Investigation of the risk factors for tuberculosis: a case–control study in three countries in West Africa

Abstract

The development of tuberculosis (TB) in man is a two-stage process in which a susceptible person exposed to an infectious TB case becomes infected and may later develop the disease, depending upon various factors. In individuals infected with Mycobacterium tuberculosis, any condition modifying the balance established in the body between the host's immune defences and the tubercle bacilli can have an impact on the risk of developing the disease. Factors that have been shown to influence this balance have been investigated for a long time, and were reported to be both ‘intrinsic’ and ‘extrinsic’ to the host.¹ However, in most studies investigating risk factors for TB, host-related and environmental factors were usually investigated separately, using different designs, making it impossible to assess their respective effects.^2,3 In addition, most of these studies were conducted in industrialized countries, and apart from the studies showing the effect of HIV infection,⁴ very few were carried out in resource-poor countries.² With the re-emergence of TB, affecting mainly developing countries, there is a need to re-examine the contribution of environmental and host-related factors for TB and identify factors that can influence the development of TB in man, in order to adjust and adapt TB control policies.

To explore these, we conducted a multicentre case–control study in three countries in West Africa, The Gambia, Guinée Conakry, and Guinea Bissau, from January 1999 to March 2001. These countries were chosen as they had almost similar ethnic mix, socioeconomic indicators and geographical environment, as well as a comparable epidemiological burden of TB to enable suitable recruitment. In each country, the study has been reviewed and approved by the local/national Ethics Committee, and informed consent was obtained from cases and controls prior to enrolment.

Methods

The detailed methodology of the study has been described elsewhere.⁵ Briefly, in each study site, newly detected sputum smear positive pulmonary TB patients >15 years who had presented at major urban health centres were eligible for inclusion. Pulmonary TB was confirmed by two consecutive smears positive for acid-fast bacilli and/or a positive culture. The household (in this study, household was defined as the extended family living together in the same area and eating from the same pot) of each case was visited and information collected on various demographic and socioeconomic variables.

For each case, two types of controls were recruited. First, a healthy control was selected at random within the household of the TB case. This control was age-matched to within 10 years of the case. Then, a healthy community control was selected at random in the neighbourhood of the case's household, by choosing a random direction from the case's home and visiting the fifth dwelling on the right. If, as commonly observed, several households lived in the same dwelling, one household was selected by drawing lots. After explaining the study to the members of the household, a control was identified, age-matched to within 10 years of the case. If there were several possible candidates within the household, the control was chosen by drawing lots. If the household head declined to take part in the study, the same procedure was repeated to select another household in the neighbourhood.

Information was collected from cases and controls on a wide range of potential host-related and environment-related risk factors for TB. These included factors already identified in previous studies^2,3 and factors expected to predispose to TB in the particular situation of developing countries. Standardized questionnaires were addressed to the study subjects by field assistants using the appropriate colloquial language, after checking the accuracy of translation with the interviewers. The presence of a BCG scar on the left or right deltoid region of the arm was checked. Blood samples were obtained from each case and control for a full blood count and HIV testing, after the patient was pre-test counselled and gave informed consent. Individuals positive for HIV infection were referred for post-test counselling and appropriate care, according to the national guidelines in use in each site.

Data were double entered using Epi-Info (version 6, CDC, Atlanta, GA) and transferred to STATA (version 7, Stata Corporation, College Station, TX) for analysis. Analysis was conducted on data from the three sites combined, using index case and household control pairs to assess the effect of host-related factors and index cases and community control pairs to assess the effect of environmental risk factors. Finally, cases and community controls were compared to assess the combined effect of host-related and environmental factors. Odds ratios (OR) and their 95% confidence intervals (CIs) were estimated using conditional logistic regression, with TB as an outcome. The likelihood ratio test was used to test the association between the explanatory variables and the risk of TB and for linear trend and interaction. Univariate analysis was performed to examine the effect of each variable of interest on the risk of TB. Multivariate models were then constructed, including variables that showed a significant statistical effect in the prediction of TB in univariate analysis (P < 0.05). In this paper, we present the results for the three sites combined. (Individual country site analysis was carried out in order to explore any specific aspect when data differed substantially among the three sites (large variations of effect or different direction of effect) but the results are not reported here.)

Results

During the study period, 846 cases, 702 household controls, and 828 community controls were recruited in the three countries. From the analysis, 24 TB cases were excluded, as smear positive pulmonary TB was not confirmed or because major data were missing. As a result, 822 index cases who satisfied the case definition were included in the analysis. Of these, 687 had a household control and 816 had a community control, allowing matched-pair analysis. The distribution by country is shown in Table 1.

Investigation of host-related factors

Results from univariate analysis carried out on the 687 cases/household control pairs are displayed in Table 2. As can be seen, male gender was found to increase the risk of TB in the three countries. Widowed/divorced and single individuals were at a higher risk of developing TB than married individuals. A former episode of TB in the case, or among his/her family, strongly increased that risk. The risk of TB was also found to increase with smoking, with a significant dose–response trend according to duration of smoking (P < 0.001), as well as with alcohol intake and drug use. Drinking tea in small groups of adults ('attaya') was not associated with the risk of TB. The presence of a BCG scar appeared protective against TB. HIV infection, history of worm infection, and recent treatment of worms were found to increase the risk of TB, but there was no effect of history or treatment of asthma. Finally, TB was associated with diabetes and anaemia, with a significant linear trend for decreasing levels of haemoglobin (P < 0.001).

A multivariable model was then constructed, based on the results of the univariate analysis. The inclusion of age-group as a categorical variable in the model did not change the effect of the other variables, indicating that there was no residual confounding of age. The haemoglobin variable was not kept in the model, as its inclusion dramatically reduced the number of pairs on which to perform the analysis. Former history of TB was also not included in the model, being on the causal pathway to TB. A total of 601 matched pairs of cases and household controls were available for analysis (Table 3). TB was found to be associated with male sex, family history of TB, HIV infection, and smoking, with a persistent dose–response effect according to duration of smoking.

Investigation of environmental factors

A total of 816 matched case–community control pairs were available for the analysis (Table 1). Environmental factors were investigated using a wide range of variables collected at the household level (Table 4). The risk of TB increased with the number of households in the compound, and with the number of persons in the household. More particularly, the risk increased with the number of adults in the household (P-value for linear trend <0.01) although TB was not associated with the number of persons per room in the household. The risk of TB increased if houses were built with mud-brick compared with cement walls, and if the floor was made of earth/soil rather than cement, but these effects were not statistically significant. The absence of a ceiling in the house was protective against TB. The risk of TB was not associated with the number of windows in the house. Variables assessing utilities and hygiene in the household (water source, electricity, presence of latrines, and waste disposal) were not associated with TB.

As observed with household controls (see above), the risk of TB increased with a reported family history of TB. More specifically, the risk increased with the number of persons in the household having had TB in the past (P-value for linear trend <0.001). Based on further data collected in The Gambia and Guinee Conakry (data were not available for Bissau), it was found that the risk of TB was even higher if the former TB case was among closely linked family members (father, mother, son, daughter, brother, or sister) than among more remote family members (OR: 1.60, 95% CI 0.62–4.18), although this effect was not statistically significant (data not shown).

At an individual level (Table 4), the risk of TB was higher in singles than in married individuals. Skilled manual workers and farmers were the most at risk of TB compared with unskilled manual workers. There was no association with religion, years of schooling, and the type of school (secular vs koranic). The risk of TB was not affected by ownership of the house by the case/control or his/her family. There was no association either with ownership of specific items or with the presence of animals in the household.

A multivariate model was then constructed, including variables of interest that showed a statistically significant association with TB both at the household and individual level, adjusting for sex and HIV infection (Table 5). The risk of TB was found to be associated with marital status, family history of TB, number of adults in the household, and ownership of the house by the case's family.

Combined host and environmental factors

Finally, in order to assess together the effect of host and environmental factors, a model was built including in the above model host-related variables that were found to have an effect in earlier analyses (see Investigation of host-related factors), using the case-community control pairs (n = 688 pairs) (Table 6). The risk of TB was found to be higher in males than in females. Among host-related factors, not surprisingly, TB was associated with HIV infection. History of asthma appeared strongly protective against TB in this population. Smoking was consistently found to increase the risk of TB, with a significant dose–response effect according to duration of smoking. Among social and environmental factors, family history of TB, marital status, number of adults in the household, and ownership of the house remained independent risk factors for TB.

In order to evaluate a possible country effect, we tested in the multivariate model for a possible interaction of each variable with country. No interaction was found for sex, smoking, HIV infection, marital status, family history of TB, or ownership of house, which showed a consistent effect on TB throughout study sites. However, there was an interaction between number of adults and country, the effect on the risk of TB being higher in Bissau (OR: 6.19, 95% CI 2.25–16.96) than in Conakry (OR: 3.15, 95% CI 1.25–7.94) and in The Gambia (OR: 1.99, 95% CI 1.05–3.77) (P-value for interaction <0.01). The effect of history of asthma was consistent throughout sites, but strongest in The Gambia. Finally, socioeconomic variables showed variable directions of effect according to countries, reflecting the complexity of the situation, but the effect of ownership of house was consistent in the three sites.

Discussion

The present study is one of the first to examine risk factors of TB consistently in a large sample and in a multicountry design, allowing the investigation of the respective effects of these factors and their consistency between countries.

Two types of controls were recruited in the study. The recruitment of household controls was based on the assumption that the effect of potential host-related risk factors would be best estimated in situations where the effect of environmental factors could be controlled for. Thus, it was expected that variations in the risk of TB observed with given variables would truly reflect the effect of these variables for a given environment.

As recruitment of community controls was difficult in the absence of a sampling frame (street address, postcode, telephone number, and recent census enumeration), we decided to select control households by choosing a random direction from the case's home and visiting the fifth dwelling on the right.⁵ The nature of this sampling process for controls means that households, rather than persons, were selected with equal probability. This was considered appropriate, as community controls were selected mainly for the evaluation of environmental risk factors. However, a person in a small household would have had a higher probability of selection as a community control than a person in a large household. In order to account for this potential bias, we did control for household size in the multivariate analysis.

Recall bias occurs in case–control studies if patients remember past exposures differently from controls. In this study, recall bias might have led to the underestimation of the effect of intercurrent disease (asthma and worm infection), as sick persons are usually more likely to remember a pathological event than healthy persons. For that reason, we limited our question on concomitant disease to a short recall period (1 year), and differentiated between history and treatment of given diseases.

Another concern in case–control studies is that the disease may affect exposure, so the results could in fact be a sign of reverse causality. This could be the situation with smoking, as, in the sites of the study, health workers strongly advise newly diagnosed TB cases to give up this habit. For that reason, we collected information on the duration of smoking, in order to differentiate past from current smoking, so reverse causality is unlikely to explain the findings.

The observed association of TB with male sex is in agreement with what has been reported in most countries in Africa, where notification rates are higher for men than for women.^6,7 Whether the male excess among the study population reflects a selective under-reporting or under-diagnosis of female patients or a true background sex difference is difficult to assess.⁸ There are considerable sex differences with regard to stigma and its social consequences, which may result in differential health-seeking behaviour and access to care between males and females.^9,10 However, the consistent effect of male gender observed, both with household and community controls and throughout the sites, suggests that selective under-diagnosis relating to gender is unlikely. Of note, a genome-wide linkage study searching for regions of the human genome containing TB susceptibility genes suggested a linkage between regions of the chromosome X and TB, with a lod score of 1.77, which could contribute to the excess of TB in males in many populations.¹¹

Smoking results in histological changes in the lower respiratory tract, including peribronchial inflammation, fibrosis, vascular intimal thickening, and destruction of alveoli.¹² This leads to alterations in the epithelial function, such as reduced ciliary activity, decreased clearance of inhaled substances, and abnormal vascular and epithelial permeability. In this study, cigarette smoking was an independent risk factor for TB, with a clear dose–response effect with duration of smoking. These results are consistent with those reported in two case–control studies carried out in Spain and in India, where a similar dose-dependent association between tobacco smoking and pulmonary TB was reported.^13,14 Our findings thus support the hypothesis of an increased vulnerability of smokers to the infection and development of TB, most probably owing to patho-physiological changes in the lungs induced by chronic smoking.^12,13 As smoking is becoming an increasing public health problem in resource-poor countries, especially in the younger age groups, these results reinforce the need to take appropriate public health measures, following the recommendations contained in the WHO Framework Convention on Tobacco Control, issued in 2003 by the WHO.¹⁵ This matter should also be addressed in TB control programme policies, and TB control staff should be trained in providing advice to TB patients and their relatives on the cessation of smoking.

Since the start of the pandemic, HIV infection has emerged as the most important risk factor for the development of TB in individuals infected with M. tuberculosis.⁴ According to WHO, ∼30% of TB cases arising among the 15–49-year-old adults in sub-Saharan Africa are attributable to HIV,⁶ and a continuous 10% per year increase in TB is projected in countries most severely affected by HIV infection.¹⁶ The major factors contributing to the HIV associated TB epidemic in Africa are the high risk of reactivation of latent TB infection in HIV infected persons and the high risk of progressive TB disease owing to HIV infection, associated with the high TB transmission rate in the community. Therefore, it is essential to include HIV infection as a risk factor when trying to identify the role of other factors on the risk of developing TB.¹⁷

The presence of a BCG scar was not independently associated with protection against TB. The sensitivity of scar reading is extremely variable, and BCG scar size has been shown to depend on vaccine strains and doses, prior tuberculin sensitivity, and age.¹⁸ A study in Malawi showed that BCG scar was a highly sensitive indicator of vaccination status when vaccine was properly handled and given at over 3 months of age, but in infants <1 month of age vaccinated in health centres, sensitivity of BCG scar had declined to <80% by 4 years post-vaccination.¹⁹ According to the authors, this low sensitivity may be attributable to vaccinations being carried out under non-optimal health centre conditions—which is likely to be the situation in the three sites of this study, where BCG vaccinations are given at birth or first contact with health services. However, it must be noted that our study was not set up to assess the efficacy of BCG in the study sites, but our data are consistent with others reporting an absence of a relationship between BCG scar and protection against TB in adults in Africa.²⁰

We observed a consistent effect of former experience of TB within the household, and this effect increased with the number of persons who had TB in the past. Furthermore, there was some evidence that this effect was higher when the former TB case was in close family link with the index TB case, as compared with unrelated household members. We showed earlier that, within the households of TB cases, the risk of TB infection increased with social proximity to the case, and that this effect was persistently higher in first-degree than in more distant relatives.²¹ This clustering of TB within families, already noticed by epidemiologists decades ago,^22,23 could reflect not only the facility of transmission within the household, but also a genetic contribution to the susceptibility to TB.²⁴ This aspect has been investigated in the present study²⁵ and results from combined case–control and linkage analysis have so far showed that vitamin D receptor variants might have an impact on susceptibility to TB.²⁶

We found that the risk of TB was associated with the number of adults in the household. Data were consistent across the three countries, with a stronger effect in Bissau, probably related to the particular housing structure in that country, with all adults sharing a single airspace. Classically, studies have measured crowding as the density of persons per room in a given household.^27–29 The fact that we did not find an association between TB and the number of persons per room, but found an independent association between TB and the number of adults in the household is probably related to the wide mobility of persons within the household, polygamy, as well as to the variability of housing within the sites.

Adult crowding reflects the increased likelihood of coming into contact with persons excreting the bacilli in crowded environments, but is also a marker of socioeconomic status (SES). TB has indeed always been associated with poverty. In industrialized countries, TB was classically associated with poor living conditions.^29–33 However, few studies were conducted in resource-poor countries, to investigate the effect of socioeconomic factors on TB and their results were inconclusive: while some studies found that the risk of TB was higher in poor living areas,^34,35 others failed to show an association between low SES and active TB in adults,³⁶ or even showed a reverse association.¹⁷

SES is difficult to measure, and there are no uniform criteria to assess it. Former studies conducted in industrialized countries to evaluate the effect of poverty on the risk of TB have been heavily dependent on the definition of specific indices and on the methodological design used.² Most of these were ecological studies based on the measurement of SES indices at the population level,^28,37–39 but few were based on an assessment of SES at the individual level.^32,40 In addition, indicators developed in the north might not be appropriate for the south, owing to socioeconomic and cultural differences.² In the present study, we collected information on a range of variables expected to reflect SES. Several of them showed some degree of association with TB in univariate analysis, but they did not have an independent effect on the risk of TB in multivariate analysis, when adjusting on main demographic variables, showing probable correlation. Information on income was difficult to obtain, owing to the fact that income-generating activities were frequently carried out within the ‘informal’ economic sector or were highly seasonal and variable in time. In addition, it was difficult to assess whether the income of the case/control would indeed reflect the overall income of the household. However, ownership of the house by the case's family appeared to be a consistent marker of SES throughout study sites.

Marital status similarly reflects both demographic and SES of individuals. In West Africa, widowed and divorced persons, especially women, rely heavily on family solidarity, and in the absence of this solidarity, may experience severe socioeconomic difficulties. Single persons are usually younger than married persons and have a different lifestyle, especially males, who often migrate to towns in search of a job where they frequently live alone or with friends. Thus, in Bissau, being single and living alone was an independent risk factor for TB.⁴¹

Subjects with TB were less likely to report a history of asthma than community controls. A putative link between exposure to mycobacteria and a decreased risk of atopic disease was suggested in a study of Japanese children in whom strongly positive tuberculin responses in early life were associated with a reduced risk of asthma in later childhood.⁴² In an ecological study carried out in 23 industrialized countries, TB notification rates were inversely associated with lifetime prevalence of wheeze and asthma.⁴³ These findings are consistent with the hypothesis that natural exposure to M. tuberculosis in childhood may have a role in inhibiting atopy. It has been shown that mycobacteria elicit a Th1 immune response in humans leading to the production of IL-12, IFN^*, and TNF^*,⁴⁴ which has been shown to cross-inhibit Th2 responses.⁴⁵ Thus, it is possible that mycobacterial exposure decreases the development of atopy through the inhibition of Th2 immune mechanisms, which is consistent with both the above epidemiological findings and our earlier findings of a high Th1 activity in exposed household controls.⁴⁶

In conclusion, this combined analysis carried out in three countries in West Africa confirmed the multifactorial aspect of TB disease, with host-related factors interacting with the environment to contribute to the overall phenotype. Findings of the study were presented and discussed with NTCP staff in the three countries. This led to recommendations for the improvement of TB control with the identification of specific targets, such as enhanced health education on TB and its risk factors and improved case-finding activities within the families of TB cases.

The authors wish to thank all field assistants who carried out the study with much dedication in each site. This study was funded through a grant from the European Commission (DG Research, contract IC18CT980375). K.F. is supported by the Medical Research Council, UK.

References