The coinfection HIV-TB presented a high long-term mortality in our cohort. The nadir CD4 cell count was the independent marker for six-month mortality, while a HIV diagnosis up to three months before hospital admission was protective among these severely ill HIV-TB patients. No aspect of tuberculosis presentation or treatment influenced the outcome. Among the organ dysfunctions, neurological dysfunction was more frequent in nonsurvivors, even after the exclusion of those patients with primary TB CNS involvement.
HIV infection and critical illness
Since the beginning of the HIV epidemic, respiratory failure has been the main reason for ICU admission with the spectrum of etiological agents changing according to the geographic region and cART access [
25‐
28]. Influenza, pneumococcal vaccination and TB background prevalence also influence the main causes of respiratory disease in this population [
29‐
32]. Tuberculosis is highly prevalent in Brazil, mainly in Rio de Janeiro, and the HIV infection epidemic has made TB a common cause of opportunistic infections responsible for AIDS diagnosis in our population [
2,
33]. Rio de Janeiro has one of the worst scenarios in regards to the Tuberculosis epidemic in Brazil. In 2012, 10,649 new tuberculosis cases were diagnosed in Rio de Janeiro State, with an incidence rate of 65.6/100.000 person-years, a rate much higher than the overall country incidence rate of 35.8/100.000 person-years. The mortality rate among TB patients from Rio de Janeiro is the second highest in the entire country, reaching 5.1/100.000 person-years.
Respiratory failure is the main reason for ICU admission among pulmonary HIV-TB patients [
8‐
10], as we described in our population, even though 45 % of patients present with disseminated tuberculosis. Isolated extrapulmonary disease was uncommon (5 %).
Severe immunosuppression caused by HIV infection is a known risk factor for tuberculosis [
7]. As immunosuppression evolves, the TB disease spectrum changes from its classic pulmonary presentation of unilateral upper lobe infiltrate to a broad range of nonspecific signs and symptoms related to hematogenic spread of the bacillus. The myriad of presentations of TB in the population with HIV-related disease is associated with challenges in establishing the diagnosis and higher mortality [
33]. Our high proportion of disseminated TB (45 %) is due to the advanced immunodeficiency status of our population, with median CD4 cell counts of 72 cells/mm
3, which were lower than previously reported (109 and 112 CD4 cells/mm
3) [
28,
29]. CD4 cell count is thought to be an unreliable predictor of ICU mortality, according to Morris and colleagues, when compared to other known risk factors, such as mechanical ventilation, low serum albumin and APACHE II score [
11,
34‐
37]. However, others authors believe that CD4 cell count does have a correlation with a poorer prognosis [
25,
38‐
40]. Low CD4 counts (<200 cells/mm
3) have been associated with ICU mortality in HIV patients with TB [
41]. In the present study, we found that patients with higher nadir CD4 cell counts were more likely to be alive six months after hospital discharge. These data are supported by the idea that CD4 cell counts at ICU admission are a better marker of long-term prognosis [
42]. We preferred nadir than recent CD4 cell count because opportunistic infections and concomitant acute infections can lower the CD4 cell count [
45]. Alongside with the low nadir CD4 cell count, we wanted to emphasize that we also observed a low median CD4 count close to ICU admission meaning that our population represents a profoundly immunocompromised population. Also, nadir CD4 cell count it is better marker of prognosis and disease evolution in HIV infected population [
43,
44]. We also found that recent HIV diagnosis was associated with protective effect for long term survival. This variable should be evaluated with caution: there is not an accurate correlation of the moment of first HIV positive serology and the time since HIV infection began. Some patients are admitted with recent HIV diagnosis and very low nadir CD4 count, although they were previously less exposed to bacterial and multilple oportunistic infections and to the HIV infection per se, than their peers with long known HIV diagnosis and non-adherence to cART. This state of chronic immunosuppression by HIV and opportunistic infections may have contributed to a lower survival rate in this population, as shown previously in a heterogeneous group of studies with and without cART influence [
11,
30,
58].
As show earlier from our institution’s cohort, the HIV diagnosis frequently occurs during an advanced immunodeficiency state, which can be associated with a high prevalence of tuberculosis [
46]. Earlier diagnosis is fundamental to diminish the complications of the advanced immunodeficiency.
Delivering cART in critically ill patients is still controversial. The limited availability of intravenous medications, potential drug-to-drug interactions, erratic gastrointestinal absorption and the risk of unmasking infections and IRIS are considerations in the setting of critically ill patient with HIV-related disease [
37,
38,
53]. The administration of cART did not influence short- or long-term survival in our study, but we cannot conclude that cART administration in the ICU is not beneficial since it was not possible to assure adherence of all patients taking cART after hospital discharge, which is fundamental to the success of the treatment. In the present study, we did not find Immune Reconstitution Inflammatory Syndrome (IRIS) criteria in any case, nor did we find any severe adverse reactions attributed to cART.
Characteristics of tuberculosis infection
The recovery of
M. tuberculosis from blood cultures is suggestive of disseminated TB, and we found bacteremia in 14 % of patients. This is a common etiologic agent of sepsis in population with HIV-related disease, as observed in two published works by our group and other groups [
11,
29,
43‐
49]. Research suggests that patients with disseminated and extra-pulmonary forms of tuberculosis may represent a special group that may be associated with worse outcomes, including greater risk for the tuberculosis immune reconstitution inflammatory syndrome [
50]. TB treatment in the acute phase of critically ill patients can often be suboptimal [
51,
52] because first line drugs are only available in oral formulations. One could suppose that intravenous treatment could improve outcomes (fluoroquinolones and aminoglycosides). However, in this case series, we found a high proportion of critically ill patients using rifampin, isoniazid and pyrazinamide, and half of our patients received intravenous quinolones and/or aminoglycosides as the first treatment for TB with no difference in six-month survival.
Risk factors for long-term mortality
The six-month mortality observed in our study was high (52 %), but lower than reported from other retrospective studies involving patients with TB admitted to the ICU (hospital mortality 22–67 %) [
10,
41,
54]. Another interesting point was that the disseminated form of TB had a similar mortality to pulmonary presentation (45 % and 50 %, respectively). This may be different from another study, which enrolled 46 HIV-TB patients, including suspected cases of TB, although we included only confirmed cases [
10]. Finally, improvements in the care of critically ill patients, such as implementation of protective ventilator strategies with low tidal volume, conservative fluid strategy for Acute Respiratory Distress Syndrome (ARDS), and early sepsis goals, have added benefits and have recently had an impact on lower mortality rates [
28,
34,
55].
This study highlights neurological dysfunction as being an important marker for mortality among critically ill HIV-TB patients. These findings cannot be attributed to encephalitis or meningitis that is primarily caused by TB. We excluded CNS involvement (encephalitis or meningitis), which carries a poorer prognosis [
56,
57]. All patients with GCS ≤ 11 had a lumbar puncture performed. The majority of our neurological findings were mild; most of them were classified as torpor. As shown previously, low GCS at admission can be a reliable predictor of ICU mortality [
35,
58]. In our analysis, neither SOFA nor SAPS II was able to predict survival. Another two studies have found correlation between higher SAPS II and ICU mortality [
35,
51]. Zahar et al. investigated laboratory-confirmed tuberculosis cases, 38 % were patients with HIV-related, all with respiratory failure, and the majority with a pulmonary presentation; higher SAPS II (approximately 45 points) and the presence of more than one organ failure were significantly associated with mortality [
51]. Meybeck et al. showed that SAPS II was significantly higher in nonsurvivors (approximately 55 points) with an odds ratio of 1.05 per point [
35]. The SAPS II score was similar between six-month survivors and nonsurvivors in our study, while SOFA scores on the day of ICU admission trended towards being higher in nonsurvivors (
p = 0.08).