Directly observed therapy and risk of unfavourable tuberculosis treatment outcomes among an international cohort of people living with HIV in low‐ and middle‐income countries

Abstract Introduction Identification of persons living with human immunodeficiency virus (HIV)‐associated tuberculosis (TB) at increased risk for unfavourable TB outcomes would inform efforts to improve such outcomes. We sought to identify factors associated with a decreased risk of unfavourable TB treatment outcomes among people living with HIV‐infection (PLHIV) in low‐ and middle‐income countries (LMIC), with a specific focus on directly observed therapy (DOT) compared with self‐administered therapy (SAT) during the continuation phase of anti‐TB therapy. Methods We conducted a retrospective cohort study among adults diagnosed with HIV‐associated TB in Africa, Asia and the Americas from 2012 to 2013; data were collected from 2012 to 2016. Unfavourable TB treatment outcomes (death during TB treatment, and TB treatment failure or recurrence) were defined according to World Health Organization criteria. Receipt of DOT was obtained at the site level and defined as ≥5 days of DOT per week. The person administering DOT and treatment location varied by site. Lack of receipt of DOT was defined as SAT. Multivariable logistic regression estimated the adjusted odds of unfavourable TB treatment outcomes. Results Among 1862 adults with HIV‐associated TB included, 252 (13.5%) had unfavourable TB outcomes (226 deaths, 26 recurrences/failures). Overall, 1825 (98%) received DOT in the intensive phase and 1617 (87%) received DOT in the continuation phase. DOT in the continuation phase was not significantly associated with unfavourable TB outcomes (aOR 1.43, 95% CI 0.86 to 2.38) compared to SAT. Body mass index (BMI) change during anti‐TB treatment (per 2 units increase, aOR 0.74, 95% CI 0.68 to 0.82) and CD4+ count at TB diagnosis (200 vs. 50 cells/µL, aOR 0.54, 95% CI 0.39 to 0.73) were both independently associated with decreased odds of unfavourable TB treatment outcomes. Conclusions In this large, international cohort of people living with HIV‐associated TB in LMIC who received intensive phase DOT, DOT during the continuation phase of anti‐TB therapy was not associated with a decreased odds of unfavourable TB treatment outcomes compared to SAT. Randomized trials evaluating the effect of continuation‐phase DOT on TB outcomes among PLHIV are needed.


| INTRODUCTION
Tuberculosis (TB) remains a serious public health problem, particularly among people living with human immunodeficiency virus (PLHIV). TB is the leading cause of death among PLHIV worldwide, and people with HIV-associated TB have higher rates of other unfavourable TB treatment outcomes, including TB recurrence or TB treatment failure, that require re-treatment. In 2016, TB treatment success rates were 82% overall but only 77% for HIV-associated TB [1]. TB re-treatment, in turn, is associated with even higher rates of unfavourable TB treatment outcomes (including death, recurrence and failure) [2,3], drug resistance [4,5], and more complicated and costly drug regimens.
Standard therapy for patients without a prior history of TB consists of a 2-month intensive phase including isoniazid, rifampicin, pyrazinamide and ethambutol (HRZE) followed by a 4-month continuation phase including isoniazid and rifampicin (HR) [6]. Suboptimal adherence has been associated with unfavourable TB treatment outcomes [7] and the World Health Organization's End TB Strategy emphasizes patient-centred treatment supervision [8]. Therefore, directly observed therapy (DOT) has been considered the standard of care for TB treatment, regardless of HIV status [9]. However, two previous systematic reviews [10,11] and one meta-analysis [12] did not reveal an association between DOT and decreased risk of TB recurrence or failure compared to self-administered therapy (SAT). Most studies included in these analyses either excluded HIV-positive persons or only included participants with unknown HIV test results.
While DOT in the two-month intensive phase is associated with a decreased risk of TB relapse [13], we are unaware of studies assessing the effect of DOT specifically during the longer continuation phase of therapy. DOT is often not provided in low-and middle-income countries (LMIC) [14] as it is less feasible due to limited staffing, increased cost and transportation difficulties compared to higher income countries. The objective of this study was to identify factors associated with unfavourable TB treatment outcomes including mortality, TB treatment failure and recurrence among PLHIV in a large, international cohort collaboration. We specifically sought to determine if DOT during the continuation phase of therapy was associated with a decreased risk of unfavourable TB treatment outcomes compared to SAT.

| Study population
We conducted a retrospective cohort study among PLHIV aged ≥16 years from the International Epidemiology Studies to Evaluate AIDS (IeDEA) cohort collaboration (http://www.iedea.org/ ), a consortium of ART treatment programmes predominantly established in low-and middle-income countries (LMIC). Sites were located in 18 countries and included a mix of healthcare facility types: referral hospitals, regional hospitals and primary care facilities. Consecutive participants were included if they were diagnosed with HIV-associated TB between 1 January 2012 and 31 December 2013. Patients were followed until death, TB failure/recurrence or administrative censoring on 31 December 2014 in order to capture ≥12 months of follow-up per individual. Patients who died within eight weeks of TB diagnosis were excluded as it may not have been possible to initiate ART within eight weeks, an important covariate of interest. Patients were excluded if they had documented resistance to isoniazid or rifampicin, were not initiated on the standard HRZE regimen for patients without a prior history of TB, or had unevaluable TB treatment outcomes ( Figure 1).

| Ethical considerations
Independent Ethics Committee (IEC) or Institutional Review Board (IRB) approval for this study was obtained by each of the local IeDEA sites as well as coordinating centres (see Data S1).

| Data collection
A standardized case report form (CRF) designed for this study was used to collect data on patient demographics, vital status, laboratory data, TB treatment regimens and TB treatment outcomes. The CRF was developed in Research Electronic Data Capture (REDCap) and was available in English, French and Spanish [15]. Local IeDEA site investigators completed a CRF for each TB case after medical record review from January 2012 to January 2016. Routine data audits for quality assurance were completed by the principal investigator (PI). Information on ART initiation dates were obtained from the electronic regional IeDEA data repositories.

| Study definitions
TB case definitions and treatment outcomes were defined according to WHO definitions [16]. Our primary outcome of interest was a composite unfavourable endpoint of death during TB treatment and TB treatment failure or recurrence, as both of the latter two endpoints would result in the need for TB re-treatment. TB recurrence was defined as re-initiation of therapy for a diagnosis of TB following a TB episode with an outcome of cure or treatment completion. TB cure was defined as a case of bacteriologically confirmed TB at the beginning of treatment with negative smears or cultures in the last month of treatment and on ≥1 previous occasion. TB treatment completion was defined as a case for whom treatment was completed, but without record of smear or culture results. TB treatment failure was defined as a TB case for whom sputum smear or culture was positive at five months or later during TB treatment. Unevaluable TB treatment outcomes included those lost to follow-up or without an assigned TB treatment outcome.
Our primary exposure of interest was receipt of TB treatment by DOT during the entire continuation phase of therapy. DOT data were obtained at the site-level and applied to each individual patient receiving care at the site. Receipt of DOT was defined as ≥5 days of DOT per week; <5 days of DOT was defined as SAT. DOT was defined according to local guidelines with respect to the person administering therapy (family member/friend vs. healthcare worker) and the location of treatment (home/community-based vs. healthcare facilitybased).
Covariates identified for inclusion in multivariable models a priori included age at TB diagnosis, sex, baseline body mass index (BMI), change in BMI from the date of anti-TB treatment start to the date of anti-TB treatment stop, smear status, culture status, bacteriologic TB confirmation, site of disease (extrapulmonary, pulmonary/both/unknown), CD4 + count at TB diagnosis, IeDEA region, year of TB diagnosis and ART initiation timing in relation to TB diagnosis. Bacteriologically-confirmed disease was defined as a positive acid fast bacillus (AFB) smear, culture, Xpert MTB/RIF or other nucleic acid amplification test (NAAT). Extrapulmonary TB (EPTB) was defined as a case of TB involving organs other than the lungs. Pulmonary TB (PTB) was defined as a case of TB involving the lung even when extrapulmonary disease was also present [1]. The CD4 + lymphocyte count at TB diagnosis was defined as the nearest available value 180 days before or up to 30 days after the date of TB diagnosis.
ART was defined as ≥3 antiretroviral drugs including nucleoside reverse transcriptase inhibitors (NRTIs), protease inhibitors (PIs), non-nucleoside reverse transcriptase inhibitors (NNRTIs), or integrase inhibitors. Early ART initiation was defined as initiation within eight weeks of TB diagnosis. This ART initiation timing cutoff was chosen to be consistent with the WHO guidelines recommending ART initiation within eight weeks of TB diagnosis dating back to 2009 [6]. TB diagnosis date was defined as the date of anti-TB therapy initiation.

| Statistical analysis
Categorical and continuous variables were compared using the chi-squared and Wilcoxon rank sum tests respectively. The association of DOT with the composite outcome was assessed using unadjusted and adjusted logistic regression in order to estimate the odds of unfavourable TB outcomes compared to SAT. Continuous predictor variables were modelled using restricted cubic splines with three knots. Covariates with missing data (excluding missing data for the combined primary outcome) were imputed once using predictive mean matching [17]. Subjects with missing outcome data were assumed to have been alive and event free until the end of follow-up. All p-values were two-sided and considered statistically significant if <0.05. R version 3.4.2 (www.r-project.org) was used for analysis and analysis scripts are available online (http://biostat. mc.vanderbilt.edu/ArchivedAnalyses).

| Sensitivity analyses
We identified a priori two important sensitivity analyses to perform. In the first, we explored the most conservative assumption that all persons with missing vital status had died.
In the second, only participants with baseline CD4 + counts <50 cells/µL who survived at least two weeks after TB diagnosis were included, and the ART initiation timing covariate cutoff was changed from eight weeks to two weeks. This sensitivity analysis was done to be consistent with results of the STRIDE study which demonstrated that ART initiation within two weeks of TB diagnosis for patients with HIV-associated TB and a baseline CD4 + count <50 cells/µL was associated with a lower rate of AIDS defining illnesses and death [18].
For the second sensitivity analysis, inverse probability weighted propensity score methods [19] were utilized rather than standard logistic regression to avoid overfitting the models due to a low event rate. We adjusted for confounders by weighting observations using inverse probability weights (IPW); specifically, the inverse of the probability of the observed exposure level given the observed values of covariates (confounders). Weighting by this inverse probability created a pseudo-population in which the covariates of interest were balanced. We performed this analysis for our primary exposure of interest (DOT) as well as our other covariates of interest.
Standardized mean difference (SMD) plots [20] were used to assess the appropriateness of the weights. Standardized  difference in means was used to compare balance in measured variables between exposed and unexposed subjects. For continuous exposures, the groups were divided at the median to calculate SMDs; for >2 categories, the groups were divided by the most prevalent versus the rest to calculate SMDs. An SMD equal to 0.10 means that the difference in means is within 10% of the pooled standard deviation.

| RESULTS
There were 1862 adults aged 16 years or older diagnosed with HIV-associated TB included in this analysis. Participants were from 18 countries in 5 IeDEA regions (Caribbean, Central, and South America network for HIV epidemiology (CCA-SAnet), Western Africa, Central Africa, Eastern Africa and the Asia-Pacific) ( Figure 2). The clinical and demographic characteristics of the study population are in Table 1. Overall, 252 (13.5%) adults with HIV-associated TB developed unfavourable TB outcomes. This included 226 deaths during TB treatment (12.1%) and 26 TB recurrences or treatment failures (1.4%). Among 1351 (72.6%) PLHIV not on ART who survived at least eight weeks after TB diagnosis, 477 (35%) were started on ART within eight weeks and 874 (65%) were started on ART >8 weeks after TB diagnosis or not started on ART before the end of follow-up. PLHIV who had unfavourable TB treatment outcomes were more likely to be older at the time of TB diagnosis (38 vs. 36 years, p = 0.012) and have lower median CD4 + counts at the time of TB diagnosis (74 vs. 123 cells/µL, p < 0.001); they were less likely to gain weight over the course of TB treatment (0 vs. 2 kg, p < 0.001). There were also regional differences, with those who had unfavourable treatment outcomes being more likely to reside in the CCA-SAnet or Central and Eastern Africa regions ( Table 1).
The details regarding DOT at each site with respect to the number of days per week of DOT in the intensive phase, the number of days per week of DOT in the continuation phase, the person giving DOT, and the location of DOT are in Table 2. There were 37 (2%) PLHIV in care at one site who did not receive DOT during the intensive phase and 245 (13%) PLHIV in care at seven sites who did not receive DOT during the continuation phase of TB treatment. Only 13 (5%) of those categorized as not receiving DOT during the continuation phase of therapy (<5 days per week), received thrice weekly DOT; the remainder (95%) did not receive any DOT during the continuation phase of therapy.
In the adjusted models, continuation phase DOT was not significantly associated with a decreased odds of unfavourable TB treatment outcomes (aOR 1.43, 95% CI 0.86 to 2.38). Both a higher CD4 + count at TB diagnosis (200 vs. 50 cells/ µL, aOR 0.54, 95% CI 0.39 to 0.73) and BMI increase during anti-TB treatment (2 vs. 0 kg/mg 2 , aOR 0.74, 95% CI 0.68 to 0.82) were associated with a decreased odds of unfavourable TB treatment outcomes in the multivariable model. There was no association with the timing of ART initiation and TB treatment outcomes among those not already on ART at the time of TB diagnosis (Table 3).
In the first sensitivity analysis, 31 (1.7%) persons with missing vital status were assumed to have died and results were largely unchanged (see Table S1). In the second sensitivity analysis, 468 (25.1%) persons with CD4 + counts ≤50 cells/µL at TB diagnosis were included (Table 4). Of the 364 persons not on ART who survived at least two weeks after TB diagnosis, 49 (13%) were started on ART within two weeks of TB diagnosis and 315 (87%) were started on ART >2 weeks after TB diagnosis or not started on ART before the end of followup. The adjusted model using IPW indicated a significant association between BMI change during anti-TB treatment and a decreased odds of unfavourable TB treatment outcomes (aOR 0.60, 95% CI 0.49 to 0.73 for each 2 kg/m 2 increase in BMI).

| DISCUSSION
In this study conducted within a large international HIV cohort collaboration, DOT during the continuation phase of TB treatment was not significantly associated with a decreased odds of unfavourable TB treatment outcomes (TB death, TB recurrence and TB treatment failure). Only 2% of this study population did not receive intensive phase DOT. It is possible that among a population with a very high frequency of DOT in the intensive phase, that DOT during the continuation phase does not affect TB treatment outcomes. There also could have been confounding by indicationthat persons at increased risk of unfavourable TB treatment outcomes preferentially received continuation phase DOT. A previous randomized trial conducted in Pakistan did not reveal an association between DOT and TB outcomes when given only during the intensive phase, compared to SAT throughout the entire course of anti-TB treatment [21].
Several previous studies have attempted to evaluate the impact of DOT on TB treatment outcomes but did not focus on the population of PLHIV, did not focus on DOT in the continuation phase of anti-TB therapy, and did not use a composite endpoint of unfavourable TB treatment outcomes (death, failure and recurrence). In three prospective studies (two conducted in Thailand and one in Taiwan), DOT was shown to be associated with a decreased risk of TB treatment default [22][23][24] and increased rates of TB treatment cure [23,24]. One of these studies also showed a trend towards a decreased risk of TB treatment failure with DOT compared to SAT [23]. DOT has not been shown to be associated with decreased risk of TB relapse [25][26][27], although these studies were also limited by low numbers of relapse events.
The rate of unfavourable TB treatment outcomes in this population of persons with HIV-associated TB who survived at least eight weeks was 13.5%; the TB mortality rate was 12.1% and the TB recurrence or treatment failure rate was 1.4%. Globally in 2017, the WHO estimated that the TB case fatality rate among people living with HIV was 32.6%. Moreover it is estimated that 23% of TB notifications among PLHIV had unfavourable treatment outcomes [1]. It is possible that the rates of unfavourable outcomes in this population are lower than global estimates because this population was enrolled from a cohort of PLHIV who were in care in an ART programme. It is also likely that the exclusion of persons who  died within the first eight weeks of TB treatment (n = 106, 32%) and/or the possibility that some deaths were missed contributed to the lower TB mortality rate in this study population.
The majority of patients not on ART at the time of TB diagnosis (65%) in our study who survived at least eight weeks after TB diagnosis, were not started on ART within eight weeks. Among those with CD4 + counts <50 cells/µL at TB diagnosis, even a higher proportion who survived at least two weeks after TB diagnosis (87%) were not started on ART within two weeks. This finding is consistent with that of a previous study in South Africa, which found that only 51.1% initiated ART within eight weeks of TB diagnosis. These data suggest that barriers to expedited ART initiation exist and that improved integration of TB and ART services may lead to improvements in ART initiation delays [28].
We did not find an association with the timing of ART initiation and TB treatment outcomes among those not already on ART at the time of TB diagnosis. Several clinical trials have shown that earlier ART initiation in relation to TB diagnosis is associated with improved survival [18,29,30]; however, there are several significant differences between those studies and our study. Only one of these trials, the STRIDE study, allowed enrolment of persons with smearnegative TB as our study did. The STRIDE study used a primary endpoint of AIDS-defining illness and death. Of their 116 endpoints, 63 (54%) were due to AIDS-defining illnesses and 53 (46%) were due to death [18]. In contrast, our study used a primary endpoint of TB recurrence or treatment failure and death. Of our 252 endpoints, 26 (10%) were due to TB recurrence or treatment failure and 226 (90%) were due to death. It is possible that these differences in eligibility criteria and primary outcome definitions led to our differing findings with respect to the timing of ART initiation.
An increase in BMI during TB treatment was associated with a decreased risk of TB recurrence or treatment failure. This finding is consistent with the findings from Tuberculosis Trials Consortium Study 22 which found that participants who were underweight at enrolment and had a <5% weight gain over the intensive phase of anti-TB treatment had nearly twice the risk of relapse compared to those who gained more than 5% [31]. It is unclear if the association between lack of weight gain and TB recurrence or treatment failure is due to a poor response to anti-TB treatment stunting the rate of weight gain, or if the presence of adequate weight gain is required in order for patients to respond appropriately to anti-TB therapy. Regardless, the WHO recommends nutritional evaluation, counselling and support as part of TB care [32].
The main limitation of this study is that we did not have detailed data at the individual patient level with respect to adherence either in the setting of DOT or SAT. Moreover we did not have individual level data on the person administering DOT (family member/friend vs. healthcare worker) or the location of DOT (home/community-based vs. healthcare facility-based). However, a previous meta-analysis found no difference in TB treatment outcomes between community-observed versus family-observed and between home-observed versus healthcare facility-observed DOT [10]. Another limitation was that the number of TB recurrences and treatment failures was low, which decreased the power of the study. While we did not collect information regarding whether a smear and/or culture at months 5 or later during TB treatment was performed or not, we do know that 22% of cases did not have a smear performed at any time during TB treatment and 86% of cases did not have a culture performed at any time during TB treatment. So, misclassification of TB treatment failure as another outcome is possible due a lack of smear and/or culture for TB diagnosis. Additionally, it is possible that some of the cases that transferred care to another facility or defaulted may have actually had an unfavourable TB outcome including death, recurrence or failure. Moreover it is possible that some recurrent cases were not identified during the 12-month follow-up. Two studies reporting combined results from TB treatment trials suggest that 12 months of follow-up is sufficient in order to capture most TB recurrences [33,34]. However, it may take longer to identify recurrent TB cases in programmatic settings in which follow-up is less active compared to clinical trials. Additionally, it is possible that some patients moved out of the local IeDEA site catchment area which would also contribute to underascertainment of recurrent TB cases.
Another limitation was the lack of data on medical comorbidities known to be associated with unfavourable TB outcomes. One important example, includes diabetes mellitus (DM). In a recent systematic review, patients with TB and comorbid DM had a higher odds of death and relapse [35]. Future studies evaluating the impact of factors associated with unfavourable TB outcomes should include DM and other important medical comorbidities.

| CONCLUSIONS
In this large, multi-centre, international study among individuals with HIV-associated TB who received intensive phase DOT, we found that DOT given during the continuation phase of TB treatment was not associated with a decreased odds of unfavourable TB treatment outcomes. A higher CD4 + count at TB diagnosis was independently associated with a decreased odds of unfavourable TB treatment outcomes, highlighting the need for early HIV testing and treatment. Increasing BMI during anti-TB treatment was also associated with a decreased odds of unfavourable TB treatment outcomes supporting the importance of nutritional evaluation and support during anti-TB therapy. Randomized trials evaluating the impact of DOT during the continuation phase of therapy on TB outcomes among PLHIV are needed.

C O M P E T I N G I N T E R E S T
All authors report no conflicts of interest related to this work.