Study Design and Participants
The study used data from the North Staffordshire Osteoarthritis project; a population-based prospective cohort study [
22]. All individuals aged 50 years and over registered with eight general practices were mailed a baseline questionnaire, in 2002, that collected data on health, individual socio-demographic factors and pain, and follow-up questionnaires 3 years later. Reminders were sent to non-responders 2 and 4 weeks after the initial mailing. The North Staffordshire Local Research Ethics Committee approved this study; all participants gave written consent to participate.
For this study we selected a cohort who (1) were aged 50–59 years old at baseline, (2) had hip, knee or foot pain for 1 day or more during the past year at baseline (to indicate lower limb osteoarthritis), (2) indicated that they were in employment at baseline and 3 years and (4) completed the items on work restriction at both time points (n = 716; mean age 54.5 (standard deviation 2.6 years), 54.7 % were female). Compared to subjects who were aged 50–59, had lower limb joint pain, were in employment and free of work restriction at baseline but withdrew, did not respond or had incomplete data at 3 years (n = 651), the participants in this analysis were no more likely to be older (p = 0.06), female (p = 0.71), have an inadequate income (p = 0.51), have better physical (p = 0.82) or mental health (0.61) but were more likely to have gone onto further education (p = 0.002).
Data Collection
Work restriction was measured by one item from the Keele Assessment of Participation (KAP) [
23]; “During the past 4 weeks, if you work, have you taken part in paid or voluntary work as and when you have wanted?” (all/most/some/a little/none of the time). The reliability and validity of the KAP are adequate for providing estimates of perceived participation restriction in population studies [
23]. Three year onset of work restriction was defined as moving from no restriction at baseline (all/most of the time) to work restriction at 3 years (some/a little or none of the time).
The independent variables in the analysis represented lower limb joint pain and functional limitation, comorbidities, age, gender, individual socio-economic, environmental factors and area-level socio-economic factors.
Lower limb pain and functional limitation were measured using the Western Ontario and McMaster Universities Osteoarthrits Index (WOMAC) [
24] for those with hip and knee pain and the Foot Disability Index [
25] for those with foot pain. The WOMAC offers a five point ordinal scale (none/mild/moderate/severe/extreme) to measure the amount of pain experienced during five tasks and the amount of physical limitation in seventeen tasks. The Foot and Disability Index consists of 19 items designed to measure the effects of foot pain on physical activities. Responses are on a three-point scale (none of the time/on some days/on most or every day). Respondents were categorised as having severe lower limb joint pain and functional limitation if either (1) those with hip or knee pain indicated “severe” or “extreme” pain in any of the five pain items or limitation on the sixteen items for physical functioning or (2) those with foot pain indicated foot pain “on most or every day” on any of the items of the functional limitation or pain intensity constructs (i.e. items 1–11, 14–17).
The co-morbidities included in this analysis were previously found to be associated with participation restriction in at least one aspect [
26]. These were: musculoskeletal comorbidity (number of affected body sites), number of self-reported health conditions, anxiety, depression, abnormal weight, and cognitive impairment.
A pain manikin was included to measure musculoskeletal comorbidity. The pain manikin allowed responders who had body pain over the previous 4 weeks to shade their painful body sites (0–44) on a full body manikin (front and back views). The number of shaded body sites was calculated and responses categorised into groups with approximately equal numbers of responders (0, 1–6 affected body sites, 7–44 affected body sites) [
20]. Number of health conditions was a simple count of the presence of four self-reported health conditions common in older adults (chest problems, heart problems, diabetes and raised blood pressure). Anxiety and depression during the previous week were measured using the hospital anxiety and depression scale (HADS)—raw scores were calculated and used to categorise individuals as non-cases (0–7) and possible/probable cases (8–21) [
27]. Cognitive impairment was measured using the Cognitive and Alertness behaviour subscale of the Functional Limitations Profile—raw scores were categorised to no impairment (score of 0) and cognitive impairment (score > 0) [
28]. Body mass index (BMI) was calculated from self-reported height and weight—responders were categorised into standard BMI groups (1) normal weight (BMI 20–24.9 kg m
−2), (2) underweight (BMI <20 kg m
−2), (3) overweight (BMI 25–29.9 kg m
−2) and (4) obesity (BMI ≥30 kg m
−2) [
29].
Individual socio-economic characteristics included were those previously found to be associated with participation restriction in the general population [
26]: occupational class (manual/non-manual) [
30,
31], educational attainment (those who finished their education on leaving school/those who went onto further education such as college or university) and perceived adequacy of income (adequate/inadequate) [
32].
Demographic details collected were age, gender, living arrangement (live alone/live with someone), and social networks (measured with the Berkman–Syme Social Network Index [
33]).
Data were collected by single items for three environmental factors relevant to work restriction: one to measure if responders required assistance or aids (i.e. “During the past 4 weeks have you required the assistance of others or aids to move around outside your home?”), one to measure access to transportation (i.e. “Do you have access to a car or public transport when you personally need it?”), and one to measure access to health care “Do you have good access to a GP or chemist?”). These items had a simple yes/no response option.
Area-Level Socio-Economic Factors
The development of the Index of Multiple Deprivation (IMD) 2004 [
34] for England has meant that seven specific socio-economic features of local areas (income, employment, health, education/skills/training, housing, crime, environment) can now be investigated for their effect on an individual’s health. We focused particularly on local area employment deprivation because of its relevance to work. Employment deprivation is conceptualised as involuntary exclusion from the labour market and the more working adults there are in an area that are unemployed, seeking work or on incapacity benefit the greater the employment deprivation, and in a sense this is a proxy for job opportunities and employment in the local area. By focusing on this we could examine if there was a link between good access to job opportunities and the onset of work restriction. The other six domains were included as putative confounders in the multivariate analysis. The index is based geographically at the lower level super output area (SOA) of which there are 32,482 in England with a mean population of 1,500. Subjects are allocated to a SOA based on their postcode. For each domain and for the combined scale, SOAs are ranked from 1 (most deprived) to 32,482 (least deprived). The SOAs from which the subjects in this study were drawn were split into tertiles for each domain of deprivation, the lowest one consisting of most deprived participants and the highest of least deprived participants.
Statistical Analysis
The frequency of onset of work restriction within each level of area employment deprivation (least, mid, most) was determined. Two level logistic multilevel modelling was then used to examine the associations of individual and area-level variables with onset of work restriction. Prior to the examination of associations, the variance components model (i.e. with no explanatory variables included) was derived, to assess the amount of variation in onset of work restriction that was at the area-level compared to that between individual subjects. The variation at area-level was calculated using the variance partition coefficient defined as
\( {{\sigma_{u0}^{2} } \mathord{\left/ {\vphantom {{\sigma_{u0}^{2} } {\left( {\sigma_{u0}^{2} \; + \;3.29} \right)}}} \right. \kern-0pt} {\left( {\sigma_{u0}^{2} \; + \;3.29} \right)}} \) where
\( \sigma_{u0}^{2} \) is the variance of the area-level random effect [
35]. The unadjusted associations of individual health, demographic, socio-economic and environmental factors with work restriction onset were then assessed.
The independent effect of each health, demographic, individual socio-economic and environmental factor and area employment deprivation on work restriction onset was then assessed over three stages with reference to the conceptual model of the International Classification of Functioning Disability and Health [
36]. In the first stage the “health” model was derived: all health factors were entered simultaneously into the model with age and gender as potential confounders. In the second stage a full multivariate model was derived: all variables significant at 5 % level or with OR >1.3 or <0.77 in the “health” model were included in the model together with the individual socio-economic variables significant in the unadjusted analysis or with OR >1.30 or <0.77 [
37]. In the third stage, due to correlation between area-level domains, each of the seven domains was added separately, adjusting for health, demographic and individual socio-economic factors in the multivariate model in stage 2. Associations are summarized by odds ratios with 95 % CIs.
Interaction terms were added to the multivariate model separately. First, we considered the potential for interaction with age to be of prime importance so we added an interaction term between age and (1) severity of joint pain and functional limitations, (2) number of areas affected body sites, (3) depression and then (4) employment deprivation. Second, to examine the role of employment deprivation we added an interaction term between employment deprivation and (1) severity of pain, (2) number of affected body sites and then (3) depression.
Analysis was performed using MLwiN 2.02 [
38] via residual iterative generalised least squares with the second order penalised quasi-likelihood approximation.