To avoid any influence of the effects of pandemic on recent publications,26 Pubmed, EMBASE, Cochrane and Medline databases were searched from inception until March 2020. Keywords included: (“Outcome*” or “mortality” or “impact” or “recovery” or “effect*”) and (“Mycobacterium tuberculosis” or “tuberculosis” or “TB” or ‘MTB”) and (“intensive care unit*” or “intensive treatment unit*” or “critical care” or “CCU” or “ARDS” or “Acute respiratory distress syndrome” or “mechanical ventilation” or “respiratory failure”) and (“scor*” or “severity” or “APACHE” or “APACHEII” or “GCS” or “SOFA” or “SAPS” or “Charlson”), “Intravenous” or “antibiotic*” or “Rifampin” or “Isoniazid” or “ethambutol” or “Pyrazinamide”.

Published studies were included if they reported on outcomes of cohorts of patients with pulmonary TB admitted to ICU Studies involving, individuals < 18 years and those involving <10 patients were excluded. Conference abstracts, posters, patient case studies and articles with no reported outcomes were excluded.

In the first stage, we screened the titles and abstracts of all citations for potentially relevant papers. In the second stage, we examined in detail the full texts of the retrieved papers.

Information on study design, setting, population characteristics including comorbidities, reason for ICU admission as well as ICU outcomes were obtained (see Table 1). Factors affecting outcomes were also recorded in a separate Table 2, and including information on mechanical ventilation, length of hospital and ICU stay, ICU related complications were obtained. Tuberculosis related outcomes such as time to initiation of anti-TB treatment, drug susceptibility pattern, concomitant treatments, were recorded (Table 2). All ICU related severity scores were recorded. (Table 3).

The Newcastle-Ottawa assessment scale (NOS) for cohort studies was used to assess study quality and risk of bias.35 The Newcastle-Ottawa assessment scale evaluates three parameters; selection, comparability and outcome, awarding a certain number of points. The maximum number a study can receive is 9 points, indicating low risk of bias. Less than 5 points indicate a high risk of bias. The outcome used for the checklist was mortality.

Fig. 1. PRISMA flow diagram of selected studies.

Figure 1.

Flow-chart of study selection.

(0.16MB).

Seventeen out of 529 studies fulfilled the inclusion criteria and were included in the review. The studies ranged from 1995 to 2018. The studies included were from high (South Africa, South Korea, India) and low/intermediate TB incidence countries (Canada, Germany, Taiwan, France, Turkey, Portugal). All studies were retrospective except Balkema et al.29 which was prospective. A total of 947 cases with active pulmonary TB who required ICU admission were included across all studies, of which 652 were male.

Quality of studies was generally high when assessed using NOS checklist. Selection bias across studies was greatest risk due to clinician selected cohort groups, with small sample sizes. All had follow up resulting in outcomes with all subjects accounted for, and outcomes were clearly defined in all studies. No study had an overall outcome <7 points indicating low risk of bias (Table 4).

To aid an inclusive qualitative analysis, the averages of medians and means were calculated, with each study weighted equally, regardless, of the number of cases. The mean or median age of cases ranged between 31.6–76.9 years with 12/17 studies having a mean/median age > 41 years. Common comorbidities included HIV co-infection (27.1%), alcohol abuse (12.5%), diabetes (7.7%) and malnourishment (5.0%). 21% of cases were smokers. Thirty-eight% of cases had a diagnosis of TB prior to ICU admission. The most common indication for ICU admission was respiratory failure and ARDS (36.3%) followed by pneumonia (9.3%), sepsis (4.3%) and massive haemoptysis (3.8%).

Acute Physiology and Chronic Health Evaluation II was the most commonly used scoring system, reported in 13 studies, however SAPS II, quick Sequential Organ Failure Assessment (qSOFA) which identifies high-risk patients for in-hospital mortality with suspected infection outside the ICU and the Glasgow Coma Scale (GCS) were also used across the studies. The average of the mean APACHE II score was 20.2 and median 19.1, across 8 and 6 studies, respectively. The average of the median SAPS II score was 42.8 across 4 studies and median SOFA was 5.8 across 6 studies. The mean and median values for severity scores were consistently higher in fatalities than survivors in all studies except for Pecego et al. with survivors having a higher SAPS II score (Table 3).36 The average of the mean APACHE II score was 22 and 16.4 for fatalities and survivors, respectively across 5 studies. The average of the median of these scores were 23.3 and 16.5 for fatalities and survivors, respectively, across 5 studies.

Individuals requiring mechanical ventilation ranged from 37.5% to 100%. Across all studies 67.2% of cases required mechanical ventilation and the duration in days was 14.5 and 13.25 days for the median and mean values, respectively. There was a large variation for example Erbes et al. provide a mean of 26 and a range of 1-106,14 similarly Lanoix et al.30 provide a median of 8 with an interquartile range (IQR) of 1-129.

Duration of hospital stay was reported in 9 studies. The average of the median was 20.2 days across 6 studies and for the mean 51.2 days across 3 studies. The duration of ICU stay was reported in 14 studies, the average of the median was 7.8 days across 9 studies and mean was 15.6 across 4 studies. Similarly, for the duration of stay for both hospital and ICU, there was a large spread of data throughout some studies, reflected by the large interquartile ranges in Table 2.

Delay in initiation of anti TB treatment (ATT) within hospital was only reported in 8 studies, the lowest being 0 days and the largest mean value was 45 days in Penner et al.27 The prevalence of drug resistance pattern was reported in 11 studies and ranged between 0% to 28.6%, 4.9% of cases having drug resistant strains when combining all studies. Steroids were given to 11.5% of cases and vasopressor support was given to 15.0% of cases. Other treatment management was given to a smaller number of individuals including extracorporeal membrane oxygenation, tracheostomy and renal replacement therapy.

The most common reported complication was ARDS affecting 19.5% of all cases, followed by ventilator associated pneumonia (10.8%), multiple organ failure (10.5%), sepsis (9.5%) and hospital acquired infections (8.2%). Other reported complications included shock, disseminated intravascular coagulation, acute kidney infection, single organ failure and pneumothorax. In-hospital mortality ranged from 29% to 95.1% giving a mortality rate of 52.9% across all studies. In two studies a lower ratio of arterial oxygen tension to fractional inspired oxygen (PaO2/FiO2) indicated a poorer prognosis.38,33 Causes of death were reported in 6 studies with septic shock and organ failure (including respiratory failure) with, respective values of 4.7% and 3.8% of total cases as the most common causes. Other causes of death included hospital acquired infection, raised intracranial pressure, pulmonary embolism and hypoxaemia.

No studies using first line intravenous anti TB medications in ICU were found.

In this systematic review, average in-hospital mortality across 17 studies was 52.9%. This value is especially high considering availability and efficacy of ATT worldwide and the advancements in intensive care medicine. Attributable factors include delay in diagnosis and ATT initiation, altered drug absorption in critically ill patients, comorbidities and TB related complications. The most common complication and indication for ICU admission across the studies was found to be ARDS/acute hypoxaemic respiratory failure. Tuberculosis related acute respiratory failure carries a mortality rate of up to 60%, pneumonia carries a 25% mortality.38–40

Multiple organ failure and sepsis were present in 10.5% and 9.5% of cases, respectively and were included as predictors of mortality in 9 studies. Three studies documented individuals with disseminated intravascular coagulation. This can be caused by miliary TB and is a negative predictor for survival, with individuals more likely to develop ARDS than those with isolated pulmonary TB; it carries a high mortality in the ICU setting, mostly attributed to septicaemia and subsequent multiple organ failure.10 HIV/AIDS, alcohol abuse, diabetes, smoking status and chronic pancreatitis identified as independent risk factors for mortality.14

Tuberculosis is the main cause of death in people living with HIV.41 People living with HIV are 30 times more likely to develop active TB, with more severe and atypical pulmonary forms as the most common presentation.37 Two studies reported an earlier age of hospitalisation and higher rate of respiratory failure.36,37 Threshold for clinical suspicion should be lower in these individuals, given their diminished symptom presentation.12,42

Several ICU complications were reported. Critically unwell individuals are prone to drug interactions and adverse effects due to complex pharmacology, polypharmacy, disease severity and organ failure.43 Hepatotoxicity is a particular risk with isoniazid, rifampicin and pyrazinamide. Patients with underlying hepatic sequelae including prior hepatitis, alcoholic liver disease are more vulnerable. Acute kidney injury and glomerular hyperfiltration can affect anti-TB drug elimination with pyrazinamide and ethambutol renally excreted.44 Decompensated or end stage renal failure in ICU negatively influences patient outcome especially in those requiring dialysis, 45 individuals across these studies had renal failure.14 Patients in multiple organ failure are less tolerant to the toxic side effects of anti-TB drugs, creating clinical dilemmas as therapy interruption can increase risk of drug resistance and death.

Hospital acquired infections (HAIs) were found to be a negative predictor of survival and were present in 8.2% of cases. Tuberculosis suppresses monocyte activity, causing immunosuppression and increasing infection risk.34 Lin et al reported nosocomial pneumonia incidence was four times higher in non-surviving individuals with pulmonary TB.28 Ventilator associated pneumonia was found in numerous individuals who had been ventilated,14 and was independently associated with hospital mortality.14 Other infections include urinary tract and central venous catheter associated bloodstream infections which are associated with length hospital stay. Hospital acquired infections can prolong length of stay, contributing to an already elevated mortality rate.45

Smear microscopy and culture have turnaround times of few days and several weeks, respectively. GeneXpert NAAT TB-PCR test and urinary LAM (Fujifilm) may allow for results within hours.9 Despite the growing availability of fast and reliable point of care tests, thinking of TB remains a challenge. Misinterpretation of clinical and radiological presentation, and lack of resources contribute to unreliable diagnosis and delays in treatment initiation. It can be challenging to radiologically distinguish TB from severe bacterial pneumonia as a cause of ARDS, many individuals are treated incorrectly before TB is considered in the differential. Empirical fluroquinolone could be beneficial covering both conditions, Tseng et al reported oral fluoroquinolone usage as independently associated with better survival in those with TB mimicking severe pneumonia in ICU.46

Survival of individuals with TB can be significantly improved if therapy is started within 14 days of hospitalisation.46 Erbes et al found a significant increase in mortality in individuals not receiving optimal treatment including isoniazid and rifampicin.14 In addition, Duro et al found that starting ATT within 3 days of ICU admission improved survival.32 Two studies with the longest delay in treatment initiation were from lower incidence countries.13,27 Delays are common in areas with fewer TB cases, probably as a result of lack of experience.47 Almost half of the studies did not report on treatment delay. The variation in delay ranged from 0-45 days globally, and may contribute to poorer prognosis. This review found only 38.3% of individuals diagnosed prior to admission.

Rifampicin resistance is increasing and a major threat, with half a million people currently estimated to be infected with rifampicin resistant strains carrying a higher mortality.2,24 The number of individuals with drug resistant TB was 46 (4.9%). Drug resistance may have been under reported in these studies and this might explain why resistance was not found to be a predictor of mortality.

Tuberculosis treatment in ICU is complicated by organ dysfunction, drug toxicity and sub-therapeutic levels. First line drugs such as rifampicin and isoniazid are generally well absorbed when administered orally at the correct dose. In critically unwell individuals, absorption and pharmacokinetic drug properties are altered. The pharmacokinetic profile of anti-TB drugs has shown that there is a dose dependent relationship between concentration and clinical outcomes.48 Critical illness alters gut motility, impairs mucosal barrier integrity, distorts commensal flora, delays gastric emptying leading to reduced absorption.10,25,49 Hypoalbuminemia was found to be a predictor of mortality in this review with 47 individuals suffering from malnutrition pre-admission.37 Hypoalbuminemia may lead to oedema, increasing the volume of distribution of drugs, as well as impair drug absorption all leading to lower drug concentrations in serum.44,50 Parenteral administration or higher doses of drugs may be required to reach therapeutic effect.

Although no studies regarding intravenous antibiotics were found, a study by Hill suggested a role for their use.25 They compared patient groups over 2 weeks, administering standard oral versus a 33% higher dose of intravenous rifampicin, finding a three times higher ‘geometric mean area under the time concentration curve’ up to 6 h, in the intravenous group. Mortality was substantially lower in individuals given intravenous rifampicin with no reported increase in toxicity. They also found an increased survival compared to the standard oral dose, including more rapid resolution of coma and reduced mortality at 2 months and 8 months.25 Koegelenberg et al investigated the pharmacokinetics of enteral anti-TB drugs in intensive care individuals, finding that a fixed dose of rifampicin administered via nasogastric tube resulted in sub-therapeutic plasma concentrations in the majority of individuals.48

Although intravenous rifampicin is available, it is not widely accessible in low income countries.51 Other first line drugs are not always accessible or available,48 with no intravenous ATT formulation included current WHO Model List of Essential Medicines (2019).52 This leads to use of second line drugs such as fluroquinolones and aminoglycosides in the ICU setting.53

Several studies identified mechanical ventilation as a risk factor for mortality.8,30 The four highest mortality rates reported were from Kim et al. 2016 (95.1%),21 Ferreira et al. (78.3%),37 Tatar et al. (72.5%),19 and Penner et al. (69.0%),27 having the highest proportion of mechanically ventilated individuals (75-100%). Studies with the lowest proportion of mechanically ventilated individuals had the lowest reported mortality, such as Erbes et al.14 with 37% ventilated and 25.9% mortality.13,14,30 Those with more severe, disseminated forms of disease were more likely to require mechanical ventilation and develop ARDS, reflecting a referral bias, most unwell more likely to die.12 Duration of mechanical ventilation has been associated with worse prognosis, possibly due to more HAIs, and pneumothorax.14

Adjuvant corticosteroid use is indicated for meningeal and pericardial disease, and pulmonary TB related ARDS.12,32 Some studies have shown that systemic glucocorticoids are associated with improved prognosis, however this was non-specific for the critically unwell population.54 The benefit of steroid use in TB individuals in ICU specifically remains unclear. We found that steroid use did not alter prognosis. Vasopressor support was found to be a predictor of fatality.

Scoring systems for critically ill individuals are commonly used for estimating general ICU mortality, guiding clinical decision making and influencing distribution of hospital resources.55 Individuals with a higher mortality risk may benefit from earlier, targeted and potentially more aggressive treatment, given the small intervention window and a higher risk of death; this may outweigh risk of iatrogenic harm.56

Many studies have shown APACHE II and SAPS consistently underestimate mortality among individuals with pulmonary TB, especially those with ARDS and the mechanically ventilated.55,31 This highlights a shortfall in accurate risk stratification in these individuals, with a need for better tailored, ARDS specific scoring systems. APACHE does not include mechanical ventilation as an adverse outcome predictor a factor in its inaccuracy.22 In the literature it has been reported than an APACHE score >18 is associated with a higher mortality giving a predicted mortality of >29%.44 The average of mean APACHE II produced about 36% predicted mortality and using median a value about 32%. The median SAPS II and SOFA scores gave an estimated about 25% and <10%, respectively. Most of these results drastically underestimate the calculated mortality of 52.9%. The data set in Table 3 showed that the fatalities vs survivors had a higher score throughout (except for Pecago et al. 36).

Villar et al. designed an outcome score calculating 24hr post ARDS diagnosis, age, PaO2/FiO2 and plateau pressure.56,57 Similarly Kim et al. developed a mortality prediction model for individuals with TB-destroyed lung on mechanical ventilation.33 This model included age, vasopressor use, PaO2/FiO2 ratio and Brain Natriuretic Peptide (all predictors of ICU mortality in these individuals) finding this score more accurate at mortality prediction than APACHE II and SOFA.33 Lung injury severity 24 h after ARDS onset is a key determinant of outcome, reflecting the necessity for a reliable mortality prediction.56 Two studies found a low PaO2/FiO2 ratio to be a predictor of fatality.38,33 Although promising results have been obtained, further studies with perhaps additional variables are needed for external validation.58

Nine out of the 17 studies were from high burden areas. Percentage of individuals diagnosed before admission was higher in low prevalence, resource rich areas, ranging from 53.8% to 75% over 4 studies (one not reported).13,14,30,27 In comparison to 24.4% to 38.6% (two not reported) over 4 studies,21,29,33,28 showing that more individuals are living with undiagnosed tuberculosis in poorer areas. This difference may be due to better diagnostic tools available in wealthier regions. The mortality in the low prevalence areas was 41.5% compared to the high prevalence at 64.9% with the highest mortality being the Kim et al 2016 study at 95.1%.21 The association between TB and low-income areas is known, with poverty being a cause and consequence of infection. Many risk factors for disease reactivation and predictors of mortality in ICU are associated with a lower socio-economic background, including HIV infection, malnutrition, alcohol use disorder and smoking.

More individuals were mechanically ventilated in high prevalence areas with higher mortality. Mechanical ventilation remains a predictor of mortality even in low burden areas. In these areas renal failure, sepsis, ARDS and APACHE II scoring are non-specific risk factors to TB.30 There was no difference in the APACHE II score, in contrast to the differing mortality rates between the grouped studies. This may reflect the inefficiency of severity scoring systems to accurately estimate mortality in critically unwell TB individuals.

This review only included individuals admitted to ICU which may reflect referral bias, as some lower income countries may not have had access to ICU beds. There was study heterogeneity in data reported, making meta -analysis challenging. No publication bias was assessed due to small sample sizes. No long term outcomes were reported.