Impact of Antiretroviral Therapy on the Incidence of Tuberculosis: The Brazilian Experience, 1995–2001

Background The human immunodeficiency virus (HIV) fuels tuberculosis (TB) epidemics. In controlled clinical trials, antiretroviral therapy (ART) reduces TB incidence in HIV-infected patients. In this study we determine if, under programmatic conditions, Brazil's policy of universal ART access has impacted TB incidence among HIV-infected patients. Methods We abstracted clinical information from records of HIV-infected patients managed in the public sector in 11 Brazilian states between 1/1/1995 and 12/31/2001. Case ascertainment (TB and HIV) utilized guidelines (with added stringency) published by Brazil's Ministry of Health. We determined TB incidence and hazards ratio (HR) for ART-naïve and ART-treated [including highly active ART (HAART)] patients employing Cox proportional hazards analysis. Results Information from 463 HIV-infected patients met study criteria. The median age of the study population was 34 years, 70% were male, and mean follow-up to primary endpoints—TB, death, and last clinic visit—was 330, 1059, and 1125 days, respectively. Of the 463 patients, 76 (16%) remained ART-naïve. Of the patients who never received HAART (n = 157) 81 were treated with ART non-HAART. Of the patients who received any ART (n = 387), 306 were treated with HAART (includes those patients who later switched from ART non-HAART to HAART). Tuberculosis developed in 39/463 (8%) patients. Compared to HAART- and ART non-HAART-treated patient groups, TB incidence was 10- (p<0.001) and 2.5-fold (p = 0.03) higher in ART-naïve patients, respectively. The median baseline absolute CD4+ T-lymphocyte count for patients who developed TB was not significantly different from that of patients who remained TB free. In multivariate analysis, the incidence of TB was statistically significantly lower in HAART-treated [HR 0.2; 95% (CI 0.1, 0.6); p<0.01] compared to ART naïve patients. A baseline CD4+ T-lymphocyte count <200 cells/mm3 [HR 2.5; (95% CI 1.2, 5.4); p<0.01], prior hospitalization [HR 4.2; (95% CI 2.0, 8.8); p<0.001], prior incarceration [HR 4.1; 95% CI 1.6, 10.3); p<0.01], and a positive tuberculin skin test [HR 3.1; (95% CI 1.1, 9.0); p = 0.04] were independently and positively associated with incident TB. Conclusion In this population-based study we demonstrate an 80% reduction in incident TB, under programmatic conditions, in HAART-treated HIV-infected patients compared to ART-naïve patients.


INTRODUCTION
Tuberculosis (TB) is the leading cause of morbidity and mortality in people living with the human immunodeficiency virus (HIV) [1][2][3][4]. Patients infected with HIV are at increased risk, up to 10% per year, of reactivating latent Mycobacterium tuberculosis infection, and of accelerated progression to TB disease soon after infection [5][6][7][8].
In addition, HIV appears to increase the rate of TB re-infection and recurrent disease [9,10]. During the last decade, TB case notification rates have risen sharply in areas of high HIV prevalence, such as sub-Saharan Africa, largely attributable to the HIV epidemic [8,[11][12][13]. The World Health Organization (WHO) recommends that affected countries implement collaborative programmatic TB/HIV activities, including antiretroviral therapy (ART), as part of the health sector response to this syndemic [14,15].
Before 1996, ART consisted of an antiretroviral (ARV) agent (primarily a nucleoside reverse transcriptase inhibitor, or NRTI) given alone or in combination with another ARV agent. Thereafter, the introduction of highly active ART (HAART), consisting of three ARV agents [including NRTI, non-nucleoside transcriptase inhibitors (NNRTI) and protease inhibitors (PI)], has had a significant impact on reducing the incidence of opportunistic infections (OI) in HIV-infected patients [16][17][18][19]. In prospective clinical studies, HAART has been shown to reduce the incidence of TB by as much as 80% [20][21][22][23]. Further, epidemiological models of TB in a theoretical setting of ART accessibility have demonstrated considerable impact of HAART on TB incidence, but these estimates have relied on assumptions such as ART being initiated early in the course of HIV infection and very high levels of treatment adherence being attained [24]. The true impact of ART on TB among people living with the human immunodefi-ciency virus under normal program conditions, however, has not been previously reported.
In 1996, Brazilian federal law established a mandate granting universal access to ART for all eligible HIV-infected individuals [25]. In a nation where the high national TB burden [estimated (2005) all case TB incidence and prevalence of 60/100,00 population (pop)/yr and 76/100,000 pop, respectively] points to significant community transmission, where the adult (15-49 yrs) HIV prevalence (2005) among TB patients is 14%, and where the vast majority of TB and HIV patients are managed in the public sector, Brazil's public policy guaranteeing universal ART access has important implications for TB control [25]. According to the WHO, in 2005, 9529 new TB cases and 2281 TB deaths (all forms) were attributable to HIV, despite the fact that 86% of known HIV-infected adults eligible for ART in Brazil were receiving this therapy [26,27]. However, up to 64% of prevalent HIV infections remain undiagnosed in the country, according to estimates (2002) by Brazil's Ministry of Health (MoH), Programa Nacional de DST e AIDS (national program for STI and AIDSunpublished data). This estimate suggests that, in addition to the need for improved HIV counseling and testing access, the total burden of HIV infection in Brazil represents a large reservoir of people at increased risk of TB [20][21][22][23].
This population-based study sought to evaluate the impact of universal access to ART on TB disease incidence in HIV-infected patients managed in the public health sector in Brazil.

Study population
The study population comprised a retrospective cohort of HIVinfected individuals receiving care in public HIV treatment facilities in Brazil.
The study sample size, 474, was calculated to discern a minimum protective effect of 35% among HAART-treated compared to ART-naïve patients with a 95% confidence level and 80% statistical power [28]. In 2000, 11% of all new TB cases in adults, globally, occurred in persons infected with HIV; and we used this figure to estimate the frequency of TB among ART-naïve HIV-infected individuals in our sample size calculation [13]. The study population was drawn from systematically selected public HIV treatment facilities in Brazil using population-proportionalto-size sampling, cluster size of 20 patients, and selection of records at facilities by generating random number lists.

Inclusion Criteria
The medical records of patients were abstracted if (a) the initial facility visit was recorded to be between January 1, 1995 and December 31, 2001, and (b) HIV was diagnosed and confirmed during that period. Only an HIV Western Blot, indirect immunofluorescence testing, or evidence of viremia (by polymerase chain reaction) were accepted as confirming the diagnosis. Medical records of patients were excluded if there was evidence that the patient had attended clinic only once during the study period, were ,18 years of age, pregnant, or wards of the state.

Data abstraction
Data abstractors, trained by investigators, utilized standardized and validated data collection tools to transcribe information from patient records. Information collected included patient demographics; historical information about TB, incarceration, hospitalization, injection drug use (IDU), and homelessness; presumptive HIV transmission route; date of and vital status at last recorded facility visit; HIV-related diseases included in the WHO clinical staging system [29]; baseline and follow-up CD4+ T-lymphocyte counts and plasma viral load; tuberculin skin testing (TST) and acid-fast bacilli (AFB) smear and culture results; diagnostic x-ray and pathology reports; and, if administered, date and duration of prescribed ART, TB treatment, co-trimoxazole (CTM) prophylactic treatment, and isoniazid preventive therapy (IPT).

Antiretroviral therapy
According to the Brazil national HIV clinical guidelines, in brief, all patients, regardless of symptomatology, are eligible for ART if their CD4+ T-lymphocyte count (in Brazil determined by flow cytometry) is ,200 cells/mm 3 ; or if patients develop HIVassociated clinical manifestations such as Pneumocystis jiroveci (carinii) pneumonia [25,30]. Antiretroviral therapy is considered in asymptomatic patients with CD4+ T-lymphocyte counts between 200 and 350 cells/mm 3 according to the clinical and laboratory evolution of their immunological features (e.g., presence of OI and measured plasma viral load) and patient characteristics (e.g., adherence and motivation) [30].
The ARV agents and ART regimens received by patients between the initial facility visit and a study primary endpoint (see below) are listed in Tables 1 and 2. We classified ART as, (1) HAART: a combination of (a) 2 NRTI+1 PI, or (b) 2 NRTI+1 NNRTI, or (c) 1 NRTI+1 NNRTI+1 PI; (2) ART non-HAART: any single ARV agent or combination not defined as HAART; and (3) ART-naïve: no ART received.

Tuberculosis case definition
We classified a case of TB based on a modified (i.e., more stringent) case definition as published in the Brazil national TB guidelines as, (a) confirmed: Mycobacterium tuberculosis isolated in culture from a clinical specimen, (b) probable: AFB identified by Kinyoun or Ziehl-Neelsen staining in a clinical specimen, and (c) presumed: documented evidence of an abnormal chest x-ray consistent with TB disease, caseous granulomatous reaction in a tissue specimen, or the prescription of anti-TB treatment by a physician [31]. Pulmonary TB was defined as TB disease involving the lungs, irrespective of whether disease was present in other locations; whereas extrapulmonary TB was defined as TB disease excluding the lungs.

CD4+ T-lymphocyte count
A baseline CD4+ T-lymphocyte count was defined as the result of this test at initial facility visit; whereas, the pre-diagnostic CD4+ T-lymphocyte count was defined as the result of the test closest to the time of TB diagnosis.

Data analysis
Data analysis was performed using Epi Info v.6 and v.3.1 (CDC, Atlanta, GA, USA) and SAS v.8.01 (SAS Institute, Cary, NC, USA). Observations were censored at the date of TB diagnosis, date of last clinic visit, date of patient death, or January 1, 2002 (i.e., primary endpoints). We excluded from analysis patients in whom TB was diagnosed within 30 days of the recorded initial clinic visit because the immunorestorative effect of ART [we used the immune reconstitution syndrome (IRIS) as a marker for this] would not be expected to develop in most HIV-infected TB patients within the first 4 weeks (range 0 to12 weeks) of initiating ART [32,33].
Cox proportional hazards modeling was used to calculate hazards ratios, 95% confidence limits, and p-values in univariate screening and multivariate analysis. Variables associated with a p-value#0.2 in univariate screening were considered for the multivariate model. In the univariate screening, patients whose treatment was switched from ART non-HAART to HAART during the study period were considered in the HAART group irrespective of the amount of time they would eventually spend on HAART. In multivariate analysis, the proportional amount of time on ART non-HAART and HAART were treated as timedependent co-variables. All two-way interactions were tested. Manual backward stepwise regression was used to construct the final model. A p-value#0.05 was considered statistically significant.

Ethical considerations
The protocol for this study was approved by Brazil's MoH ethical review committee and determined as non-research by the National Center for HIV, STD, and TB Prevention of the US Centers for Disease Control and Prevention. Patient medical records kept at facilities contained routinely-collected clinical information and only this information was obtained. The study included no patient contact, and the abstraction forms and databases created for this study recorded no patient identifiers. Data abstractors were trained in issues of confidentiality and privacy by investigators.
Except for patients with a history of injection drug use (IDU; X 2 , p = 0.02), no statistically significant differences were observed between the patients who eventually developed TB and those who did not develop TB with respect to sex, age, mode of sexual transmission, or state where the patient was clinically managed (Table 3). For whom the information was available in the patient record (i.e., denominators vary), 229 and 116 of 379 patients presented with a baseline viral load of $10,000 and $100,000 copies/ml, respectively; and 263/429 (61%) presented with a baseline CD4+ T-lymphocyte ,200 cells/mm 3 . Two-hundred and twenty-three of 271 (82%) patients presented or developed WHO stages 3 or 4 clinical disease. Recorded modes of HIV transmission (not mutually exclusive) included 136/232 men who had sex with men (MSM), 61/461 with history of IDU, 232/297 with multiple sexual partners (heterosexual and MSM), 151/244 with HIV-positive partner contact, and 12/331 with prior blood transfusion.
Except for the 76/463 (17%) patients who remained ARTnaïve, all patients who initiated any ART (n = 387) did so at some point after their initial recorded facility visit. At the time of the primary endpoints, 245 [patients treated exclusively with ART non-HAART (n = 81)+those initially treated with ART non-HAART and later switched to HAART (n = 164)] and 306 [patients treated exclusively with HAART (n = 142)+those initially treated with ART non-HAART and later switched to HAART (n = 164)] patients had contributed survival time to ART non-HAART and HAART regimens, respectively (Fig. 1). One hundred-fifty seven patients received no HAART. The drug classes and specific agents that composed the most commonly

Protease inhibitors (includes ritonavir-boosted regimens)
A total of 177 specific ART regimens were cumulatively used for at least one day, 1421 times during the study period , , Note that the number of ART regimens cumulatively used during the study period exceeds the total study population because, for those on ART, more than one regimen may have been prescribed during the study period doi: 10 prescribed ART regimens in this study are listed in Table 2. At least one dose of CTM was prescribed for 204/441 (46%) patients.
Mean duration of follow-up from the initial recorded facility visit to a primary endpoint was 330, 1059, and 1125 days for TB diagnosis , patient death, and last clinic visit, respectively. Of the 43 (9% of study population) patient deaths recorded during the study period, 3 had been diagnosed with TB disease (X 2 : p.0.05).
We analyzed potential risk factors for TB disease in the study population (Table 4). Patients with a baseline CD4+ T-lymphocyte count of ,200 cells/mm 3 had more than twice the incidence of TB disease than those with a baseline CD4+ T-lymphocyte count .200 cells/mm 3 {hazards ratio (HR) 2.3; [95% confidence interval (CI)] 1.2, 4.4}. A history of IDU, incarceration, hospitalization, and prior TB were also associated with a statistically significantly higher incidence of TB disease (Table 4). No statistically significant difference was observed in the incidence of TB disease among patients with baseline (i.e., at initial visit to the facility) WHO clinical stages 3 and 4 vs. 1 and 2, [HR 1.4; (95% CI 0.5, 3.6)].
In univariate screening (Table 4), ART-naïve patients had a statistically significantly higher incidence of TB disease than those receiving any ART [HR 5.0; (95% CI 3. In multivariate analysis, controlling for type of ART intervention, a baseline CD4+ T-lymphocyte count ,200 cells/mm 3 , IPT, and prior history of TB, hospitalization, incarceration and IDU (  1, 9.0); p = 0.04] were independently and positively associated with incident TB. All two-way interaction terms were tested and none were found to be statistically significant (i.e., p.0.1).
Ninety-five percent, 86%, and 69% of patients who received HAART, ART non-HAART , and remaining ART naïve, respectively, remained free of TB disease at primary endpoints (p,0.001 for all three situations, when evaluated together; see Figure 2).  The mortality rate of the study population at study end was 0.24 per 100-person years of follow-up. The proportional mortality of severely immunocompromised TB patients (CD4+ T-lymphocyte,200 cells/mm 3 ) at 6 and 12 months after starting TB treatment, was 0.8% and 1.1%, respectively.

DISCUSSION
In this population-based retrospective cohort analysis of adult HIV-infected patients managed under normal programmatic conditions in public HIV treatment facilities in Brazil, we observed a statistically significantly lower (80%, p,0.01) incidence of TB in patients prescribed HAART compared to patients who remained ART-naïve. This effect of HAART remained statistically significant even when controlling for the patients' baseline immunological status, and other important characteristics and interventions (Tables 4 and 5). Though treatment with ART non-HAART also resulted in a lower TB incidence rate, the reduction did not reach statistical significance (p = 0.08). The findings in this study serve to validate similar findings in previous, more controlled studies, and demonstrate that it is possible for national programs to provide access to HAART under ''real world'' conditions, and in doing so, reduce the risk of the most common OI of adults living with HIV-TB [19][20][21][22].
In Brazil, it is customary for HIV treatment facilities to take over the care of newly diagnosed HIV-infected patients in the public sector, and thereafter, provide longitudinal comprehensive clinical management and ART, as indicated [25]. According to the WHO, 86% of diagnosed and eligible HIV-infected patients in Brazil were receiving ART in 2004, a proportion superior to that of low and middle income countries (e.g., Peru, 18%); and similar to that of developed countries in South America such as Argentina and Chile (both 100%) [26]. However, while access to ART remains very important, a significant impact in the reduction TB incidence for people living with HIV is unlikely unless a large proportion of the subset who are unaware of their HIV serostatus are tested, and national programs are able to deliver high quality comprehensive clinical care, including effective provision of ART to this population.
The patients who developed TB disease in our study had similar risk factors for TB as HIV-infected patients in other settings, and  Denominators vary due to missing data from the patient medical record for characteristic (missing values deleted from analysis) Reference value appears bolded on the second line of each characteristic. Note that those patients who received ART non-HAART and were later switched to HAART were classified as patients having received HAART (see data analysis section of Methods)^T he word ''Total'' in the ''No. with feature/Total'' column refers to the denominator for each particular characteristic (e.g., 81 out of a total of 157 patients who did not receive HAART were treated with ART non-HAART exclusively) # The total number of patients who developed TB (n = 39) varies for each characteristic according to whether or not patients were excluded from analysis due to incomplete information (in the patients' medical record) for that characteristic doi: 10 66% eventually received HAART [1,8,11,22,34,35]. Nevertheless, 11 (4%) patients developed TB disease despite receiving HAART. A higher proportion of patients with TB had median baseline CD4+ T-lymphocyte counts ,200 cells/mm 3 compared to patients without TB (57% vs. 37%, respectively. X 2 , p = 0.01).
This suggests that those who developed TB may have been more immunocompromised at baseline than patients remaining TB free; however, the median baseline absolute CD4+ T-lymphocyte count and baseline WHO clinical staging characteristics were not significantly different between the two groups. We cannot exclude the possibility, however, that the study lacked the power to detect subtle but clinically important immunological differences among groups. In areas of high TB prevalence, TB can occur at all levels of CD4+ T-lymphocyte count, though ART-naïve patients who develop TB frequently present with low CD4+ T-lymphocyte counts prior to TB diagnosis [36,37]. Both HAART and non-HAART regimens have been shown to raise CD4+ T-lymphocyte counts and improve the clinical status of HIV-infected patients with and without concomitant OI [20,38,39]. This suggests that other factors may be affecting TB disease risk after initiating ART. Indeed, evidence exists to suggest that HAART is unable to fully restore the immune response to Mycobacterium tuberculosis in HIVinfected patients, leading to a chronically heightened TB risk over an increased lifespan [39][40][41]. Addressing this heightened risk poses a challenge for the future control of TB; however, growing evidence also suggests that this risk may be mitigated by early initiation of HAART with corresponding high levels of treatment adherence [24,41,42]. Fortunately, currently and widely recommended clinical guidelines for the treatment of HIV infection and TB disease are helping to eliminate an important negative factor from the TB/HIV equation-suboptimal ART. Because of the timeframe selected for this study, 1995-2001, it was not Finally, there is now reliable information showing that HIVrelated morbidity and mortality in adults and children, including that due to TB disease co-morbidity, can be significantly reduced with the adjunctive use of CTM prophylaxis [46][47][48]. We found that the mortality rate and overall mortality of the cohort and of severely immunocompromised TB patients was comparable, if not lower, than that observed for HIV infected patients managed and reported in high income countries after 1995 [16,17,49]. And, while many factors clearly confound any mortality considerations in this study, the fact that 46% and 33% of the cohort and of TB patients, respectively, in the study received at least one dose of CTM probably contributed to favorable outcomes in these patients. As is probably the case in many other clinical settings globally, the use of CTM prophylaxis among HIV-infected populations with TB can clearly be improved and, in doing so, independently reduce patient morbidity and mortality.
Our study population specifically excludes children (age,18years) and pregnant women. Due to the retrospective and observational design of this study, we were also unable to control for factors specific to (a) patients (e.g., adverse nutritional and psycho-social issues); (b) facilities (e.g., infection control measures); and, (c) programs [e.g., erratic pharmacy drug stocks and diagnostic rigor (e.g., low sensitivity and specificity of X-ray imaging)]. In addition, patients with only one clinic visit were explicitly excluded from the study, and we were unable to consistently determine the degree of patient adherence to therapeutic regimens from patient records. With regards to treatment adherence, however, WHO and others estimate adherence to be approximately 75% among patients on ART in Brazil [27,50].
An additional limitation of our study was that the majority of the TB diagnoses recorded were made, primarily, on clinical grounds (i.e., presumed); and therefore, an alternative diagnosis to TB cannot be excluded. As the most common opportunistic disease of HIV-infected patients, however, the clinical experience of health care staff in facilities with large HIV patient populations probably lessens this possibility. In addition, TB misdiagnosis would only tend to underestimate the degree of TB incidence reduction by HAART.
Despite the limitations of the study and the fact that a proportion of study patients received sub-optimal ART before being switched to HAART, the significant reduction in TB disease observed is all the more remarkable. Our findings, however, are not applicable to pediatric patients, and should be interpreted in light of the study limitations.
We conclude that, in settings of unrestricted access to ART, HAART can have a significant impact on the TB incidence of HIVinfected patients developing. We found that, even under programmatic conditions, a statistically significant reduction in TB incidence of 80% was observed among HIV-infected patients receiving HAART in the public sector in Brazil compared to patients who remained ART-naïve. We demonstrate that, although broad access to ART remains an elusive goal of low and middle income countries, a national policy that drives the effective implementation of a program that manages HIV-infected patients using HAART is an initiative that is not only beneficial in reducing TB disease in HIV-infected patients, but also a feasible one.