Article Text
Abstract
Background: Tuberculosis remains an important public health problem in Brazil where over 100 000 new cases and 6000 deaths are reported every year. Current drug efficacy and wide availability should have curbed this toll. The goal was to study the factors associated with death in incident cases of pulmonary tuberculosis (PT) during treatment.
Methods: This is a case–control study including all new cases of PT reported between 2000 and 2004; cases were patients who died of any cause and controls were those cured after initial PT treatment. Data analyses included unconditional multiple hierarchical logistic regression.
Results: A total of 313 502 new cases of tuberculosis were reported between 2000 and 2004; 224 355 (71.6%) were cured after initial treatment and 20 721 (6.6%) died during the surveillance follow-up. Over 82% of all cases were diagnosed with PT. After controlling for significant variables, the factors associated with a higher risk of death included gender (males: odds ratio (OR) 1.4; 95% confidence interval (CI) 1.33 to 1.47), age (<5 years of age: OR 1.90; 95% CI 1.51 to 2.38; 30–59 years: OR 2.78; 95% CI 2.61 to 2.97; over 60 years: OR 10.92; 95% CI 10.09 to 11.81), positive HIV serology (OR 10.59; 95% CI 9.76 to 11.48), alcoholism (OR 1.49; 95% CI 1.36 to 1.65), mental disorder (OR 1.80; 95% CI 1.43 to 2.27) and presence of additional lung pathology on chest x ray (OR 2.22; 95% CI 1.83 to 2.70). Protective variables included education (highest level: OR 0.67; 95% CI 0.63 to 0.70).
Conclusions: Preventive interventions should target the most vulnerable patients, in particular the very young and the elderly, those infected with HIV and those presenting with a mental disorder or additional lung pathology.
Statistics from Altmetric.com
Tuberculosis (TB) remains an important public health problem in Brazil. Although significant improvements have been made, there are still approximately 100 000 new cases and 6000 deaths reported every year; these figures rank Brazil as having the 15th highest incidence in the world.1 Socioeconomic development is the main determinant of TB incidence.2 3 However, in spite of that, death from TB should be avoidable given current drug efficacy and the wide availability of such drugs in the country.4–6
The national Tuberculosis Control Program (TBCP) has implemented important control measures, but it has not yet succeeded in decreasing mortality as expected. In order to better understand why, we proposed to study the factors associated with death among TB patients using data from the TBCP program. Our main goal was to identify the factors that could guide the development of more focused control strategies, ultimately leading to better allocation of resources and improved outcomes.
STUDY POPULATION AND METHODS
Study design and population
This is a case–control study based on secondary data. Newly diagnosed cases of pulmonary tuberculosis (PT) reported between January 2000 and December 2004 were eligible. Cases were defined as all eligible patients who died from any cause during initial treatment for TB. Controls were all eligible patients classified as cured after completing the TB treatment. The Brazilian Health Ministry-run TBCP defines a new case as a patient with a first diagnosis of TB and as cured, patients who completed treatment independent of bacteriological confirmation (patients presumed cured).
Data collection and variables studied
Data were obtained from the Reportable Disease Information System (SINAN, Brazilian Ministry of Health), a population-based, decentralised information system in which only confirmed cases of tuberculosis are entered. Cases are confirmed when they fulfil either the clinical and laboratory criteria or the clinical and epidemiological criteria. Details entered include place of notification, sociodemographic characteristics, past medical history, details of treatment and follow-up.
Consistency analyses were performed by the Department of Health Analysis (Ministry of Health) for quality control and to ascertain data completion. Explanatory variables were: (1) demographic and socioeconomic factors: sex, age group, place of residence (state capital (yes/no), region (south, southeast, west central, north and northeast) and type of municipality (non-prioritised or prioritised as set by the TBCP)), year of diagnosis (2000–2004), self-reported ethnicity or skin colour (white, black, brown, yellow and native), formal education (number of completed school grades); (2) pre-existing medical conditions and comorbidities: HIV serology, alcoholism, diabetes, mental disorder, other pathologies diagnosed through chest x ray examination; (3) variables associated with the disease or with its treatment: use of directly observed treatment (DOTS), tuberculin skin test and treatment modifications including changes due to medication intolerance, toxicity or treatment failure. When the variable included a large number of missing responses, an additional category was created for subjects with unknown (missing) responses.
Statistical analysis
Unconditional, hierarchical, multiple logistic regressions were used to evaluate the association between selected independent factors and the dependent variable.7 8 All analyses were carried out using the statistical software STATA.9 Hierarchical analysis levels were established based on a death causality hypothetical model applied to patients with PT as follows: the first hierarchical level (FHM) included the demographic and socioeconomic variables; in the second hierarchical level (SHM), the model was supplemented with information on pre-existing medical conditions and comorbidities and, in the third hierarchical level (THM), variables associated with the disease or with the treatment were added.
The logistic regression analyses observed the following stages: (1) crude and multivariate analyses of all variables (non-collinear) from FHM; (2) multivariate analyses of all variables (non-collinear) from SHM (adjusted by all variables kept in the final FHM); (3) multivariate analyses of all variables (non-collinear) from THM adjusted by all variables kept in the final FHM and SHM. Backwards selection was used to construct the final model for each hierarchical level. The Wald test was performed for all dummy variables. The “unknown” category was included in studying some variables because its exclusion could introduce selection bias. For all associations, the conservative significance level of 0.01 was chosen on account of the large sample size and in order to limit identification of associations statistically significant at the level of p<0.05, which may be of no biological, clinical or epidemiological relevance.
Odds ratios (OR(h)) were estimated by the hierarchical process, and OR(full model) originated from full models given no hierarchy. As it is possible to invoke the “rare disease assumption” (proportion of deaths <7%), the OR in this study is interpreted as a good estimator of the relative risk or, more specifically, the “case fatality proportion” ratio.7 Conceptually, the term “case fatality proportion” in this article refers to the proportion of reported deaths (from any cause) among patients with PT.
RESULTS
From 2000 to 2004, a total of 464 741 diagnoses of tuberculosis were reported to SINAN. Under the normal surveillance procedures, an entry form and a treatment outcome questionnaire were completed for 391 492 of these patients; 313 502 (80.1%) were new cases of TB and 27 773 (7.1%) were classified as relapses (table 1). The remainder were cases with incomplete treatment or incomplete information.
Almost 72% of new cases (n = 224 355) with a known outcome completed the initial therapeutic regimen and were thus considered cured by the TBCP. The overall case fatality proportion was 6.6% among new cases; this rate was higher for patients who entered the SINAN system due to a relapse (8.3%), default (8.7%) or a transfer (14.7%) (table 1).
Most patients had TB confined to the lungs (83% or 297 285 = study population). In this group, there were 15 863 deaths (the case series for the present analyses) and 187 664 cures (the group taken as the control).
Cases, compared with controls, were more likely to be males, older, residents in the southern and southeastern regions, not presently residing in a programme priority municipality, illiterate, with HIV positive serology, reported alcoholism and presenting with other pathologies on chest x ray examination (tables 2 and 3). A higher proportion of controls were not assigned to DOTS strategy and were more frequently found with a strong reaction in the tuberculin test result and to be less likely to have a missing tuberculin test result (table 3).
Tables 4 to 6 depict the results of crude (OR) and adjusted (OR(h)) odds ratios from hierarchical multiple logistic regression models. In addition, the OR(full model) originated from full models given no hierarchy are also presented below.
A significantly higher likelihood of death was found for men compared with women (OR(full model) 1.32; OR(h) 1.40, 95% CI 1.33 to 1.47), and for extremes of age when compared with patients aged 5–29 years: children younger than 5 years (OR(full model) 1.78; OR(h) 1.90, 95% CI 1.51 to 2.38), adults 30–59 years (OR(full model) 2.50; OR(h) 2.78, 95% CI 2.61 to 2.97) and those 60 and older (OR(full model) 11.72; OR(h) 10.92, 95% CI 10.09 to 11.81).
Patients living in state capitals had a slightly higher chance of dying than those living elsewhere (OR(full model) 1.03; OR(h) 1.21, 95% CI 1.14 to 1.28). When compared with those living in southern Brazil, a lower risk of death was found for patients living in the southeastern (OR(full model) 0.83; OR(h) 0.81, 95% CI 0.76 to 0.86), west central (OR(full model) 0.73; OR(h) 0.67, 95% CI 0.60 to 0.74), northern (OR(full model) 0.67; OR(h) 0.56, 95% CI 0.51 to 0.61) and northeastern regions (OR(full model) 0.60; OR(h) 0.52, 95% CI 0.49 to 0.56). Likewise, patients living in municipalities prioritised by the TBCP had a lower risk of dying than those in other towns (OR(full model) 0.83; OR(h) 0.88, 95% CI 0.83 to 0.93). As expected, educational level was a very significant protective factor: patients who completed at least 4 years of schooling had significantly lower risk of dying than their counterparts (OR(full model) 0.64; OR(h) 0.67, 95% CI 0.63 to 0.70). The set of variables selected for this first hierarchical level (FHM), however, explained poorly the risk of death among the population studied.
In the second hierarchical level (SHM), a strong positive association was found between the risk of dying from PT and HIV seropositivity (OR(full model) 10.60; OR(h) 10.59, 95% CI 9.76 to 11.48), alcoholism (OR(full model) 1.61; OR(h) 1.49, 95% CI 1.36 to 1.65), coexisting mental disorder (OR(full model) 1.92; OR(h) 1.80, 95% CI 1.43 to 2.27) and the presence of other pathologies on chest x ray examination (OR(full model) 2.21; OR(h) 2.22, 95% CI 1.83 to 2.70).
In the third hierarchical level (THM), new cases of PT with a strong reaction in the tuberculin test had a significantly lower risk of death than those with a negative or small reactive result (OR(full model) OR(h) 0.42, 95% CI 0.36 to 0.49) (table 6). Interestingly, cases were more likely to be included in the DOTS strategy than controls (OR(full model) OR(h) = 1.11, 95% CI 1.04 to 1.20).
DISCUSSION
Tuberculosis has proved to be a major public health challenge worldwide.10 11 It is well known that case management in local health sectors plays an important role in preventing death from tuberculosis.10 In the present study, factors associated with death after first treatment were identified among patients with incident PT in Brazil. In addition, this study may provide ammunition to the debate on the usefulness and limitations of the Reportable Disease Information System (SINAN) database. Although SINAN reports deaths from all causes among tuberculosis patients, it is expected that the most likely cause of death is tuberculosis, as the underlying or associated cause of death that somehow contributes to death from other causes.12 Lillebaek et al12 documented that, among patients with tuberculosis in Denmark, only 8.3% of all deaths were due to other causes of death.
Several factors associated with the risk of death among tuberculosis patients were identified including gender, age, place of residence, education level, HIV serology, alcoholism, mental disorders, other pathologies identified by a chest x ray examination, DOTS strategy and tuberculin test result. These findings are corroborated by other studies. For instance, male gender and extreme age have been implicated as important risk factors for death among tuberculosis patients in different countries.13–17 The greater vulnerability of the elderly has been described previously; this increased risk is partly explained by the presence of other concomitant health conditions and a higher incidence of atypical PT presentations.15 18–20 It has been postulated that the latter may lead to a higher case fatality proportion by prompting delays in diagnosis and treatment.15 Similarly, children younger than 5 years old may also be at higher risk because of diagnosis and treatment delays, but for different reasons, which include lack of standardised case definitions and the relatively low incidence of the disease in this age group with the rarity of a positive bacilloscopy, which may decrease the awareness of both physicians and laboratory technicians.21
Previous studies have also found male gender to be associated with a higher tuberculosis incidence, mortality and case fatality proportion.14 22 23 Oliveira et al14 reported the case fatality proportion among men to be 1.4 times higher than that for women (14.6% vs 10.6%). Biological and behavioural characteristics, as well as those associated with the utilisation of health services, may be involved in conferring different levels of risk by gender. However, these determinants are not yet fully understood.
It is important to be careful in interpreting associations between gender or age and death among PT patients. Compared with younger people, the elderly are expected to present a risk of death higher than that of the general population, for several reasons (other than tuberculosis), for example due to cardiovascular disease. In the same manner, men are expected to present higher mortality rates than women, for example due to injuries.23 Because we were not able to differentiate deaths “with” PT from deaths “due to” PT, this may have generated an overestimation of the differences in the case fatality proportions between these and the reference groups in this study.
Patients living in municipalities prioritised by the TBCP experience death rates lower than their counterparts. This protective effect may seem to point towards the effectiveness of the set of actions deployed in these places. If such effectiveness is confirmed in future evaluations, such programmes should be considered for implementation in other TB areas.
Educational level was one of the most important protective variables studied, leading to a significant reduction in the case fatality proportion among study patients. The reasons are at least twofold. First, formal education along with intellectual capacity would allow higher assimilation of information, more adequate use of resources, including health services, and several other self-protective and self-care behaviours. Second, this may be accomplished by the high correlation between educational level and socioeconomic status. Other authors have already described the close relationship between socioeconomic indicators and the risk of becoming ill with and dying from tuberculosis across regions in Brazil.24–26
Our analyses also revealed a higher probability of death for patients with a positive HIV serology, reporting alcoholism, mental disorder and/or alterations in the chest x ray examination. These additional elements are useful in the identification of target groups for specific interventions aimed at preventing death from PT.
The association between HIV seropositivity and higher chances of death among tuberculosis patients is well documented.24 27 To fully understand the multifaceted relationship between the probability of dying from tuberculosis in the presence of AIDS and vice versa, more information is required. Once infected by the bacillus, the AIDS patient has a much higher risk of developing tuberculosis than the general population. The presence of tuberculosis also accelerates AIDS evolution, augments immunodeficiency progression and increases the patient’s vulnerability to several other opportunistic diseases. Therefore, health professionals should be aware of the possible concomitance of these diagnoses in order to intervene adequately and opportunistically.
Treatment non-compliance is a well-documented risk factor for recurrence, drug resistance development and death, and the DOTS strategy has been described as a useful tool to diminish non-compliance.10 11 28 However, in the present study, we did not find a protective effect afforded by this measure: PT patients who had DOTS available to them actually presented a slightly higher chance of death compared with those for whom the strategy was not offered or had missing information. We speculate that this may be due to the inclusion criteria used to select patients for DOTS; ie, patients at higher risk of non-compliance and death may be preferentially selected, characterising a potential for indication bias. The multivariate analyses we performed did not take into account all possible confounding factors that could better prevent this bias, eg homelessness, nutritional status, the presence of other comorbidities (not included here) and unmeasured daily therapeutic compliance (even under DOTS). These variables are not collected through the SINAN system. In addition, the reported coverage of the DOTS strategy is low (∼15%) and its notification to SINAN is quite poor (∼60% missing information), potentially presenting low sensitivity and high specificity. However, there have been reports that this association, also seen recently in India, may derive from poor-quality monitoring of DOTS.29
We found that patients with a negative or low reaction to the tuberculin test had higher risk of dying than those with large reactions. It is possible that a negative or low reaction to a diagnostic test may cause diagnosis and treatment delays. We also postulate that a negative/low tuberculin test result could be an indicator of the severity of the disease. Anergy or hypoanergy are common in elderly patients and those with AIDS or other immunodeficiencies, especially in the final stages of the disease when the patient is extremely undernourished. Owing to missing information, we were only able to control partially for the confounding effect of advanced age and AIDS; therefore residual effects may be present.
One should interpret these results with caution. We chose to select as controls all patients classified as cured and to exclude all those with unknown outcomes (ie, “transferred out”, “unknown” and “default”). In this manner, we prioritised the comparison between the two most extreme categories. However, we found this strategy methodologically justified because the excluded individuals would have had a differential probability of being identified as cases (death); we have thus prevented the contamination of the control group with potential cases and/or with individuals not representative of the base population in this study.
This study has identified vulnerable groups of patients who are at higher risk of death and who would benefit from early identification, treatment and follow-up. It is worth mentioning that this study was the first national study to identify factors associated with death among tuberculosis patients in the entire country, presenting great potential for generalisation of the results.
What is already known on this subject
Tuberculosis (TB) remains an important public health problem in Brazil. There are still approximately 100 000 new cases and 6000 deaths reported every year. It is well known that socioeconomic development is the main determinant of TB incidence; however, in spite of that, death from TB should be avoidable given current drug efficacy and the wide availability of such drugs in the country.
The national Tuberculosis Control Program (TBCP) has implemented important control measures, but it has not yet succeeded in decreasing mortality as expected.
Factors associated with death among TB patients are not clearly studied and may be an important piece of evidence to fully inform the Brazilian TBCP.
What this study adds
This study offers the identification of factors associated with death during first treatment among patients with incident pulmonary tuberculosis (PT) in Brazil including gender, age, place of residence, education level, HIV serology, alcoholism, mental disorders, other pathologies identified by chest x ray examination, DOTS strategy and tuberculin test result.
These findings may help in the opportune identification of more vulnerable patients and the implementation of preventive measures and follow-up.
This is the first national research that identifies factors associated with tuberculosis-related deaths among Brazilian patients in the entire country, greatly improving the external validity of study results.
Acknowledgments
The authors acknowledge the logistic support received from the Secretariat of Health Surveillance, Brazilian Ministry of Health and from the Department of Social Medicine, University of Brasilia, Brazil.
REFERENCES
Footnotes
Funding: This study received logistic support from the Department of Social Medicine, University of Brasilia, Brazil. The Secretariat of Health Surveillance, Brazilian Ministry of Health, provided access to the Tuberculosis database. No direct funding was involved in this study.
Competing interests: None.
Ethics approval: This study included only public secondary data stripped of any individual identifiers. Therefore, ethics committee approval was not required.