Brazilian cohort study of risk factors associated with unsuccessful outcomes of drug resistant tuberculosis

Background Treatment outcomes were evaluated of a cohort of new pulmonary tuberculosis (TB) cases that were rifampicin resistant, multidrug-resistant, or extensively resistant during 2013 and 2014 in Brazil. The objective of this study is to identify factors associated with unfavorable treatment outcomes for drug-resistant TB cases. Methods The Brazilian Special Tuberculosis Treatment Information System (SITE-TB) was the main data source. The independent variables were classified into four blocks (block I: individual characteristics; block II: clinical characteristics and proposed treatment; block III: treatment follow-up characteristics; and block IV: TB history). The category of successful therapeutic outcome was compared with lost to follow-up, failure, and death. Considering the multiple outcomes as the dependent variable, the odds ratios (OR) and its respective 95% confidence interval (95% CI) were estimated by multinomial logistic regression. Results After applying the exclusion criteria, 980 (98.8%) individuals were included in the study. Of these, 621 (63.4%) had successful treatment, 163 (16.6%) lost to follow-up, 76 (7.8%) failed, and 120 (12.2%) died. Important factors associated with lost to follow-up in the final model included use of illicit drugs (OR = 2.5 95% CI: 1.57–3.82). Outcome failure was associated with having disease in both lungs (OR = 2.0; 95% CI: 1.09–3.62) and using more than one or not using injectable medication (OR = 2.8; 95% CI: 1.05–7.69). Major factors for the death outcome were at least 60 years old (OR = 3.4; 95% CI: 1.90–6.03) and HIV positive (OR = 2.7; 95% CI: 1.45–4.83). Conclusions The factors associated with unfavorable treatment outcomes were different. Some of these factors are specific to each outcome, which reflects the complexity of providing care to these individuals. Supplementary Information The online version contains supplementary material available at 10.1186/s12879-021-06756-7.


Introduction
Tuberculosis (TB) is one of the top 10 causes of death worldwide. In 2019, about 10 million people fell ill and 1.2 million died from TB [1]. According to Falzon et al. [2], multidrug-resistant (MDR) TB, which is defined as resistance to rifampicin (R) and isoniazid (H), plays a vital role in global TB control. In 2019 worldwide, 465 thousand people developed rifampicin resistant (RR) TB diagnosed by GeneXpert MTB/RIF ® (RR) of which 78% had MDR-TB. Only 206 thousand (44%) were diagnosed and reported, and 177 thousand (38%) started treatment with second line drugs [1].
In Brazil, the strategies to contain the main causes associated with the development of drug-resistant TB (DR-TB) in the community include free treatment for Open Access *Correspondence: patricia.bartholomay@gmail.com 1 Tropical Disease Post-Graduation Program, University of Brasilia, Brasília, DF, Brazil Full list of author information is available at the end of the article

Study design and population
This analytical study, of the historical cohort type, studied a population of new cases with pulmonary and mixed TB (pulmonary and extrapulmonary), starting treatment between 2013 and 2014 and with the following patterns of drug resistance at the beginning of treatment: RR-TB, MDR-TB, and TB with extensive resistance (XDR). All individuals included in the study had laboratory-confirmed tuberculosis. The diagnosis of resistance was made by sensitivity test or GeneXpert MTB/RIF ® . Cases were excluded who were less than 15 years old, without definitive treatment reported, or were duplicated.

Data source
The SITE-TB was the main source of data [11]. The history of TB treatment that occurred before the notification of SITE-TB was retrieved from the Information System on Notifiable Diseases (Sinan). The treatment outcomes were qualified with the death records in the Mortality Information System (SIM), through the probabilistic relationship of the database [12].
In Brazil, all TB cases must be reported in the Sinan. Cases diagnosed with DR-TB should be concluded in the Sinan and notified in the SITE-TB to monitor treatment [3]. Since the implementation of SITE-TB in 2013, all cases under treatment with special regimens, that is, different from the basic treatment regimen, must be recorded in this system. SITE-TB is an online information system. All information until the outcome is registered on the system, which makes it possible to follow patients even if they switch the health facility during the treatment [11].

Study variables
Independent variables were classified into four blocks (block I: individual characteristics; block II: clinical characteristics and proposed treatment; block III: treatment follow-up characteristics; and block IV: TB history).
Individuals were divided into two age categories: 15 to 59 and over 59 years old. The " no " and " do not know " categories of the variables on diseases and associated conditions (alcoholism, diabetes, smoking, use of illicit drugs, and others) were grouped together forming the " no/do not know " category. Confirmation of alcoholism or smoking was defined by the physician. The use of illicit drugs was reported by the patient.
The HIV test variable was classified as yes, no, and missing (no test performed or result not recorded). Education was categorized into the following categories: up to seven years of education, eight or more years, and information not provided.
In block II, the initial scheme type was classified as standardized scheme (cases that started treatment using schemes define by the MoH recommendations [3]) and individualized scheme (for those cases that started treatment with any scheme different from the standardized schemes).
The fluoroquinolone treatment was defined as: used only levofloxacin or moxifloxacin, used only ofloxacin, and used more than one fluoroquinolone or did not use fluoroquinolone. The injectable drug treatment was categorized as: used only amikacin, used only streptomycin, used only capreomycin, and used more than one injectable or did not use injectable drugs. More than 80% of the cases identified with RR-TB also exhibited resistance to isoniazid, with the MDR resistance pattern confirmed after the sensitivity test [13]. Therefore, we classified the categories of the initial resistance pattern into MDR/RR and XDR.
In Block IV, the previous tuberculosis events variable classified the number of previous tuberculosis events into two categories: up to three events and four or more events. The time between the first diagnosis of tuberculosis and the start of drug-resistant tuberculosis treatment was divided into three categories: up to one year, between one and three years, and three years or more.
The classification, categories, and description of each variable are provided in the Additional file 1: Table S1.

Data analysis
To identify the factors associated with unfavorable outcomes in the treatment of DR-TB, records with unreported information were excluded when they represented less than 5% of all records. When the percentage was 5% or more, a category called " unreported " was created. The outcome category " success treatment " was used as a reference for the variable response of treatment outcome and compared with the other categories ( " lost to followup " , " failure " , and " death " ). Cases with lost to follow-up are those who used the drugs for 30 days or more and stopped the treatment for 30 consecutive days or more. Failure was defined in the following situations: persistence of smear positive sputum at the end of treatment; cases who had strongly positive smear (+ + or + + +) at the beginning of treatment and maintained this situation until the fourth month; initial positive smear followed by negative; and new positive results for two consecutive months, from the fourth month of treatment. Cases that died from tuberculosis or from other causes during the treatment were considered in the outcome death. For the multiple outcomes of the dependent variable, the odds ratios (OR) and their respective 95% confidence interval (95% CI) were estimated by multinomial logistic regression.
Initially, we included in the model all variables with a statistically significant association (α = 0.10) in the bivariate analysis of multinomial regression. Using the backward strategy, only the statistically significant variables were kept in the final model, considering the value of the significance level α = 0.05. The final model adequacy was evaluate with goodness-of-fit test by Hosmer-Lemeshow [15].
The final model was graphically represented by the OR logarithm and respective interval confidence of 95% (95%CI). The logarithm was chosen to represent the risk and protection factor on the same scale. A

Results
After excluding three cases under 15 years old, five duplicate records, and 17 records without the final information, 980 laboratory-confirmed cases of pulmonary RR/ MDR/XDR-TB reported in the years 2013-2014 (98.8%) were included in the study (Additional file 1: Figure S2). Of these, 621 (63.4%) were successfully treated, 163 (16.6%) were lost to follow-up (LFU), 76 (7.8%) failed, and 120 (12.2%) died (Tables 1, 2 The percentage of LFU was higher in men (18.4%) and the age group of 15-59 years (17.8%). For associated diseases, the percentage of LFU was higher in individuals with reported alcoholism (22.4%), no diabetes or missing information about diabetes (17.8%), smoking (19.5%), and use of illicit drugs (32.1%) ( Table 1). For the variables in block II, the absence of cavitation (62.9%), bilateral disease (68.9%), primary resistance type (71.9%), and MDR/ RR initial resistance pattern (63.8) presented the best treatment results ( Table 2). In block III, the highest indicators for death were the same municipality of residence and treatment (12.9%), recorded unfavorable clinical evolution (18.1%), and no recorded regimen change (13.1%) ( Table 3). Individuals who had up to three previous TB events registered on the Sinan exhibited 64.7% treatment success ( Table 4). The nonadjusted model is in Additional file 1: Table S3.
The final model is presented in Figs. 1, 2, and 3 with OR logarithm and respective CI for each outcome. Variables whose confidence intervals do not cross the null value are statistically significant. The results of the final model (OR and IC) are also provided in Table 5.

Discussion
The treatment success for individuals in the study cohort (63.4%) was higher than that achieved worldwide (55%) in 2016 for cases of MDR/RR-TB; however, it was still far short of the values considered acceptable by the WHO [1]. Our analysis evidenced that the factors related to the unfavorable outcomes studied are distinct and multifactorial. Individual and follow-up treatment factors were related to LFU. For failure, the clinical characteristics, proposed treatment, and follow-up stand out. For death, issues related to individual, clinical, treatment characteristics, and follow-up treatment are related to this outcome. The protective effect against LFU of recorded minor adverse reactions compared to no recorded adverse reactions suggests that factors related to the quality of health services, which were not included in the study, may also influence unfavorable outcomes of DR-TB. The hypothesis is that health units with professionals that were more sensitive to user complaints, where the presence of minor adverse reactions is valued, obtained better treatment outcomes. Low education was associated with LFU and death. Several studies have already associated education with adherence to sensitive and MDR-TB treatment [10,[16][17][18]. The significance of this finding can be extrapolated to their conditions, as education is recognized as a proxy of socioeconomic status [19]. The same can be said for brown/black race/color [1620], which was also associated with death. Research has already proven that the mitigation of social determinants is essential to control TB [21]. However, for DR-TB, this is not enough for therapeutic  success. A study that assessed the impact on the outcome of DR-TB of social and economic intervention demonstrated satisfactory results for adherence but no impact on treatment failure [22].
Having lesions detected in both lungs suggests a late diagnosis and this factor was related to failure and death, as was also found in other studies [18,23,24]. In the study period, the GeneXpert MTB/RIF ® was just being implemented in Brazil and culture was taken for 26% and 42% for new and retreatment cases, respectively [25]. These results confirm the difficulty in making the diagnosis of MDR/RR-TB and corroborates the hypothesis that individuals arrived too late for treatment, influencing therapeutic success. The GeneXpert MTB/RIF ® established in Brazil can diagnose 57% of new TB cases; however, only 33% of new cases in the country performed GeneXpert MTB/RIF ® [25]. These results suggest that the available laboratory network is still not suitable for sending samples to laboratories with equipment or that the TB diagnostic algorithm is not being complied with according to national recommendations, making access to the exam more difficult [3]. The protective association against LFU of living in a municipality other than the treatment location also may indicate that access to DR-TB diagnosis and treatment may have influenced the outcome. The individuals who obtain DR-TB diagnosis in smaller municipalities, without TB reference services are probably different from those who live in large cities, where access to diagnosis DR-TB is broader.
Alcoholism has already been associated with the worst outcomes of sensitive and MDR-TB [9,16,26,27]. This same association occurred in the bivariate analysis of this study for LFU and death, but it was not maintained in the final model. The inclusion in the final model of variables related to alcoholism, including illicit drug use and education, may explain why alcoholism did not remain associated. A study that evaluated the relationship between alcohol consumption and the outcome of MDR-TB treatment concluded that this consumption may be an indicator for other behavioral disorders [27], such as illicit drug use, which maintained the association with LFU in the final model. Another factor that may have influenced this result is the lack of a standardized definition of alcoholism in the TB notification forms.
The gender difference was not found to influence the DR-TB treatment outcomes assessed in this study, which corroborates other studies carried out in Brazil and in other BRICS countries [10,27,28]. However, studies about sensitive tuberculosis have already associated males with less chances for cure, which could be because men seek health services less often and how the health services organize their programmatic activities, prioritizing the maternal and child population [29,30].
The association between the XDR resistance pattern and death was expected [3]. With the launch in 2019 of the TB Guidelines in Brazil, the MoH reinforced the need for early diagnosis of these cases, recommending considering failure of the first treatment of tuberculosis to be MDR-TB when there is no bacteriological conversion and no clinical improvement after eight months of treatment. In addition, these Guidelines also standardize the therapeutic regimen for XDR-TB, which before publication was only performed individually [3].
The final multivariate model associated HIV positivity with death. Studies have already found this result [9,31,32], as well as the greater occurrence of adverse effects  and drug interactions in these cases [33]. The early diagnosis of drug resistance, the appropriate definition of the treatment, and the rapid initiation of antiretroviral therapy (ART) two weeks after starting TB treatment, if the patient is not already on ART, are recommendations for the treatment of co-infected patients [33,34].
Having reported unfavorable clinical evolution at any time during treatment was strongly associated with all the unfavorable outcomes studied. This information is included in SITE-TB when the physician responsible for the follow-up appointments observe a worsening of clinical symptoms, imaging, or laboratory tests. Thus, this information can be used by professionals who monitor patients with DR-TB as a sign of the clinical evolution of a potential unfavorable outcome during treatment.
This study has some limitations in addition to those inherent to retrospective observational studies based on secondary databases. Few patients had undergone a sensitivity test for second-line drugs, which may have caused classification bias of the initial resistance pattern and the change in the resistance pattern of the cases during the treatment. Another limitation was the way in which SITE-TB registers associated diseases and conditions as well as adverse reactions, not differentiating missing data from the absence of the disease/condition/adverse reaction, generating a possible classification bias and underestimating the occurrence of these factors. The variable HIV result does not have this limitation; however, for the HIV and the education variables, the option " missing " among the analysis categories was necessary. In addition, there is no standardized definition of smoking and alcoholism for filling out the notification forms. The SITE-TB does not allow exporting the treatment scheme used in the case of individualized schemes. Therefore, it was only possible to know the drugs used in each case, which was included in the analyzes.
The exclusion of 17 records (1.7%), given the type of conclusion completed at the time of exporting the database, may have influenced the results. Likewise, 774 (88.1%) MDR-TB cases and 104 (11.9%) RR-TB cases were included as a single category of initial resistance pattern and this categorization may also have influenced our findings. Finally, the unavailability of the body mass index, also predictors of unfavorable outcomes already studied [35], could be a limitation.

Conclusions
In the historical cohort of DR-TB case analyses, the factors associated with unfavorable treatment outcomes were different. Some factors were specific to each outcome, which reflects the complexity of the care provided to these individuals.
These results may be used in the development of an index to identify the risk of unsuccessful outcomes for DR-TB cases, helping services to identify cases with a greater chance of unfavorable treatment outcomes. Although the WHO recommends short or long-term regimens, including bedaquiline and pretomanid, to facilitate adherence to MDR-TB treatment, these are not yet available in Brazil [3,36]. The implementation of these new technologies may provide benefits to the therapeutic success of DR-TB and modify the results found. Implementation of DR-TB diagnosis, with wide access to molecular tests to detect antimicrobial resistance markers and to first-and second-line sensitivity tests, is another factor that could assist in the outcome of cases by permitting early diagnosis of anti-TB drug resistance. Finally, studies should be conducted that include in the analyzes the association between the quality of TB reference services, the phenotypic diversity of strains in the cases, and the outcomes of DR-TB.