After an accurate procedure of selection and evaluation of 178 studies, the following twelve were deemed as relevant (Table 2):

-7 case-control studies [18,19,20,21,22,23,24].

-5 cohort studies [25,26,27,28,29].

Study Characteristics

Population. In the case-control studies, the cases were classified as RA according to the ACR/EULAR classification criteria (American College of Rheumatology) [19, 20, 22] or the American Rheumatism Society (ARA) criteria[26]. In the case-control mortality studies [21, 23], subjects were identified through the specific death cause code ICD-9 (International Classification of Diseases, 9th revision) in their respective death certificates.

With regard to cohort studies, when declared, subjects with RA were selected using the International Classification of Diseases, 10th revision (ICD-10) [25,26,27] with M05 codes for seropositive rheumatoid arthritis and M06 for seronegative.

Exposures

Most of the population studied was dusty trade workers subject to moderate or high occupational exposure to FCS.

Controls

Subjects were randomly selected and frequency-matched to the cases by the principal indicators (e.g., age, sex).

Outcome

May the occupational exposure to FCS be a significant risk factor in the development of RA? [18,19,20, 22, 24,25,26,27,28,29]. May the RA mortality rate possibly be related to occupational exposure to FCS? [21, 23]. The studies were published between 1986 and 2017.

Exposure to FCS concerns different jobs (mainly “dusty trades”). The examined populations are predominantly men. Smoking habit was recorded and studied in four of the studies included in the meta-analysis.

Risk of Bias Within Studies

The quality and risk of bias appraisal was conducted using the Newcastle Ottawa Scale (NOS)[15] by two independent evaluators (AM, IS). We used a modified version of this tool in order to better adapt it to the studies we reviewed (see Supplementary Data S1 and S2).

An overview of the risk of bias for each of the studies included in this work is shown in Supplementary Table 2.

The NOS produced a final score of 7 [18], 6 [19], 8 [20], 7 [22], 5 [24], 7 [26], 4 [27], 3 [21], 4 [23], 5 [25], 7 [28], and 2 [29].

A few study in particular [21, 23, 27, 29] scored low on NOS scale, with significant shortcomings in both case selection and exposure assessment.

We found that the exposure assessment in all the studies has some critical limitations; this can be attributed to the use of self-completed questionnaires or a face-to-face interview in which the patient is asked to remember events dating many years in the past.

A widely used tool is the Job Exposure Matrix (JEM) [30] to estimate exposure to silica-based on work tasks, exposure measurements, and information on the work process[21, 23, 24, 27].

It should also be taken into account that personal Interviews might be subjected to recall biases. The exposure assessment in the case-control mortality studies [21, 23] is particularly weak, as the job and duration was inferred from the death certificate. These are often incomplete and inaccurate. It is also conceivable that the reported employment on the death certificate is attributable to the last job or to that mainly carried out with the risk of misclassification.

Cases of the different studies can be defined as dusty trade workers affected by RA and defined according to the American College of Rheumatology criteria (ACR). In this regard, it is important to note that over years, the classification criteria have undergone different revisions: in 1987 [18,19,20, 22] and in 2010 (ACR/EULAR criteria) [25].

The 1987 revised ACR classification criteria have been criticized for including late manifestations (e.g., rheumatoid nodules and radiological damage) failing to identify patients with early inflammatory arthritis. Nevertheless, they are able to classify established RA with high sensitivity and specificity.

With regard to cohort studies and mortality studies, the diagnosis of RA was defined according to the International Classification of Diseases (ICD) 10th [26, 27] and 9th Edition [21, 27]. ICD-10 is a classification introduced in 2003 and with greater specificity as compared with ICD-9. The specific RA codes used in ICD-9 were 714.0, 714.1, and 714.2 and in ICD-10 M06.9, M05 for seropositive rheumatoid arthritis and M06 for seronegative rheumatoid arthritis.

A non-negligible limit of the studies that use the ICD classification is to define RA+ patients based on the ICD-M05 code without distinction between antibodies for rheumatoid factor or citrullinated peptide.

In one study [25], the diagnosis of RA was self-reported by cases during the telephone interview, therefore with a high risk of misclassification or reporting bias.

All the studies controlled for the main and potential confounding agents (age, sex, geographical area) with a few exception [21, 29].

Several studies [32,33,34], reported a strong association between tobacco smoking and RA; however, only a few of the studies adjusted the OR for potential confounding from smoking [19, 20, 25, 27].

Some studies have also investigated the correlation between ever/never smokers exposed to FCS and RA [19, 22, 25, 27].

As usual, in these kinds of studies, participants were selected on a voluntary basis and this may introduce a selection bias.

Some main biases affected the identification of cases and controls; specifically, in the Schmajuk et al. study [25], patients were asked to self-report the diagnosis of their physicians and the possible intake of glucocorticoid treatment. In Yahya et al. study [19], both hospital and general population controls have been enrolled.

In the studies [21, 24] enrolling mine workers for long periods, we are aware that such individuals have a greater chance of multiple diseases and therefore false associations are more likely to report.

Synthesis of Results Primary Meta-Analysis Results

he meta-analysis of eleven studies [18,19,20,21,22,23, 25,26,27,28,29], applying a random effect model, yielded an overall OR of 1.94 (95% CI 1.46–2.58) with I2 = 95% (pronounced heterogeneity). This result was statistically significant (p ≤ 0.05).

Subgroup Analysis Results (By Autoantibodies)

We performed an additional meta-analysis on seven studies[18,19,20, 24,25,26,27], using a random effect, to investigate the relationship between occupational exposure to FCS and the effect on seropositive (RA+) or seronegative (RA−) RA patients, either for ACPA and/or RF (Fig. 2).

RA+ patients: We obtained the following result: OR 1.74 (95% CI 1.35–2.25 (I2 = 59%)). This result was statistically significant (p ≤ 0.05) (Fig. 3).

Fig. 3

Meta-analysis of studies including seropositive (RA+) RA patients

RA−: We obtained the following result: (OR 1.23 (95% CI 1.06–1.43) (I2 = 0%)). Significant association has been observed (p ≤ 0.05) (Fig. 4).

Fig. 4

Meta-analysis of studies including seronegative (RA−) RA patients

Subgroup Analysis Results (By Smoking)

Finally, we performed a further meta-analysis on five studies [18,19,20, 25, 27] to investigate the relationship between smoking habits and occupational exposure to FCS and the effect on RA+ development (Fig. 5).

Fig. 5

Meta-analysis of studies investigating interaction between occupational exposure to FCS, smoking habits, and seropositive (RA+) RA patients

Using a random effect, we obtained the following result: OR 3.30 (95% CI 2.40–4.54 (I2 = 49%)). This result was again statistically significant (p ≤ 0.05).

Risk of Bias Across the Studies (Funnel Plot)

We used the Funnel Plot, Egger’s test, and Higgins index to detect possible biases between the studies included in the meta-analysis (Supplementary Figures S6-S9).

Using the Higgins I2 index, we obtained the following results:

Any RA meta-analysis: I2 = 95% (Figure S6), RA+ meta-analysis: I2 = 59% (Figure S7), RA− meta-analysis: I2 = 0% (Figure S8) and RA+ , FCS and cigarette smoke meta-analysis: I2 =  49,% (Figure S9).

The visible asymmetry in the Funnel Plots confirms the presence of heterogeneity in the majority of the meta-analyses. The p-value of the Egger’s regression test resulted significant (p < 0.05) for publication bias, in all the performed meta-analyses.