Subjects and design
Data were collected cross-sectionally by postal questionnaire in 2001–2002, using previously validated measures of neck and upper limb pain and occupational exposure . North Staffordshire Local Research Ethics Committee approved the study, and recipients of the questionnaire consented to participation by completing the questionnaire and returning it to the principal investigator.
The study population was adults aged 18–75 years on the general practitioner (GP) database of the North Staffordshire District Health Authority, UK. We randomly sampled 10 000 adults equally from four age groups: 18–44, 45–54, 55–64, 65–75. Approximately 98% of the UK population are registered with a GP  and this sampling frame is considered representative of the general population . Assuming an exposed:unexposed ratio of 1:5, this sample size would provide ≥ 96% power to detect a 5% difference in prevalence between exposed and unexposed, at a two-tailed 5% significance level.
We sent a reminder to all non-responders two weeks after the first questionnaire mailing, and sent remaining non-responders another questionnaire after a further two weeks.
Measurement of outcome
The focus of this study was non-specific neck and upper limb pain; we did not attempt to identify specific diagnoses in specific body areas. For the primary outcome, we used a pre-shaded manikin question, in which the neck, shoulder, arm and hand areas are treated as one region . This question asked: "In the past 4 weeks, have you had pain that has lasted for one day or longer in any part of the shaded area?" (yes/no). Although interest in this study centred on the neck and upper limb region as a whole, we also asked respondents to shade a blank body manikin, from which we calculated the prevalence of pain in each of five areas within this region: neck; shoulder(s); elbow(s); forearm(s); hand(s) .
Measurement of exposure
We asked respondents to enter, in a grid, details of up to five most recent jobs (job title, area of work, start date, end date) held for at least 12 months. There was no limit to the length of time retrospectively in relation to which respondents could complete the grid. Occupational data were coded according to Standard Occupational Classification 2000 .
We asked whether or not ('yes/no' response) these jobs involved any of six activities – involving repetitive movements or sustained postures of the neck or upper limbs – on most or all days of the working week (which we deemed to represent substantial exposure):
• repeated lifting/carrying of heavy objects
• prolonged gripping/holding of an object
• bending the neck forwards for prolonged periods
• carrying out repeated movements with the fingers
• carrying out repeated movements with the wrists
• working with one/both arms at shoulder height or above.
From the jobs recorded, that held for the longest time was coded as the respondent's 'main job'. Psychosocial factors, relating to the main job only, were assessed by the following questions, on a five-point adverbial scale ('none of the time' to 'all of the time'):
• Can/could you control the way you worked in this job?
• Is/was your work physically demanding in this job?
• Do/did the tasks and activities that you perform/performed in this job change during your time in the job?
• Do/did you get job satisfaction from your work in this job?
• On the whole, are/were your supervisors/managers supportive?
Similar methods of collecting information on work activities and psychosocial factors – based on the control-demand model  – have been used previously [13, 22, 23].
Other neck and upper limb symptoms assessed included severity of neck and upper limb pain (0–10 numerical rating scale, categorized: 0–5 mild; 6–7 moderate; 8–10 severe ), time since onset of neck and upper limb pain (five-point ordinal scale: less than 4 weeks; 1 to 6 months; more than 6 months but less than 12 months; 1 to 5 years; more than 5 years). The questionnaire also included questions about previous neck or upper limb injury, spare-time activities involving repeated movements of arms or hands, and demographics. We used the Townsend Deprivation Index as a measure of deprivation .
The reliability of the pre-shaded manikin and of the measure of occupational history has previously been shown to be good .
For the primary outcome, we analyzed the association between a one-month prevalence of neck and upper limb pain and pottery work, compared to non-pottery work, as main job. Cross-sectional associations with neck and upper limb pain were estimated by odds ratios (ORs) with 95% confidence intervals (CIs). Associations between potential risk factors and neck and upper limb pain were investigated through chi-square tests. Multivariable analysis used binary logistic regression. Two such analyses were performed: (i) partial regression – covariates were age, sex, Townsend category (ii) full regression – covariates were age, sex, Townsend category, pottery work as main job, all work activities and psychosocial factors.
To investigate selection bias, associations between neck and upper limb pain and work in the pottery industry were compared across the three waves of questionnaire response, to simulate the effects of non-response. This assumed that factors leading to non-response resemble those leading to late-response, and that late responders are therefore most representative of non-responders; a similar strategy has been used previously . Confounding was investigated by adding to the full regression model: duration of main job, time since end date of main job, spare-time activities involving repeated movements of arms or hands, previous neck, shoulder, arm or hand injury, and comorbid pain marked on the blank body manikin. To account for recall bias, we repeated the full logistic regression (as described above) but additionally included interaction terms for each of the work activities and psychosocial factors with time since end date of the main job, quantified on a discrete numerical scale ranging from 0 years (current job is the main job) to 57 years.
The attributable fraction (AF) is the proportion of cases of a disease attributable to exposure to a particular risk factor or group of risk factors, or alternatively, the proportion of cases that would be prevented following elimination of these risks . The exposed AF (AFe) is the fraction of those cases with pain who were exposed who would not have been cases in the absence of the exposure . The population AF (AFp) is the fraction of all cases with pain (i.e. both exposed and unexposed) that would not have occurred in the absence of the exposure. The AFp takes into account not only the association between risk factors and outcome, but also the prevalence of the risk factors in the population as a whole, and therefore is a measure of the potential public health impact if the risk factors were removed . The AFs were used to estimate the proportion of neck and upper limb pain cases attributable to exposure to identified significant risk factors. They were calculated from relative risks, derived from adjusted odds ratios using an appropriate conversion method , after establishing significant association in the earlier analysis. Estimates of the AFp and AFe for separate risk factors, and of the AFp for categories of risk factors (physical and psychosocial), were calculated by direct age-standardization to the North Staffordshire population [27, 31].
Statistical significance was set at p ≤ 0.05 (two-tailed). Statistical analysis was carried out using SPSS version 12 (SPSS, Chicago, IL).