Figure 2

Figure 2. Summary of the data flow for case–control analysis of tuberculosis prevalence, Vietnam, 2017–2018.

Of the 1,230 preselected controls, 107 did not attend the screening event or declined to participate, and 54 had a positive Xpert result that defined them as TB cases (Figure 2). There were 1,319 interviewed participants in total, of which 250 were cases and 1,069 were controls (Table 1); 694 (52.6%) participants were men and 625 (47.4%) women. Among the controls, 2 were later reported to have positive culture results, classified as TB cases by the expert panel, and excluded from our analysis.

The distribution of men and women among the domains (Table 2) shows that men were more dominant in the most disadvantaged tertile of all domains, except for SES. Compared with women, men more often reported severe clinical symptoms (12.8% vs. 6.7%), poor access to healthcare (38.9% vs. 24.0%), and high behavioral and environmental risk (52.9% vs. 4.2%) (Table 1).

Figure 3

Figure 3. Structural equation model of the relationships between domains and Xpert-positive (Xpert MTB/Rif; Cepeheid, https://www.cepheid.com) tuberculosis prevalence for case–control analysis of tuberculosis prevalence, Vietnam, 2017–2018. For significant associations, the...

SEM results (Figure 3), full estimation results (Table 3), and the confirmatory factor analysis results (Appendix Table 1) indicate that sex was directly associated with prevalent TB (adjusted odds ratio [aOR] for men compared with women 3.0 [95% CI 1.7–5.0]). Sex was also associated indirectly with prevalent TB through behavioral and environmental risks, behavioral and environmental risks and clinical symptoms, and behavioral and environmental risks and access to healthcare. Clinical symptoms (aOR for moderate severity 5.0 [95% CI 2.8–8.7]; aOR for most severe 13.2 [95% CI 5.9–29.2]), access to healthcare (aOR for moderate access 2.5 [95% CI 1.4–4.4]; aOR for poor access 13.4 [95% CI 7.0–25.6]), and behavioral and environmental risks (aOR for moderate risk 2.1 [95% CI 1.1–3.9]; aOR for high risk 2.9 [95% CI 1.5–5.6]) were directly associated with TB as well. SES did not directly affect prevalent TB, only indirectly through access to healthcare, behavioral and environmental risks, and clinical symptoms (Figure 3, Table 3). When we stratified the SEM model by sex, we found a statistically significant association between prevalent TB and behavioral and environmental risks among men (aOR for moderate risk 3.5 [95% CI 1.5–8.2]; aOR for high risk 4.8 (95% CI 2.0–11.7]). That same association was not statistically significant among women (Appendix Figures 1 and 2).

Considering associations between domains and covariates, we found more severe clinical symptoms among participants who were >55 years of age, lived in northern Vietnam, had poor access to healthcare, were more exposed to harmful substances, or had lower SES. Poor access to healthcare was more frequently observed in participants living in rural areas and among those with high behavioral and environmental risks or lower SES. Behavioral and environmental risks were less frequently observed in women, younger participants (15–24 years of age), those living in northern locations, and those with high SES. The bootstrapped AUC result showed that our model could predict the TB status of persons living in Vietnam, assuming the input information is adequate, with a probability of 0.90 (95% CI 0.89–0.92, Table 3).

In the sensitivity analyses, when the clinical domain was excluded in model 1, there was no statistically significant difference in the direct effect of sex nor in the effect of access to healthcare, behavioral and environmental risks, and SES on prevalent TB compared with the primary SEM model (Appendix Table 2). We found no statistically significant difference between the main SEM model and sensitivity analysis in model 2, where the main outcome was replaced by TB status defined by the expert panel (Appendix Table 3). This was also the case for analysis model 3, where all Xpert-positive TB cases with a history of previous TB treatment were excluded from the analysis (Appendix Table 4).

In Vietnam, we witnessed an exceptionally large difference in TB prevalence in relation to sex, with behavioral and environmental risks as the biggest contributors to this disparity: sex determined behavioral and environmental risks leading to TB. Even after accounting for all known associated factors of prevalent TB collected in our data, we found that the odds of having TB were still 3 times higher for men than for women, and this difference remained in our sensitivity analyses. Although we cannot exclude unmeasured confounding, our data suggest that in addition to behavioral and environmental factors, access to healthcare, and symptom presentation, there may be biologic factors that render men more vulnerable to TB than women. Such biologic factors may contribute substantially to the M:F ratio for TB observed in Vietnam and globally.

The behavioral and environmental risks domain, while being strongly associated with sex, only exerted a modest effect on prevalent TB. Of note, in the sex-stratified SEM models, the behavioral and environmental risks domain had a statistically significant effect on prevalent TB among men, but not among women (Appendix Figures 1, 2). This domain consisted both of factors affecting the risk for MTB infection (working indoors, contact with TB patients) and of factors affecting the risk for TB disease progression (smoking, excessive drinking), because we believe that the exact pathophysiologic effect of these factors is unknown and may be multiple and interactive. For example, heavy drinkers may also socialize with more persons in poorly ventilated spaces, thus increasing their risk for MTB infection. In Vietnam, men smoke and drink excessively at an overwhelmingly higher rate compared with women, with a smoking rate of 45.3% vs 1.1% (16) and an excessive drinking rate of 44.2% vs 1.2% (17). This explains the tremendous influence of sex on the behavioral and environmental risks domain, even when the levels of working indoors and contact with TB patients were similar for both sexes. Also, behavioral and environmental risks were higher among the middle-aged compared with the youngest age group, which is consistent with findings of a survey of tobacco consumption in Vietnam in 2015 (17). The effect of this pathway on TB in the sex-stratified SEM models solidifies the evidence that smoking and drinking are major drivers of the difference in TB prevalence between men and women.

Although access to healthcare appeared the strongest predictor of tuberculosis prevalence, we did not find a direct effect of sex on the access to healthcare domain, because men and women in Vietnam tend to have similar access to healthcare. This opposes the findings of other studies, where the gap in access to TB health services is stated to be the main reason behind sex-related differences in the TB burden (2,18). In Vietnam, a nationwide study in 2002 showed that women have longer TB diagnostic delays than men (5.6 weeks vs 4.4 weeks on average) (19), but results from a laboratory study in northern Vietnam suggested that women are more likely than men to have sputum smear examinations (20). Our study revealed no significant direct effect of sex on access to healthcare but did demonstrate an indirect effect through exposure. This finding is in line with studies indicating an association between heavy drinking and decreased healthcare utilization (21,22), because drinkers may be less likely to seek preventive care and primary healthcare than abstainers.

Before 1945, only 10% of the population of Vietnam was literate (23); thus, in our study, elderly persons most often had no schooling. We found no difference in SES between men and women, but elderly participants (≥60 years of age) reported a much lower SES than younger participants (15–24 years of age). The SES domain showed only a marginal direct association with TB, but significant indirect associations through the other 3 domains. Studies have shown that persons with low SES are more likely to be uninsured, seek healthcare less often, and have poor-quality healthcare, which may have led some participants to visit medical facilities with more severe clinical symptoms during the survey (24,25). Persons with low SES also are more likely to smoke and drink alcohol excessively (26), which suggest the need for welfare-focused interventions for TB control and prevention.

After controlling for all domains, we still found that men in Vietnam are 3 times more likely to have TB than women. Although other latent factors might have contributed to those odds (e.g., silicosis, nutritional status, illicit drug use, social contact patterns), our findings suggest that a direct biologic effect contributes to this sexual disparity. Although the contribution of some domains might be very specific to the setting of Vietnam, we expect that the residual biologic effect might be the same across settings. Further, we suspect that the M:F ratios observed in other surveys might be underestimated, given that confounders (e.g., with respect to access and exposure) were not addressed. Further research should focus on biologic aspects that might influence TB risk among men and women, for example, innate recognition of pathogens and antimicrobial immune responses. Recent mouse model data showed that estrogen in women boosts the potential of macrophages to kill bacteria that cause pneumonia (27). Compared with male mice, female mice also produced more IFNγ, an important cytokine that increases the antibacterial functions of macrophages (28). Data also suggest that testosterone is a mediator that inhibits the immune system by inhibiting such proinflammatory factors as TNF-α and nitric oxide and that male castration increases TNF-α secretion in mice (29,30). Aside from the influence of hormones, genetics might have a critical impact on improving women’s susceptibility to infections like TB since the X chromosome expresses several immune-related genes and immune-associated microRNAs. Women therefore may benefit from having 2 X chromosomes (31).

Our SEM analysis included all known and commonly occurring associated factors of prevalent TB that were collected in the PEER questionnaire, but we acknowledge that there may be other factors that were not measured. There also might be selection bias among the case group, since this group consisted of participants with only one positive Xpert test, which may include false-positive Xpert results, causing a slight underestimation of the associations. The prevalence survey used BACTEC MGIT 960 liquid culture (Becton, Dickinson and Company, https://www.bd.com/en-us) to diagnose TB, but the turnaround time was too long to inform selection for the PEER study; therefore, 2 participants selected as controls for the study who had MTB-positive culture results had to be excluded from the analyses. In addition, 83 survey participants who were Xpert-negative but culture-positive were missed being selected to participate in the PEER study. There were also 38 cases who were likely to be Xpert false-positive due to their previous TB treatment history. In the sensitivity analysis that excluded these cases, the SEM model yielded similar results. Another limitation of this study is that gender was not explicitly taken into account in our analysis. Indeed, some factors such as access to care and risk behaviors are likely to be more closely associated with gender than with sex, making the delineation between the 2 concepts in our study not straightforward. Neyman bias might have occurred because of our choice of prevalent TB as the outcome. Using prevalence survey data offers the benefit of minimizing sex bias in reporting. However, prevalence reflects both incidence and duration of disease, and we could not distinguish TB patients with different duration of disease in our analysis. Selection bias also might have occurred in our study, because the proportion of Xpert-positive TB cases found among the preselected controls was much higher than the proportion of TB cases found in the overall TB prevalence survey population. The participation rate among the preselected controls was also higher than that of the TB prevalence survey population (7). Like the TB prevalence survey, our analysis involved an undersampling in the youngest age group because the prevalence of TB among young persons in Vietnam was very low, which limited the allocation of participants in this age group (7).

In conclusion, we attribute the strong sex difference in TB prevalence found in Vietnam to a complex interplay of factors relating to behavioral and environmental risks, access to healthcare, and clinical manifestations. However, after controlling for all these factors, there remains a direct effect that is likely biological. Further insights from basic and clinical research are needed to explore this biologic difference. Aside from addressing other controllable factors, such as access to healthcare and behavioral and environmental risks, efforts to control TB should include effective strategies focused on men and reducing that specific disease burden.