Table 1 shows the distributions of personal, clinical, and occupational characteristics of the full sample (N
1639), and also of those with and without self-reported knee symptoms. As expected, subjects with knee symptoms were older and more likely to be female; they also had higher BMI, reported more prior knee injuries, had a higher prevalence of rKOA (all with p<0.001) and had higher prevalence of either current or past smoking (p
0.05). Those with knee symptoms also had fewer years of formal education; they were more likely to be either currently or previously employed in workplaces with more physical demands (frequent squatting, standing and lifting), but more likely to be unemployed at the time of baseline examination (p<0.001) and less likely to be employed in workplaces with reduced-time accommodation (p
0.02) and better benefits policies (both with p<0.001). The duration of index employment was 13 (SD 11) years (13 (SD 10) for current jobs and for 13 (SD 11) last jobs).
Table 2 shows the availability of workplace policies in relation to sociodemographic features, individual lifestyle factors, other workplace characteristics and KOA-related prevalence. Statistically significant associations were found with age (for four policies; all with p<0.001), education (for reduced-time policy, p
0.002; for work benefits, both with p <0.001), race/ethnicity (for work benefits, both with p<0.05), frequencies of squatting (for disability payment, p
0.01), standing (for four policies; all p<0.001), lifting and walking (for reduced-time, p
0.03), at work (for work benefits, both with p<0.01), sex (for disability payment, p
0.02) and occupational group (with working in farming, forestry and fishing industries the least likely to offer better policies with p<0.001 except job-switch). There were no discernible differences or consistent patterns in the distributions of available workplace policies with regard to race, BMI or smoking behaviours.
Table 2Availability of workplace policies in relation to personal, clinical and workplace characteristics†
As shown in Table 3, the availabilities of these workplace policies tended to be clustered. Workplaces offering one benefit were very likely to offer the other, and workplaces providing one accommodation were likely to make the other also available. Also, there were appreciable positive correlations between accommodation and benefit polices.
Table 3The interrelation of available workplace policies
Results of multiple logistic regression analyses for the associations of knee symptoms and sKOA with workplace policies are presented in Tables 4 and 5. There were statistically significant differences in the frequency of knee symptoms and sKOA across six occupational groups, and self-reported physical demands were often significantly associated with more knee symptoms or higher sKOA prevalence (e.g., knee symptoms and lifting; sKOA and walking) in the adjusted models (numerical data not shown). We observed statistically significant negative association between knee symptoms and employment in workplaces offering better policies (Table 4). Lower prevalence of knee symptoms was found in workplaces offering better policies (job-switch: 31% vs. 42%; reduced-time: 34% vs. 40%; paid sick-leave: 33% vs. 48%; disability payment: 32% vs. 48%). Except for reduced-time policy, such differences remained statistically significant in the multi-variable analyses (adjusted Model I), although the magnitude of negation association was changed.
Table 4The associations between workplace policies and prevalence OR of knee symptoms
Table 5Associations of symptomatic knee osteoarthritis and asymptomatic radiographic knee osteoarthritis with workplace polices
Statistically significant lower sKOA prevalence was found in workplaces offering job-switch accommodation (8% vs. 13%), paid sick leave (9% vs. 16%) and disability payment (8% vs. 16%). In adjusted models (Table 5), for those in workplaces offering workplace benefits, the odds of having sKOA was approximately 42–46% lower than for those in workplaces without such these policies (for paid sick leave: adjusted OR 0.58, 95% CI 0.37 to 0.91, p
0.017; for disability payment: adjusted OR 0.54, 95% CI 0.35 to 0.84, p
0.007). These statistically significant associations were estimated from logistic regression models with reasonable goodness-of-fit to our empirical data, as suggested by the results of the Hosmer–Lemeshow tests (with one p
0.21 and all the others p>0.52).
Three propensity score models (with their areas under the receiver operating characteristic [ROC] curve equal to 0.75, 0.85 and 0.92, respectively) were constructed for the association of availability of disability pay with sKOA. Estimated propensity scores were all associated with increased prevalence of sKOA (with ORs 4.16, 2.53 and 3.10 associated with a one-unit increase in each estimated propensity score). However, adding each propensity score to the multiple logistic models did not alter the statistically significant negative association between disability payment policy and sKOA (estimated ORs 0.49 [95% CI 0.31 to 0.79], 0.54 [95% CI 0.32 to 0.91] and 0.55 [95% CI 0.30 to 1.00].
We found a lower prevalence of asymptomatic rKOA in workplaces offering job-switch accommodation (13% vs. 17%), paid sick leave (13% vs. 18%) and disability payment (12% vs. 27%). In adjusted models (Table 5), the odds of having asymptomatic rKOA among those in workplaces offering disability payments were estimated to be approximately 50% lower than those in workplaces without such a policy (adjusted OR 0.48, 95% CI 0.29 to 0.77). The estimated propensity scores were all associated with increased prevalence of asymptomatic rKOA (ORs 2.65, 2.25 and 2.24 associated with one-unit increase in each estimated propensity score), but further adjustment for each propensity score in the multiple logistic models did not alter the statistically significant negative association between disability payment policy and asymptomatic rKOA (with estimated ORs 0.52 [95% CI 0.37 to 0.74], 0.55 [95% CI 0.37 to 0.82] and 0.53 [95% CI 0.34 to 0.83]).
Additional sensitivity analyses were carried out to evaluate if the above findings were sensitive to further adjustment for KL grade, employing different ways to temporally characterise workplace policies, or confounding by other socioeconomic conditions. We found that the negative association between workplace benefits and knee symptoms prevalence remained statistically significant even after additional adjustment for KL grade (Table 4, adjusted Model II). When occupational variables were defined with different durations of employment, the estimated OR did not change substantially when using 1, 2 or 5 years as the cut-off (data not shown), and there was a tendency toward stronger negative associations as the employment duration increased (e.g., disability payment policy: adjusted OR 0.49 [95% CI 0.28 to 0.87] and OR 0.41 [95% CI 0.21 to 0.82] for choosing 10 and 15 years as the cut-off, respectively). Analyses restricted to the 1348 (82% of all eligible participants ) who provided additional information on household income (four levels: <US$10
000) and home-ownership showed the same robust negative association between workplaces offering disability payment and sKOA, even after also adjusting for either of these socioeconomic factors (data not shown).