Methods

Data for this study derive from two cross-sectional, random-digit dial (RDD) population-based telephone surveys. The RDD surveys included landline and cellular phone sampling targeting exchanges likely to be in counties in Appalachia (including parts of Kentucky, Ohio, Pennsylvania, Tennessee, Virginia and West Virginia) with historically high mortality rates from CWP. We identified the targeted areas using data from the National Institute for Occupational Safety and Health.10 We limited eligibility for survey participation to males 50 and over who were English language speakers, had a history of any past or current labour force participation, and confirmed current residence in one of the targeted counties.

This analysis includes data collected in two separate RDD surveys with identical sampling methods. The initial sample, described previously, included 973 men surveyed in 2017.11 In the second sample, conducted in 2019, there were 62 948 call attempts resulting in 23 729 (38%) contacts with potential participants; 15 731 (66% of contacts) were ineligible because of age, sex, language, lack of labour force participation, or county of residence. Among eligible individuals, 5990 declined participation, yielding 2008 respondents with data available for analysis (25% of eligible contacts). For analyses combining both samples, the study population consisted of 2981 survey respondents (online supplemental figure 1).

Study interviews averaged 10 min in the 2017 survey and 18 min in the more extensive 2019 survey. Both surveys contained identically worded core items addressing employment, smoking history, sociodemographics and arthritis and related diagnoses. They both ascertained duration and type of coal mining experience, as well as exposure to inhaled dusts in non-coal mining jobs. The health sections queried whether the respondent had ever received a diagnosis from a health professional of arthritis of any kind, with follow-up items specifying RA, psoriatic arthritis or gout. The two surveys also elicited diagnoses of other autoimmune conditions, including systemic lupus erythematosus, psoriatic arthritis and systemic sclerosis. Interviewers asked about joint swelling, stiffness or pain, and asked those who responded affirmatively if they had ever received oral glucocorticoids (‘prednisone or steroid pills’) in treatment for those symptoms.

The 2019 survey included expanded batteries with items that were not present in the 2017 pilot detailing additional coal mining activities in order to further delineate exposure (see Coal and Silica Exposure) and additional questions related to treatments received for arthritis, in order to improve the specificity of the self-reported diagnoses of RA (see Disease Classification, below). The 2019 survey also restructured its assessment of ergonomic factors in order to better differentiate among higher intensity exposures (see Ergonomic Exposures Classification, below).

Arthritis was defined by an affirmative response to a primary stem question about receiving a healthcare provider’s diagnosis of arthritis. We defined RA based on a follow-up item about type of arthritis, further defined by also reporting having received glucocorticoids for joint symptoms. An alternative, more specific definition for RA, limited to items elicited in the 2019 survey only, required a diagnosis of RA and report of receiving at least one of a standard list of disease-modifying antirheumatic drugs (DMARDs): methotrexate, sulfasalazine, hydroxychloroquine, azathioprine, leflunomide, etanercept, adalimumab, infliximab, golimumab, certolizumab, tocilizumab, abatacept, rituximab or tofacitinib. As a sensitivity analysis, we also created a restrictive definition requiring both glucocorticoids and a DMARD. We defined a category of non-RA arthritis for all those who responded positively to the initial arthritis question, but did not meet the main study definition of RA. This category is likely to be predominantly degenerative arthritis (osteoarthritis), but includes reported RA without glucocorticoid use (or, in 2019 survey, without DMARDs) as well as infrequent reports of other autoimmune arthritis.

We defined coal mining based on survey-reported occupational history of any coal mining employment, with additional survey items eliciting underground and/or surface coal mining. Other silica dust exposure (not from coal mining) was defined by affirmative responses to any of a list of specific job tasks including: underground mining (other than coal); work with silica, sand or concrete dust; sandblasting; rock drilling or roof bolting (in other mining); rock crushing or quarry work; foundry work; concrete finishing, cutting or drilling; or masonry work or tip-pointing. The surveys did not specifically elicit employment in other silica trades that are less common in this region, such as glass-making or ceramic manufacturing. The expanded 2019 survey elicited more detailed information about coal mining exposure, allowing for identification of coal jobs that likely would result in higher intensity silica exposure: roof bolter, slope or shaft construction work, and, in surface mining, bulldozer, dragline or scraper operator. The 2019 survey ascertained the number of years spent working underground and the proportion of that time spent at the coal face, as that also confers higher silica exposure.

The 2019 survey elicited responses on multiple ergonomic exposures from any occupation grouped by the body part involved or physical movement (eg, hand/wrist twisting). The categories surveyed included: neck and upper extremities (three items); hands (two items); back (four items); feet and lower extremities (three items); and work activities involving vibration (three items). Respondents reporting any work-related ergonomic exposures in these five categories were asked to specify the number of years worked in such jobs. We used these responses to generate a summary score in which one point was given for any group in which the respondent endorsed all of its items and reported at least 5 years exposure. This method yielded a summary score ranging from 0 to 5. Although ergonomic exposures were elicited in the 2017 survey, the battery was more limited, had a lower threshold of positive response, and did not allow application of the same scoring schema; thus those data are not included in this analysis.

In both surveys, smoking exposure was assessed in the same manner, including age started, number of years smoked (for former smokers), and average number of cigarettes per day. This allowed for categorisations based on smoking status—never/former/current—and on number of years of exposure, yielding an estimate of pack-years of exposure. Respondents who had quit smoking fewer than 3 years prior to interview were considered recent smokers and included in the same category as current smokers. Those who did not report intensity or duration of exposure were excluded from pack-year analyses.

We compared the demographic, occupational and arthritis prevalence characteristics of the 2017 and 2019 samples. As shown in table 1, the frequencies of all characteristics were extremely similar, supporting a pooled analysis of the two samples for these measures. Using multivariable logistic regression analyses, we modelled the risk of all arthritis, RA (defined by glucocorticoid use), and non-RA arthritis associated with coal mining employment and other silica exposure, or either exposure. These models controlled for age and smoking status (current, former, never). We also tested interaction terms between coal/silica exposure and smoking status for the odds of disease.

Using the maximum likelihood estimates from the regression models, we calculated the population attributable fraction (PAF) of RA prevalence to estimate the proportion of prevalent cases among men that could be attributed to coal and/or silica exposure.12 We further estimated the odds of RA from models stratified by never versus ever smoking and, among ever smokers, stratified by former vs current or recent and by years of exposure (dichotomised at the median, 25 years), also testing interaction terms in these analyses. As a sensitivity analysis, we re-estimated the main models using generalised estimating equations (GEE) including the survey group (2017, 2019) as an additional variable.

Using the additional data only available in the 2019 survey, we modelled the risk of disease using the overall approach described above but adding the two specific definitions of RA with DMARDs and with both glucocorticoids and DMARDs. In these models, we additionally included an ordinal measure of ergonomic exposure from the 2019 survey based on score (0, 1, 2–3 or 4–5 points). We further defined coal mining exposure based on high intensity silica exposure tasks (available in the 2019 survey only) and on the number of years in underground mining. Each of these models also controlled for age and smoking (never/former/current).

To reduce the chance of misclassification bias, all models of RA (any definition) or non-RA arthritis excluded subjects who reported arthritis but did not meet the criteria for the diagnosis under consideration. For example, models of RA with glucocorticoids (main study definition of RA) excluded subjects who reported arthritis but not RA or who reported RA but not glucocorticoids. Thus, the reference population in each model is comprised only of individuals without any arthritis. Statistical analyses were carried out in SAS V.9.4 and Stata V.15.