferred methods, anti-tuberculosis treatment (including steroid and other adjuvant therapies) and specific management problems with particular relevance to the intensivist are discussed. The role of therapeutic drug monitoring, judicious use of alternative regimes in the context of toxicity or organ dysfunction and when to suspect paradoxical tuberculosis reactions are also covered. Diagnostic and therapeutic algorithms are proposed to guide ICU doctors in the management of this sometimes complicated disease.

Patients with TB requiring ICU care may have high rates of co-morbidities and ICU related complications. In one German study [8] 65.5% of patients had deranged liver function, 12.1% chronic pancreatitis, 8.6% chronic renal failure and 6.9% HIV co-infection. ICU related complications were also common, with nosocomial pneu monia in 67.2% patients, pneumothorax in 13.8%, ARDS in 12.1%, acute renal failure in 12.1% and MOF in 3.4%. Rates of co-existing extra-pulmonary TB can be up to 19 to 22% [6,8].
Delay in appropriate treatment due to lack of early recognition of TB may result in progressive multi-organ involvement and ICU admission [6] and an important opportunity between hospital admission and ICU presen tation may exist. Given the high ARDS rates in ventilated patients with TB, standard mechanical venti lation strategies to reduce ARDS may be appropriate, including lower tidal volumes and a conservative fl uid strategy [12].
Mortality is high for patients with active TB and respiratory failure; a Canadian study found a signifi cantly higher in-hospital mortality of 69% for patients requiring mechanical ventilation for TB in comparison to ARDS of any cause (56%) and nontuberculous pneumonia (36%) requiring mechanical ventilation [13]. Risk factors for mortality include older age, nosocomial pneumonia, TB destroyed lung, MOF, a duration of symptoms of more than 4 weeks, and an APACHE-II score >20 [14,15].

Miliary tuberculosis
Miliary TB is a form of TB where there is haematological dissemination from focal infection into the blood, leading to seeding of multiple organs with TB bacilli. A minority of patients may present with symptoms of less than four weeks duration. An underlying predisposing condition such as diabetes or steroid use will be present in about a third of patients [16] and immunosuppression due to HIV or iatrogenic reasons may also contribute [17]. Th e CXR may be initially normal for the fi rst few weeks of the disease. Mortality is about 25% overall [16] but assumed near 100% if untreated. Patients with miliary TB may be more likely to develop ARDS than patients with isolated pulmonary TB and MOF may account for most of the mortality of patients with miliary TB on ICU [10]. A retrospective study by Silva and colleagues [4], which included high rates of extrapulmonary TB (about 63%) and HIV seropositivity, described MOF in over 80% of patients.
Disseminated TB can rarely lead to a septic shock with MOF presentation, sometimes described as Landouzy septicaemia after the original report [18]. Th is is usually described in association with HIV infection [19], and recently associated with monoclonal antibodies used in the treatment of rheumatological disease [17], but can also occur in patients with no obvious risk factors [20]. Disseminated TB may also cause adrenal insuffi ciency [21], which should be considered in the context of refractory hypotension or hyponatraemia.

Tuberculosis meningitis and other central nervous system tuberculosis
Tuberculosis of the central nervous system (CNS) occurs in approximately 1% of TB cases, and includes TBM and cerebral tuberculomas. It carries a high mortality and is probably fatal if untreated. It is the reason for about 6 to Table 1. Other potential reasons for admission of a patient with tuberculosis to an ICU 18% [4,6] of TB related ICU admissions. In a French case series of TBM admitted to ICU [22], nearly all patients were admitted due to a falling conscious level, 75% required mechanical ventilation and 33% underwent a neurosurgical procedure; 1 year mortality was 65%. Clinical symptoms range from an acute illness mimicking bacterial meningitis to a non-specifi c illness of fever and headache, with a smaller proportion having cranial nerve palsies [22]. Th e mean duration of symptoms ranges from 12 to 29 days in most series and approximately one-third will have symptoms lasting less than a week [7]. Th e majority of the literature on CNS TB comes from the paediatric age group. Cerebral tuberculomas can present as seizures (focal and generalised), as well as with focal neurological signs; occasionally a tuberculous brain abscess can form that may require neurosurgical intervention as well as ATT [23].
As well as a high mortality, TBM poses two additional particular challenges for the intensivist relevant to neurocritical care -hydrocephalus and hyponatraemia. Hydrocephalus is common and develops radiologically in about 77% [24] of cases. It is usually due to communicating hydrocephalus associated with tuberculous exudates in the basal cisterns, but may be non-communicating in a smaller (17 to 25%) proportion of cases -for example, due to obstruction at the outlet foramen of the fourth ventricle or the cerebral aqueduct by oedema or a tuberculoma [25]. Hydrocephalus may increase the intracranial pressure, leading to reduced cerebral perfusion and ischaemia, and in more advanced cases cause brain herniation. Non-communicating hydrocephalus usually requires a neurosurgical procedure such as a shunt procedure or an endoscopic third ventriculostomy [26]. Cases of communicating hydrocephalus should also be discussed with a neurosurgical centre. Th ere may be a role for external ventricular drainage in patients with a low Glasgow coma score and TBM [27]. Hyponatraemia is common in TBM patients and is independently associated with a worse outcome [26]. It is multifactorial, and the syndrome of inappropriate anti-diuretic hormone and cerebral salt wasting probably both play a role [28]. Th e best approach to TBM-associated hyponatraemia is uncertain, and hypertonic saline, fl uid restriction, fl udrocortisone and demeclocycline may all have a role depending on the fl uid state of the patient [26]. Other aspects of neurocritical care may be relevant for TBM. Th ese additional interventions may include intra-cranial pressure monitoring [25], a higher transfusion threshold [29], and control of fever [30].

Diagnosis of tuberculosis
Culture confi rmation of tuberculosis should be obtained where possible [31]; the World Health Organisation (WHO) recommends all patients suspected to have pulmonary TB submit at least two sputum specimens for microscopic examination [32]. Culture not only confi rms the diagnosis but also provides drug susceptibility testing. In the UK, 8.4% of isolates were resistant to any fi rst line drug, and 1.6% of isolates were multi-drug resistant TB (MDR TB) [2]; rates of MDR TB are much higher in some countries.
New liquid culture techniques should give a culture result in 2 to 4 weeks [31]. An initial targeted sample for auramine or Ziehl-Neelsen stain is important as this would help confi rm TB in the appropriate clinical situation. Th ere is little evidence specifi c to diagnosis in an ICU setting. TB can present acutely and a high index of suspicion should be had in the majority of ill patients with an abnormal CXR, particularly if risk factors for TB are present. An incidental fi nding of acid fast bacilli (AFB) in sputum in a patient not suspected to have TB may be due to non-tuberculous mycobacteria (NTM) and correlation with other diagnostic data such as radiology and immune status is advised; 35% of AFB seen in sputum was due to NTM in one study [33]. NTM may cause pulmonary disease especially in the immunosuppressed or may not be signifi cant. Conversely, an AFB-positive specimen in a patient with a suggestive history, clinical features and radiology should be assumed to be TB unless proved otherwise; TB PCR may have a role if there is doubt.
In patients who are expectorating, serial sputum sampling probably provides a yield similar to broncho scopy [34]. Microbiological sampling in the non-expectorating patient may be carried out via bronchoscopy if the patient is intubated. Transbronchial biopsy may provide histology to enable rapid diagnosis and may increase the diagnostic yield [35]; this can carry a risk of complications in the mechanically ventilated patient (mainly bleeding and pneumothorax [36]) but in selected patients the risk/ benefi t ratio may be favourable, particularly in patients with diff use lung shadowing or miliary disease. Specialised procedures such as bronchoscopy through a noninvasive ventilation mask may facilitate sampling in a hypoxic non-intubated patient and this is carried out in our institution. Where facilities exist, induced sputum may have a role in some patients and has been shown to have similar or better sensitivities to bronchoscopy (73 and 87% sensitivity, respectively, in smear-negative patients [37]). No data on the utility of respiratory secretions ob tained at suction for diagnosis of TB on ICU/HDU exist but transtracheal aspirates have shown a high sensitivity (88%) in smear-negative patients [38]. If there are signs and symptoms reinforced by appropriate investi gations consistent with a TB diagnosis, treatment should be started without waiting for culture results, and continued even if subsequent culture results are negative [31].

Non-pulmonary samples
Extrapulmonary TB is common in ICU patients with pulmonary TB and in the context of advanced immunosuppression associated with HIV [39]; in the latter, visualised AFB may be due to NTM. Whilst culture is central to confi rmation of the mycobacterium, PCR for TB may have a role in this patient population to help establish a diagnosis [40]. Most forms of extrapulmonary TB have a lower bacterial load than pulmonary disease and histology/cytology such as pleural biopsies and lymph node aspiration play a more prominent role in diagnosis. A lymphocytic pleural eff usion, caseating granulomas or granulomas with Langhan's giant cells are suggestive of TB [31], although other conditions may also cause these histological changes; lymphoma is a main diff erential for a lymphocytic eff usion. A lymphocytic eff usion in an ICU setting with a high index of suspicion for TB should trigger ATT. A small proportion (6.7%) of patients, especially if the duration of illness is short, may have a predominately polymorphonuclear leucocytic pleural eff usion [41]. Adenosine deaminase levels may have a role in conjuction with other pleural fl uid results, although this is a controversial area [42]. Urine culture for TB may be useful in miliary disease, with positive cultures in about 25 to 69% of cases [43,44]. Urine microscopy for AFB is not usually carried out due to high rates of NTM.
Gastric aspirates may be easy to obtain on an ICU. Th ere is little evidence for their utility in diagnosing TB in adults, and none from an ICU setting. Th e presence of AFB in a gastric aspirate may be due to atypical mycobacteria leading to a low specifi city in some studies [45] but not others [46]. A culture rate of about 5% for Mycobacterium tuberculosis (MTB) has been described in gastric aspirates in patients with pulmonary TB [46].
Th e diagnosis of TBM is based on typical cerebrospinal fl uid (CSF) fi ndings (high protein, low glucose and predominately lymphocytic CSF) and confi rmation if possible of TB elsewhere in the body. Th ere may be a role for repeat lumbar puncture in 48 hours in atypical CSF fi ndings. Culture of TB in CSF may occur in up to 80% of cases if larger volumes of CSF (>6 ml) are submitted. Repeated lumbar punctures may increase diagnostic yield [26]. TBM is probably fatal if not treated and there should be a low threshold for empirical therapy unless a clear alternative diagnosis is made.
Blood cultures for mycobacteria are particularly useful in the diagnosis of disseminated TB and NTM (especially mycobacterium avium intracellulare) in HIV aff ected individuals with a low CD4 count, with mycobacteraemia being proportional to the CD4 count [47]. Th e yield drops to zero with CD4 counts above 300 cells/μl. Th e utility of mycobacterial blood cultures in non-HIV immunosuppresion is unknown. Bone marrow exami na tion yielded positive results in the majority of patients with miliary TB when it was carried out [16], although the place of bone marrow examination in the absence of haematological abnormalities requires clarifi cation.

Interferon gamma release assays/nucleic acid amplifi cation tests
Th e past decade has seen the development of interferon gamma release assays (IGRAs) in the diagnosis of latent TB. IGRAs work on the principle of measuring the cytokine interferon gamma released from T cells in response to synthetic antigens that are also found in MTB and have an approximate 90% sensitivity and 99% specifi city for diagnosis of latent TB [48]. Th ese antigens are absent from Mycobacterium bovis and most NTM so are not aff ected by prior Bacillus Calmette-Guérin vaccination or most NTM exposure.
IGRAs cannot distinguish between latent and active TB and therefore may be positive in an individual who has latent TB but another cause for ICU admission. It is recommended that IGRAs should not be used as a routine diagnostic tool in active TB, although there may be a role in using IGRA with other complementary tests when TB is suspected, especially if samples are diffi cult to obtain [49]. A negative test does not exclude active TB. ICU admission has also been associated with a false negative IGRA [50] and so the role of IGRA in intensive care patients is less clear. Th ere is ongoing research into next generation IGRAs and T-cell-based diagnostic platforms that may overcome some of the current limitations of IGRAs [51].
Nucleic acid amplifi cation tests (NAATs) tend not to be routinely used, and the sensitivity and specifi city can be highly variable compared with culture results [40]. A negative result does not exclude TB and a positive result does not give drug sensitivities. One exception is the Xpert MTB/RIF probe, which has shown a 98.2% and 72.5% sensitivity for diagnosing TB in smear-positive and smear-negative patients, respectively [52]. Th e probe also identifi es mutations associated with rifampicin (R) resistance and may have a role where MDR TB is suspected; for this reason WHO suggests this test as a follow on test for patients with smear-negative samples. British recommendations for NAATs currently restrict their role to the case where rapid confi rmation of a TB diagnosis in a smear-positive patient would alter manage ment (for instance, if an NTM was suspected). Th ere are few recom mendations on the role of NAATs on nonrespiratory samples. Data from meta-analysis gives a 56% sensitivity and 98% specifi city for CSF, which is similar to microscopy [26]. Th ere may be a role for NAAT on CSF when ATT treatment has been started without a confi rmed diagnosis, as mycobacterial DNA may remain detectable for a month after the start of treatment [26].

Radiology
Upper lobe disease on a CXR was shown to increase the odds ratio of TB by 14.6 [53]. Th ere are fewer data specifi c to ICU patients; small nodular or cavitary patterns on a CXR as well as a duration of illness of more than 2 weeks may be predictive of TB in some studies [54], although other studies fail to identify radiological changes specifi c for TB on an ICU [55]. In HIV coinfected patients, radio logical appearances can vary and cavitation becomes less common as immunosuppression advances. An American study described a normal or near normal CXR in 19% of pulmonary TB patients with a CD4 count less than 200 cells/μl [56]. Computed tomography (CT) scanning may have a role in identifying active TB and allowing diff erentiation from old fi brotic lesions, with centrilobular nodules and a 'tree in bud' pattern often seen in active disease. Mediastinal lymphadenopathy and cavitation may also raise the suspicion of TB (Figure 2), and miliary shadowing may be present on CT even with a normal chest X-ray [57]. CT chest may help in gathering diagnostic information in an intubated patient where TB is suspected but not confi rmed, and also allow targeting of bronchoscopy.
A proposed diagnostic algorithm for respiratory failure is presented in Figure 3.

Anti tuberculosis treatment and adjuvant therapies (including paradoxical reactions)
Th e standard treatment for non-MDR TB involves combination therapy of more than three drugs; R and isoniazid (H) provide a crucial backbone to ATT regimes, allowing shorter courses of 6 to 9 months (1 year for CNS TB) to be effi cacious due to their bacteriocidal action. R and H are associated with the potential serious side eff ects of hepatotoxicity. Two other fi rst line ATT, pyrazina mide (Z) and ethambutol (E), are renally excreted and E may be associated with optic nerve toxicity. Other drugs used to treat TB include fl uoroquinolones (for example, moxifl oxacin), aminoglycosides (for example, streptomycin or amikacin) and a range of other second line anti-TB drugs such as cycloserine or prothionamide. Not all ATT is available parenterally; parenteral preparations are available for R, H, fl uoroquinolones and aminoglycosides. Th e management of HIV co-infection is compli cated and detailed discussion is beyond the scope of this article; guidelines exist for management outside of the HDU/ICU [40]. Principles of HIV/TB co-infection management include awareness of the immune reconstitution infl ammatory syndrome (IRIS), and when highly active antiretroviral therapy (HAART) should be started; this depends on the CD4 count but is recommended as soon as is practical in the very immunosuppressed (CD4 count <100 cells/μl) [40].
High rates of hepatic and renal dysfunction in ICU patients with TB provide specifi c challenges. Patients on the ICU may have uncertain enteral absorption [58]. Subtherapeutic levels of ATT have been associated with a slow clinical response, treatment failure and drug resistance [59]. Th e incidence of subtherapeutic levels of anti-TB drugs in ICU patients is not known but it is reasonable to have a low threshold for therapeutic drug monitoring in TB patients on the ICU. Our practice is to prefer parenteral therapy in severely ill patients for the initial 72 hours.
Additional bacterial infection can complicate TBrelated ICU admissions [8] and a low threshold for additional anti-bactericidal therapy should be present.

Hepatotoxicity
Hepatotoxicity may be associated with older age, malnutrition, alcoholism, HIV or viral hepatitis co-infection [60]. In miliary TB it is important to ensure that the cause of the deranged liver function is not due to TB itself (by imaging or consideration of liver biopsy), as the management of deranged liver function in this context would be management of the TB. International guidelines diff er slightly but suggest stopping TB medication if transaminases are more than three to fi ve times the upper limit of normal or there is a bilirubin rise [60,61]. If it is crucial to continue ATT in the short term (for instance in TBM where treatment interruptions are an independent risk factor for death [26]), a combination of relatively non-hepatotoxic drugs, such as an amino glycoside, E and a fl uoroquinolone, could be given [60]. Ethionamide and prothionamide may be an alternative to E as they penetrate the meninges well and do not have the concern of optic nerve toxicity [62].
As R and H are important for TB treatment, they are usually sequentially reintroduced once liver function tests improve with close monitoring, and standardised re-challenge protocols are provided [60,61]. In cases of severe or prolonged hepatotoxicity who have had R and H re-introduced, it may be reasonable not to re-challenge with Z and extend treatment to 9 months [60]. Th e management of patients with decompensated liver disease and TB is not clear, and the standard regime with close monitoring, or 18 to 24 months of E, a fl uoroquinolone and an aminoglycoside is suggested [60].

Nephrotoxicity
Dose adjustments are required with Z and E with a glomerular fi ltration rate of less than 30 ml/minute/1.73 m 2 . Aminoglycosides and cycloserine may require similar dose adjustments [63]. No data exist for patients on continuous renal replacement therapy, and collaboration between intensivists, pharmacists, TB physicians, renal physicians and monitoring of drug levels is appropriate.

Corticosteroids as adjuvant therapy
Corticosteroids inhibit release of infl ammatory cytokines, which may help lessen tissue damage and constitutional symptoms. Th ere have been many studies of corticosteroids in TB, including advanced pulmonary TB. Th e only indications for steroids recognised in most inter national guidelines are a) TBM, where corticosteroids help reduce risk of death or disability [26,31] and b) tuberculous pericardial eff usion where corticosteroids decrease the amount and rate of re-accumulation of tuberculous pericardial eff usion [31,64]. Th e largest trial of steroids in TBM used a reducing dose of dexamethasone, initially given intravenously at a dose of 0.3 to 0.4 mg/kg/day (depending on the severity of the meningitis) for a total of 6 to 8 weeks [65]; British guidelines suggest prednisolone (or equivalent) 20 to 40 mg/day with gradual withdrawal. Th ere are no direct comparisons of the dose or type of steroid in TBM. For tuberculous pleural eff usion there is some evidence for adjuvant corticosteroids resulting in a faster resolution of pleural eff usion at 4 weeks, but no diff erence in residual fl uid at 8 weeks or death rates between corticosteroid and non-corticosteroid groups [66]. Many trials of corticosteroids for pulmonary TB were carried out in the 1950s to 1960s and suggested a more rapid clinical and radiological improvement compared to control patients, particularly in severe disease [67], but an absence of longer term eff ects on survival or risk of long-term lung damage. Th e implications of this for patients with advanced TB and respiratory failure on an ICU are unclear. A recent meta-analysis suggests a non-signifi cant trend towards benefi t of steroids in pulmonary TB [68], and a 2008 study [14] suggested a lower mortality rate in patients with pulmonary TB who received corticosteroids but fi rm conclusions cannot be drawn due to the retrospective nature of the study.

Paradoxical reactions/immune reconstitution infl ammatory syndrome
A paradoxical reaction in TB is defi ned as a clinical or radiological worsening of pre-existing tuberculous lesions or the development of new lesions in patients receiving ATT in the absence of an alternative expla nation such as drug resistance, ineff ective drug delivery or a secondary diagnosis [69]. Patients with HIV are more likely to develop these reactions, which are referred to in the context of HIV co-infection as IRIS [40]. Th e aetiology of these reactions is unknown, but in HIV may relate to HAART causing a reconstitution of immunity leading to an immune response to dead bacilli. Th e incidence in HIV-negative patients is probably between 2 and 23% [70], and about 32 and 36% in HIV-positive patients, and may be more in patients with advanced TB. Paradoxical reactions are usually mild and may manifest as recurrent fever, deterioration in radiological appearances or lymph node infl ammation [70] but may cause signifi cant morbidity, including airway obstruction, splenic rupture, or worsening neurology due to new or enlarging intracranial tuberculomas [69,71]. Th e median time of onset is about 26 days after treatment [70], but can occur months into treatment. Risk factors for IRIS include a low baseline CD4 count, rapid recovery in CD4 numbers and HAART started within 2 months of diagnosis [40]. Th ere are no fi rm recommendations on how to treat paradoxical reactions but a tapering dose of corticosteroids is reasonable [69]. Th alidomide may have a role in severe CNS TB paradoxical reactions unresponsive to corticosteroids [72], and montelukast has been used in IRIS. An algorithm for treatment of pulmonary TB on the ICU based on the above review is presented in Figure 4.

Conclusion
Respiratory failure and miliary TB are common reasons for admission of patients with TB to an ICU. Th e evidence base for diagnosis and management of TB specifi c to an ICU setting is sparse and what may be true for stable TB patients may not translate to an ICU. Th ese patients have a high mortality and high rates of organ dysfunction. Diagnosis of TB if not confi rmed prior to ICU admission may be challenging but culture of mycobacterium TB should be attempted, with radiology, histology and possibly IGRAs contributing to the clinical picture. Treatment is complicated by drug toxicity, erratic absorption and organ dysfunction, and therapeutic drug monitoring and judicious use of alternative regimes may help this problem. Corticosteroids have a role in TBM and pericardial TB and may have a role in far advanced pulmonary TB on a case by case basis. Th e clinician should be alert to paradoxical reactions as a cause of apparent treatment failure or disease progression.