This analysis shows that exposure to tobacco smoke is consistently associated with TB, regardless of the specific types of exposures and specific TB outcomes. Compared with people who do not smoke, smokers have an increased risk of a positive TST, of having active TB, and of dying from TB. Although there were fewer studies for passive smoking and IAP from biomass fuels, those exposed to these sources were found to have higher risks of active TB than those who are not exposed. An important finding of this study is the suggestion that the risk of TB among those exposed to passive smoking is especially high in children who are not normally at high risk for active disease. These findings support the hypothesis that exposure to respirable pollutants from combustion of tobacco and biomass fuels increases the risk of both TB infection and TB disease.
In addition to the positive association demonstrated here, multiples lines of evidence support a causal relationship between combustion smoke and TB. A dose–response relationship has been demonstrated in most of the studies that have stratified on dose; in this meta-analysis, we found that the risk of TB increases with both daily dose of cigarettes and duration of smoking. There is also accumulating evidence for the biological plausibility of this association. Chronic exposure to tobacco as well as to a number of environmental pollutants impairs the normal clearance of secretions on the tracheobronchial mucosal surface and may thus allow the causative organism, Mycobacterium tuberculosis,
to escape the first level of host defenses, which prevent bacilli from reaching the alveoli [67
]. Smoke also impairs the function of pulmonary alveolar macrophages (AMs), which are not only the cellular target of M. tuberculosis
infection but also constitute an important early defense mechanism against the bacteria; AMs isolated from the lungs of smokers have reduced phagocytic ability and a lower level of secreted proinflammatory cytokines than do those from nonsmokers [68
]. Recent work has suggested a novel mechanism for this effect; nicotine is hypothesized to act directly on nicotinic acetylcholine receptors on macrophages to decrease intracellular tumor necrosis factor-α production and thus impair intracellular killing of M. tuberculosis
]. Wood smoke exposure in rabbits has also been shown to negatively affect antibacterial properties of AMs, such adherence to surfaces, ability to phagocytize bacteria, and intracellular bactericidal processes [70
]. Boelaert and colleagues [71
] have also proposed an alternative explanation for the impaired ability of macrophages from smokers to contain M. tuberculosis
infection. These investigators noted that AMs from smokers have an markedly elevated iron content and that macrophage iron overload impairs defense against intracellular microorganisms through reduced production of both tumor necrosis factor-α and nitric oxide.
The available data support a causal link between smoke exposure and either an increased chance of acquiring TB or progression of TB to clinical disease. Our study shows that the risk of latent TB among smokers is quantitatively similar to their risk of active disease, which would suggest that much of the impact of smoking takes place during infection. At the same time, one case-control study selected TST-positive controls, thereby comparing patients who were TST positive and had clinical TB to people who were also TST positive but had not progressed to clinical TB [54
]; that study also found a strong association between smoking and disease, suggesting that smoking may induce progression or reactivation disease in those infected. We included the outcome TB mortality in this study in order to investigate the association between smoke and TB occurrence rather than the association between smoke and TB treatment outcomes. The risk of death from TB among smokers was found to be somewhat higher than the risk of latent infection or disease, possibly because smoking has been identified as a risk factor for poor TB treatment outcomes among those undergoing therapy [57
There are several potential limitations to this study. First, our findings are based on the results of observational studies; we cannot, therefore, exclude the possibility of confounding by variables that may be associated with each of the exposures. The issue of confounding is particularly a concern in a meta-analysis of observational studies when effect sizes are relatively small, as was the case in the studies considered in this analysis [74
]. We therefore performed a stratified analysis to explore the degree to which potential confounders may have influenced the findings. Among possible confounders, alcohol use is a known risk factor for TB and is closely associated with tobacco use in many populations. Those studies that adjusted for alcohol intake in a multivariable model found that the effect of smoking was reduced, but not eliminated. Those studies that controlled for the effect of alcohol were also less heterogeneous as a group than those that did not, a finding which suggests that some of the variability may have resulted from differences in alcohol consumption. Other risk factors that may confound the association between smoking, passive smoking, and IAP and TB include socioeconomic status, gender, and age. Although it is impossible to fully exclude bias introduced by residual confounding, we found that the effects the exposures on TB remained after adjustment for these factors.
More than half of the studies in our review are case-control studies. These used different approaches to the selection of controls, including sampling from hospitals and clinics, from household members, and from the community. Since smoking is associated with a wide range of diseases, the choice of hospital- or clinic-based sampling may lead to over-representation of smokers among the controls, thereby biasing the results toward the null. Similarly, since people dwelling in the same household may share behavioral risk factors, controls chosen from households of smoking TB patients may have been more likely to smoke than would the general population [75
]. When we compared the effect estimates for studies stratified on the basis of the control selection strategy, we found that studies that had not used population-based controls tended to report lower effect estimates, consistent with our expectation of a bias toward the null among studies that used hospital- and household-based controls.
Other potential sources of bias include possible misclassification of both exposure and outcome status. The assessment of tobacco smoking relied on self-reported behavior, which may not have been accurate especially among those who consider smoking to be stigmatizing, such as women in some cultural settings. The exposure “current smoking” may also have been subject to reverse causation. Patients are often diagnosed with TB months or more after having first experienced symptoms of the disease, which may cause some patients to quit smoking. This is consistent with the finding of several studies that “former” smoking to be a stronger risk factor for TB than current smoking [34
]. Nonetheless, since “former” smoking also included very distant smoking, both current and former smoking may underestimate the effect of smoking that had occurred just prior to the onset of disease. Similarly, misclassification of passive smoking and IAP may have introduced a bias toward the null in our analysis. The classification of passive smoking among children, for example, relied on parent reports, which may have been influenced by guilt or shame at having exposed the child to an agent suspected of causing disease. Most problematic among exposures was the classification of IAP; this usually relied on the proxy “use of biomass cooking fuel,” which probably only coarsely captured the actual exposure to inhaled smoke. For example, one study that found no association between biomass fuel use and TB noted that houses in the area were well ventilated, and thus actual exposure to inhaled smoke among those using biomass fuels was probably lower.
Misclassification of outcome may have also introduced bias into this analysis. For example, we included a large mortality study conducted in India in which the odds of death among urban male smokers was 4.5 times that of nonsmokers. Since diagnosis of TB in India relies heavily on radiographic findings, TB may be overdetected, especially among patients with pulmonary lesions—such as malignancies—that may be causally linked to smoking [76
]. When we repeated our analysis excluding the two Indian mortality studies, the heterogeneity among the remaining studies was reduced. Similarly, when the mortality studies were excluded from the funnel plot, there was much less variability among the studies with the smallest standard errors. Another possible source of outcome misclassification was suggested by Plant and colleagues [32
], who noted that the frequency of small induration sizes among TSTs was higher among smokers than nonsmokers, suggesting that smokers may be less capable than nonsmokers of eliciting a vigorous skin test reaction and that latent TB infection in smokers may thus be underdetected when the 10 mm cutoff is used. Despite this possible limitation, we found that the two studies of latent infection that used 5 mm cutoffs for the TST [32
] reported effects that were not statistically different from those that used 10 mm [30
]. Finally, the diagnosis of TB in children is notoriously difficult; if children exposed to passive smoke were more likely to be successfully diagnosed with disease than those who were not, this might have introduced a bias that would explain the strong positive association between passive smoking and TB.
Although our evidence suggests that tobacco smoking is only a moderate risk factor in TB, the implication for global health is critical. Because tobacco smoking has increased in developing countries where TB is prevalent, a considerable portion of global burden of TB may be attributed to tobacco smoking (see Text S1 for an illustrative calculation of population-attributable fraction and attributable deaths in different regions of the world). More importantly, this association implies that smoking cessation might provide benefits for global TB control in addition to those for chronic diseases.
Despite heterogeneity in design, measurement, and quantitative effect estimates among the studies included in this analysis, we found consistent evidence for an increased risk of TB as a result of smoking, with more limited but consistent evidence for passive smoking and IAP as risk factors. These findings suggest that TB detection might benefit from information on exposure to respirable pollutants from sources such as smoking and biomass use, and that TB control might benefit from including interventions aimed at reducing tobacco and IAP exposure, especially among those at high risk for exposure to the infection.