The association between silica exposure, silicosis and tuberculosis: a systematic review and meta-analysis

We carried out a systematic review to assess the evidence from human controlled studies conducted from the 1970s onwards for the associations between silica, silicosis and tuberculosis in adults. We excluded laboratory studies. We registered the protocol with PROSPERO (registration identification CRD 4201912696). Protocol development and review reporting were guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) checklist [35].

Inclusion criteria, grouped according to the population–exposure–comparator-outcome (PECO) framework [36], were as follows:

We excluded studies without comparison between higher and lower (or zero) levels of silica exposure or with insufficient information to assign or impute different levels of silica exposure; and for silicosis, without a comparison between those with and without the disease. We also excluded studies of latent tuberculosis infection or of tuberculosis self-reported or based on radiology alone, cross-sectional studies and mortality studies, solely autopsy based studies, and studies of tuberculosis cases from registers without the diagnostic method specified.

We searched the following healthcare and biomedical electronic databases up to 30 April 2020 using a comprehensive search strategy as outlined in Table S1: (1) PUBMED via http://​www.​ncbi.​nlm.​nih.​gov/​pubmed; and (2) EMBASE via www.​embase.​com. The strategy was translated into the appropriate syntax for each database (Additional File 1, Table S1). The search strategy included database-specific and free text terms for [silicosis] and [tuberculosis] and was not limited by study design filters, language, or publication date. We checked reference lists of studies screened as relevant, as well as review articles for additional relevant citations. Where necessary we contacted authors of published studies for information.

The selection process is presented in Fig. 1. Duplicate records were identified and removed. Two authors with expertise in the subject (RE, DR) independently identified potentially eligible studies based on article abstracts, applying the inclusion and exclusion criteria above. Full-text articles were obtained for further independent eligibility assessment. Disagreement was settled through joint re-assessment of the article and discussion with a third reviewer (NS).

Data from alternate articles were extracted by two of the authors (DR, RE) using a piloted template, checked by the other and disagreement resolved by discussion with NS. Data included first author, publication date, location/industry, study design, study population, calendar period of study, number of individuals included [i.e. silica exposed and unexposed, with and without silicosis (including grades of exposure/disease where available), and with and without tuberculosis], method of silica exposure estimation and/or silicosis identification, method of tuberculosis diagnosis, confounders measured and/or controlled for, and overall measure of effect (relative risk - rate, risk or hazard ratio, or odds ratio) by exposure or diagnosis category, both unadjusted and adjusted where available, with 95% confidence intervals. Where degree of silica exposure or silicosis was stratified, stratum-specific measures of association were extracted.

Risk of bias of each eligible study was assessed independently by two authors (RE and DR) using the Newcastle Ottawa Scale (NOS) for case-control and cohort studies, and disagreements settled through discussion and consultation with a third author (NS). The NOS was developed by the University of Newcastle, Australia, and the University of Ottawa, Canada, to assess the quality of nonrandomized studies included in systematic reviews [37]. There was no modification of the instrument for this review. The questions and score assignment for each study design are included as Table S2 (Additional File 1). Quality assessment scores were not used to exclude studies.

Confounding by risk factors for tuberculosis likely to be associated independently with silica exposure or silicosis, was considered in relation to study design. In these occupational (adult) cohorts, age, and in some settings HIV, were judged to be the most important potential confounders [38, 39]. Additional confounders of relevance, particularly in cohort studies that were population rather than industry based, or where the general population was used as the reference population, included smoking, congregate settings (in transport, housing and the workplace), socioeconomic status, undernutrition and indoor air pollution [40‐42]. Heavy alcohol intake and diabetes were considered as potential confounders as they are established risk factors for tuberculosis [41, 42]. Finally, bias arising from an increased likelihood of investigating for and diagnosing tuberculosis among those with silica exposure or silicosis was considered as a potential selection bias.

Where studies were sufficiently clinically and methodologically homogeneous, we conducted a meta-analysis for the selected outcome of tuberculosis. Aggregated participant data were used for data synthesis. Meta-analysis was conducted on RevMan version 5.3 [43] using the random effects model given anticipated statistical heterogeneity. We assessed the presence of heterogeneity in study results using the chi-square test and quantified the degree of heterogeneity using the I² statistic.

We preferentially report on the adjusted analysis using the estimate of effect reported in the study rather than calculating estimates of effects based on the crude data. Where only crude data were presented, we calculated the crude risk ratio and 95% confidence interval for dichotomous data if appropriate, and combined these with the adjusted estimates using the generic inverse variance function in RevMan 5.3 [43]. We explored anticipated heterogeneity by tuberculosis burden in country of study, i.e. high burden versus low and intermediate burden as defined by the WHO [4]. We further examined heterogeneity by study design (cohort vs case-control) and conducted a sensitivity analysis by examining the impact on overall findings of removing studies at risk of bias.

For those studies where we were not able to pool data, we provide a narrative and graphic synthesis using histograms showing relative risk or odds over different metrics for the individual studies. Disease severity-response gradients were analysed by extent of silicosis, and silica exposure-response relationships using stratified exposure metrics (e.g. cumulative dust exposure, duration of employment, or occupational dustiness category). To study the effect of silica exposure in the absence of silicosis, we included only studies which controlled for silicosis in the analysis though adjustment or excluded those with the diagnosis.

GRADE was used to judge the overall quality of the evidence with data directly imported from Revman into GRADEPro (GRADEpro Guideline Development Tool. McMaster University, 2020) [44]. Using this instrument in our context, overall quality reflects our confidence that the effect estimates are adequate to support an aetiological inference. Owing to the observational nature of the included studies, the overall confidence commences as low on the GRADE schema. We considered the following characteristics to mark quality up or down: risk of bias (individual study limitations); consistency, directness and precision of the evidence; and publication bias and selective reporting. We also considered the following as reasons to mark upwards: the magnitude of the effect, exposure-response gradient, and the likelihood that any residual confounding would have reduced rather than exaggerated the true effect [45].

The PUBMED and EMBASE searches yielded 1674 and 1607 records respectively. Following electronic deduplication, one author (NS) reviewed all potentially duplicate records and removed 798 true duplicates, resulting in a total of 2483 records (Fig. 1). From these, two of the authors (RE, DR) identified 34 potentially eligible records, of which 23 full text articles were obtained for full eligibility assessment. Fourteen of the 23 articles were excluded after assessment (Additional File 1, Table S3). Nine articles reporting on separate studies met the inclusion criteria.

Table 1 sets out the characteristics of the nine included studies [46‐54]. There were two case-control studies [50, 54], and seven cohort studies [46‐49, 51‐53]. Publication ranged from 1986 to 2013 and included five low or intermediate tuberculosis burden countries, Sweden, Denmark, Taiwan, Hong Kong (prior to unification with China), and Iran; and only one high tuberculosis country, South Africa, predominating with four studies. There were six industry-specific studies, including mining and quarrying (46, 48–51] and foundries (46, 47]; one regional population study [54]; one based on a national kidney disease register [53]; and one on a national silicosis register [52].

Figure 2 summarises the risk of bias in the reviewed studies by nine NOS characteristics, one set for case-control and one for cohort studies. In six of the studies [47‐52], all criteria were assessed as low risk of bias except for comparability of the study groups on “any additional factor” (other than age, and interpreted here as controlling for relevant additional factors) which was assessed as a source of uncertain risk of bias. Based on a close analysis of confounding and selection bias (See Additional File 1: Note and Tables S4 and S5), we concluded that although studies dealt with confounding in different ways, the overall risk of bias in these six studies was plausibly low.

There were additional sources of uncertain bias in the remaining three studies. In the Swedish registry study [46], there was an 11-year gap between the inception of the silicosis cohort and the first ascertainment of tuberculosis, and no information on how tuberculosis was excluded at baseline. In the Taiwan study [53] based on an end stage renal disease register there was lack of information on definition of silicosis and on how tuberculosis at baseline was excluded. Finally, the Iranian community study [54] was judged as having a high risk of bias owing to low response rates, uncertain risk of bias in the representativeness of cases and the lack of blinding in interviewing cases and controls about past silica exposure.

Titles and abstracts reported in languages other than English were excluded from the review but were screened for relevance. A single study, in Czech, was identified on abstract (as the article was unobtainable) which might have qualified for inclusion [56].

Table 2 summarises the results of eight studies of silicosis as a binary exposure and Fig. 3 presents the meta-analysis. All except Yarahmadi et al. [54] adjusted for age but differed in their other covariates. All effect measures, both crude and adjusted, showed a substantial effect of silicosis, ranging from 2.2 [51] to 32.99 [46]. The summary relative risk, preferentially combining adjusted estimates where these were reported with crude estimates, was 4.01 (95% confidence interval (CI) 2.88, 5.58) with I² = 53%, indicating moderate statistical heterogeneity. Given the large outlier effect size of Westerholm et al. [46], omission of this study resulted in a reduced summary relative risk of 3.70 (95% CI 2.78, 4.93) and statistical heterogeneity of 41%. Removing, in addition, the one study with an overall high risk of bias as described earlier [54] did not change the summary relative risk (3.69, 95% CI 2.59, 5.25).

The summary relative risk for studies in low and intermediate burden tuberculosis countries was 6.59 (95% CI 3.36, 12.94) with I² = 49%, indicating moderate statistical heterogeneity. Omission of Westerholm et al. [46] reduced this summary relative risk to 4.97 (95% CI 3.18, 7.77) and statistical heterogeneity to 14%. In high tuberculosis burden setting (all South Africa) the summary risk ratio was 3.16 (95% CI 2.31, 4.32), and statistical heterogeneity 35%. Sub-group analysis by study design (case control vs. cohort) did not explain residual heterogeneity within tuberculosis burden country strata, nor overall. The difference in relative risk between the two tuberculosis burden country groups was not statistically significant (p = 0.10) after omitting the outlier study (Fig. 3).

With regard to exposure-response, Table 3 summarises the findings of five studies which used grade of silicosis severity as the exposure, four cohort studies and one case control study. Since the ordinal scale did not allow pooling, a graphic synthesis is provided in Fig. 4. Two studies included some minor scale grades of the ILO classification for silicosis (0/1, 1/0, 1/1 and >  1/1), i.e. distinguished borderline categories from more advanced disease [50, 51], while a third used only the major scale grades (0, 1, 2 and 3) [48]. Hnizdo et al. [49] based their exposure-response analysis on subradiological (histological) silicosis and active tuberculosis detected at autopsy. Of note is that just under half to two thirds of the identified silicosis cases in these studies were at the lowest grades, i.e. ILO 1/0 or 1/1, or “negligible” to “slight” on histology. All controlled for age. Four were South African studies of goldminers - two controlled for HIV directly [50, 51], one covered a low HIV period [48] and the other a low HIV prevalence population [49]. Chang et al. [52] used progressive massive fibrosis as an indirect measure of severity. Table 3 and Fig. 4 shows a consistent monotonic increase in the risk or odds of tuberculosis with increasing grade of silicosis.

Using the GRADE schema (Table 4) we rated our confidence in the evidence for a strong aetiological association (relative risk > 2.5) between silicosis and tuberculosis as high. This judgement was based on a low risk of bias (following close consideration of confounding and a sensitivity analysis), a consistent, large effect size, directness and the presence of an observed exposure response gradient. To reduce across-study bias, all study outcomes were reported. However, there were too few studies to perform a funnel plot to exclude publication bias [57].

Table 5 and Fig. 5 summarise the results of five studies which reported on silica exposure and controlled for silicosis, either by exclusion [46, 47], modelling [50, 51], or both [49]. The meta-analysis is presented in Fig. 6. In all of these studies, the association of silica exposure with tuberculosis persisted after excluding radiological silicosis (including at least one of the study analyses in those studies in which two metrics were used [50, 51]. The effect measures ranged from 1.00 (dusty occupation vs none) [50] to 2.85 (any vs no silica exposure) [54].

The summary relative risk, preferentially combining adjusted estimates where these were reported with crude estimates, was 1.92 (95% CI 1.36, 2.73), with I² = 59%. This moderate statistical heterogeneity could be explained by study design difference: specifically, omission of the two case-control studies [50, 54] increased the relative risk to 2.13 (95% CI 1.70, 2.67) while reducing I² to zero.

The summary relative risk for studies in low and intermediate burden tuberculosis countries was 2.75 (95% CI 1.70, 4.45) with I² = zero. In high tuberculosis burden countries (South Africa) the summary risk ratio was 1.66 (95% CI 1.04, 2.66), and statistical heterogeneity high at I² = 73%. As above, omission of the case-control study [50] increased the relative risk to 2.11 (95% CI 1.68, 2.66) while reducing I² to zero. The difference in summary relative risk between the two tuberculosis burden country groups was not statistically significant (p = 0.14) (Fig. 6).

Using GRADE, we rated our confidence in the effect estimate for silica exposure and tuberculosis in the absence of silicosis as low (Table 6). Although exposure-response gradients and relative risks greater than two were observed in most of the analyses, this was not the case in all - we therefore did not mark up for magnitude of association nor exposure-response gradient. We considered there to be sufficient precision, consistency and directness not to mark down on these criteria. A risk of bias stemmed from the use of proxy metrics for silica exposure, with the potential for exposure misclassification bias [58]. Given that all of the studies except one were based on registers and not self-report, we regarded this risk as being non-differential with respect to tuberculosis, and therefore highly unlikely to have produced spurious associations. We therefore did not mark down further for risk of bias.

For radiological silicosis, an increase in the risk or odds of tuberculosis was seen at the ILO profusion grade 1/0 in two studies relative to the stratum of no silicosis [50, 51] (Table 3, Fig. 4). A third study used only major ILO grades and found an increased relative risk of 2.2 (95% CI 0.7, 3.6) at ILO grade 1 and 2.9 (95% CI 1.1, 4.6) at grade 2.

Threshold effects were difficult to infer from the four studies of silica exposure that controlled for silicosis and provided exposure-response gradients (Table 5) [47, 49‐51]). One showed no duration effect [51], while in the two that did, the first increment in risk was estimated across wide strata: 10–14 vs <  10 years [50] and > 15 years vs 0.5–14.5 years [47]. The same applied to the one study which measured cumulative dust exposure, i.e. 10–14 mg-years/m³ vs < 9 mg-years/m³ [49].

We have high confidence on the GRADE schema that further evidence would not change the conclusion that silicosis strongly increases the risk of tuberculosis, i.e. with a relative risk > 2.5 With regard to silica exposure controlling for radiological silicosis, the evidence suggests an elevated risk of tuberculosis with a exposure-response gradient. However, our confidence in the effect estimate is low and future studies, particularly those with more accurate measures of silica exposure, may change this estimate.

This is to our knowledge the first review to provide a meta-analysis and critically summarise the evidence of the association between silica exposure, and separately radiological silicosis, and pulmonary tuberculosis. The evidence covered spans the 1960s to 2013, with a large proportion of silicosis cases in the lower ILO (or histological severity) grades, distinguishing it from that of the first part of the century characterised by very high levels of silica dust exposure, severe silicosis and untreatable tuberculosis.

The co-modelling of silicosis and silica exposure in one study (which used both radiological and autopsy findings) [49] allowed an estimate of the combined independent effects of silicosis and cumulative dust exposure. The risk of tuberculosis in silicotic miners in the highest category of dust exposure was 13.4 times greater (multiplying the independent effects of 4.18 and 3.22 for silicosis and dust exposure category, respectively), than that of miners without silicosis in the lowest category of dust exposure.

Exposure-response gradients showed an increased risk of tuberculosis at an early radiologic grade of silicosis [50, 51]. Further, since a significant proportion of silicosis is subradiological [31, 32] the finding of an increased risk of tuberculosis at early histological grades [49] suggests an even lower threshold than that revealed radiologically. This is consistent with an earlier autopsy study showing an elevated proportion of tuberculosis even with a “slight degree of silicosis not detected radiologically in life” [59]. The excess risk of tuberculosis in silica exposed workers without radiological silicosis may thus be due to subradiological silicosis or cumulative silica dust each on its own, or to the combination. The implication of these findings is that radiological silicosis should not be required for attribution of the excess risk of tuberculosis to silica exposure in members of silica exposed workforces. However, there remains uncertainty about the threshold, in exposure or duration (as a proxy for exposure), at which excess risk of tuberculosis begins.

The review covered six countries, both high and low/intermediate tuberculosis burden, using different metrics of exposure. The overall estimates of effect for both silicosis and silica controlling for silicosis was higher in low/intermediate tuberculosis countries, arguably because of a relatively low baseline or population incidence of tuberculosis. However, only one high tuberculosis burden country, South Africa, was included. Workers on the South African gold mines and particularly black miners have a history of high rates of tuberculosis related to the migrant labour system and deep level gold mining going back 130 years. In addition to silica exposure, this system has entailed oscillation between mines and rural areas and congregate exposures in transport, accommodation and underground work, greatly aggravated by the HIV epidemic from the 1990s onwards [60, 61]. Studies are needed from other high tuberculosis burden countries such as India, China, Indonesia and Brazil, to determine if their patterns of risk differ.

A range of industries, occupations and settings were covered – gold mining, iron and steel foundries and caisson construction work [55] with one population based study covering a wide range of occupations [54]. This is, however, still relatively low coverage of the full range of occupations involving significant silica exposure, such as construction, agriculture, ceramics, stone work, sandblasting and fabrication of artificial stone. Besides differences in local tuberculosis epidemiology, differences in industry-associated potency factors for silicosis [62] may be relevant to the silica-tuberculosis association.

There has been recent international awakening to the fact that control of silica dust is important not only for preventing silicosis but also for control of tuberculosis [17, 18]. The summary relative risk (excluding one outlier study) of 3.65 (lower confidence limit 2.79) for silicosis and tuberculosis is of the same order as that of other common risk factors for tuberculosis [42] with the exception of HIV which is associated with a very high risk [39]. However, the impact of silica exposure on miners goes beyond the workforce. A modelling study of South African gold mining communities concluded that miners and members of mining communities contributed a disproportionately large share of new tuberculosis cases nationally in relation to their proportion of the population, although the main transmission impact was local [63].

Potential biases in the review process were controlled via a structured and transparent approach. The study protocol followed PRISMA and was registered on Prospero. Structured searches of PUBMED and EMBASE were undertaken and articles selected independently by two subject experts and agreed by consensus. Studies were limited to those published after 1970 in English; however, only one potentially relevant abstract was identified among those records not reported in English. A structured tool for assessment of risk of bias was applied independently by two subject experts and agreed by consensus. In-depth exploration of confounding was undertaken along the lines recently recommended [64]. In studies with more than one outcome, all were considered; however, there were too few studies to explore for publication bias. Finally, GradePro was used to assess the degree of confidence in the effect estimate for the two primary associations studied.

The findings are consistent with those from mortality studies. Workers in silica exposed occupations have been found to have standardised mortality ratios (SMRs) for tuberculosis ranging from 329 (95% CI 233, 452) [65] to 2175 (95% CI 1837, 2556) [66]. Studies specifically of radiological silicosis have shown similarly elevated SMRs, reaching 564 (95% CI 411, 754) in silicotic subjects who had claimed workers’ compensation [67].

We excluded mortality studies from the review and meta-analysis for several reasons. Death certificate studies generally provide no information on how the initial tuberculosis diagnosis was made. Substantial misclassification of tuberculosis as the cause of death has been shown in the South African gold mining context of very high rates of silicosis, tuberculosis and HIV infection [68]. Further, silicosis may contribute to tuberculosis mortality without an association with incident tuberculosis, for example, by creating diagnostic confusion and delaying TB treatment [69], or acting as an effect modifier by aggravating the course of tuberculosis through coexistent fibrosis, lung function impairment or reducing the effectiveness of standard treatment. Some such effect is suggested by the finding of the case fatality rate among those being treated for tuberculosis to be three times higher in silicotics than non-silicotics [70].

The findings are also consistent with cross-sectional studies (Additional File 1, Table S3) which reported positive associations between tuberculosis and silicosis and metrics of silica exposure controlling for silicosis [38, 58]. However, we excluded cross-sectional studies because of potential biases arising from exposure or disease related selection of workers with silicosis or tuberculosis out of the workforce prior to the sampling; underestimation of effect by including only prevalent active tuberculosis; or uncertainty associated with a diagnosis of past tuberculosis based on self-report or old radiologic features and the temporal relationship of such diagnosis with silica exposure.

The increased risk of pulmonary tuberculosis associated with silica exposure found in observational epidemiologic studies is supported by animal experiment studies. From as early as the first decades of the twentieth century, experiments on small laboratory animals showed that exposure to silica dust and tubercle bacilli resulted in greater proliferation of bacilli, faster development of tuberculosis or more severe disease in exposed compared to control animals [71‐75]. More recent work has shown that silica exposure markedly increases susceptibility to mycobacterial infection in mice, and that macrophages transplanted from exposed into unexposed mice reproduce this susceptibility without concomitant dust exposure [76]. Biological plausibility has been further strengthened by the demonstration or elaboration of possible underlying biologic mechanisms [77, 78]. Silica inhalation or silicosis should therefore be considered an effect modifier of the progression of tuberculosis from recent infection or latent infection to active disease.

Interpretation of studies of silica exposure, silicosis and tuberculosis needs to take the phenomenon of subradiological silicosis into account in terminology, causal contrast and interpretation. Computerised tomographic (CT) scanning may improve sensitivity to a variable degree [31] but remains impractical for large scale studies and medical surveillance.

There is a need for studies which are able to provide a more accurate measure of the dose of respirable silica dust, controlling for radiological silicosis, and its relationship to incident pulmonary tuberculosis. Given that exposure is a continuous variable, the goal should be a exposure-response curve, as has been achieved for respirable silica and silicosis [79]. This is best achieved through cohort studies, ideally prospective. Such studies would provide a more accurate estimate of the size of the effect of a given cumulative silica exposure on tuberculosis risk than presently available. For prevention purposes, a central question is the exposure threshold below which there is no excess risk of tuberculosis. The question of whether elimination of radiological silicosis, i.e. under current protective dust standards, would be sufficient to protect against tuberculosis, is a corollary of this.

Understanding of time dependent phenomena is also required. This includes the effect of short-term high intensity silica exposure on tuberculosis risk as opposed to long-term lower exposure; or the extent to which, in the absence of radiological silicosis, excess risk persists once silica exposure has ended.

More explicit attention to potential confounding is needed, in this case ensuring that risk factors for tuberculosis are evenly distributed across comparison groups or controlled in the analysis. Even if age is adjusted for, general populations may be an inadequate control for silica-exposed populations or silicosis cases without additional information on confounders. Including relevant covariates would also allow assessment of effect modification of the silica exposure/silicosis tuberculosis thresholds by covariates such as age, HIV, smoking and diabetes.