The association of mobility disability and obesity with risk of unemployment in two cohorts from Sweden

Design, study population, and data

This is a prospective cohort study based on population-based surveys and one national register with up to 16 years of follow-up. Samples from two separate populations including people of working age (ages 19–64) were included. The first study population consisted of participants who took part in the National Survey of Living Conditions (ULF/SILC) between 1996 and 2011 (baseline). This survey is conducted annually by Statistics Sweden on random nationwide samples of the Swedish population [20]. The second study population was based on pooled data comprising individuals from two of the Stockholm Public Health Surveys (SPHS) conducted in 2002 and 2006 (baseline). SPHS are based on stratified random samples from the total population of Stockholm (capital of Sweden) County (n = 1,900,000), aged 18–84 years [21]. The two study populations were followed up in the Longitudinal Integration Database for Health Insurance and Labour Market Studies (LISA) between 1997 and 2012 in terms of unemployment (outcome). LISA is a national database administered by Statistics Sweden and contains annually updated labour market data of all individuals 16 years of age or older who are registered in Sweden [22]. Further, information on disability benefits (part-time or full-time) was collected from the Swedish Social Insurance Register between 1990 and 2012. Information on mortality and emigration was taken from the Cause of Death Register and the Immigration Register, held by Statistics Sweden.

Exposure

In the SPHS cohort, mobility status was evaluated by a question originating from the EuroQol EQ-5D-3 L self-rating scale [23]. Here, people who had answered “yes” on one of the following two alternatives, “I have some problems in walking about” OR “I am confined to bed”, were categorized as having MD. Self-reported height and weight were used to calculate body mass index (BMI) (kg/m²) in both cohorts according to WHO classification [24]. The weight categories used were normal weight (BMI 18.5–24.9), overweight (BMI 25–29.9), and obesity (BMI ≥ 30). Finally, six exposure groups were created at baseline (ULF/SILC: 1996 to 2011; SPHS: 2002 or 2006) based on combining MD and weight status: normal weight without MD (reference), normal weight with MD, overweight without MD, overweight with MD, obese without MD, and obese with MD.

Study cohorts

Complete information on the exposure variables, i.e., height, weight, and MD status, was required for eligibility. People with underweight (BMI < 18.5 kg/m²) were not considered in the current study. Individuals with incomplete records or with implausible values of weight (≤ 40 kg or ≥ 250 kg), height (≤ 150 cm or ≥ 210 cm), or BMI (≥ 80 kg/m²) were excluded. To account for the healthy worker selection into work [25], individuals who before baseline had disability benefits [n(ULF/SILC) = 1963; n(SPHS) = 2385] or who had been long-term unemployed for ≥180 days [n(ULF/SILC) = 133; n(SPHS) = 657] were excluded from the study samples. The final ULF/SILC sample (Additional file 1) comprised 43,163 individuals (study cohort included 39,947 individuals) with complete information on mobility status, weight status, and the covariates used in the study. The final SPHS sample (Additional file 2) comprised 43,834 individuals (study cohort comprising 40,088 individuals).

Outcome

In Sweden, there are two official ways to measure unemployment. The first way is by using the EU harmonized labour force surveys. The other way is by using unemployment records from the Swedish Employment Service (registered in the LISA database). In this study, the latter approach is used. Information on number of days in unemployment was retrieved from the LISA database. Three unemployment outcomes were defined: 1) time to first occurrence of any unemployment (≥1 day), 2) time to first occurrence of long-term unemployment (at least 90 accumulated days during two consecutive years), and 3) the average number unemployment days per year during the follow-up. Survey participation only happens once a year; however, it is technically possible to participate in more than one survey during the study period e.g. an individual can participate in a ULF/SILC survey 2000 and then later in 2008. Follow-up started the year an individual for the first time participated in either a ULF/SILC or SPHS survey and ended the year of any unemployment (1) or long-term unemployment (2), year of retirement (age of 65), year of first disability pension (1996–2012), year of emigration or death, or the end of the follow-up period (31 December 2012), whichever came first (yielding individual follow-up times between 1 and 16 years).

Confounding factors

Statistical analysis

Baseline characteristics

The reported baseline results are based on all six exposure groups; however, the overweight groups are excluded in Table 1 to improve the readability. In the SPHS analytical sample (n = 40,088), 8.9% were obese (Table 1), and 5.0% had MD (Table 1). Almost 22% of the individuals in the groups with MD were obese compared with 8.1% in the groups without MD (Table 1). In the ULF/SILC analytical sample (n = 39,947), 8.8% were obese, and 2.5% had MD. Approximately 21% of the participants in the groups with MD were obese compared with 8.4% in the groups without MD. In both samples, individuals with MD more often were women, were older, had lower education, were employed in unskilled and non-manual type of labour, and had lower disposable income compared with members of the groups without MD (Table 1). The percentage of people born outside Sweden among the groups with MD was notably higher in the SPHS cohort.

Any and long-term unemployment

In both cohorts, individuals of normal weight without MD had the lowest unemployment rate, and considerably higher rates were seen in the other groups, especially those with MD (Table 1).

During the follow-up period for the ULF/SILC cohort (mean: 5.7 years; SD: 4.2 years), 10,845 (27%) individuals were unemployed at least once, whereas 7590 (19%) individuals were long-term unemployed. Results from Model 2 (the main model) show the obese group without MD had a higher relative risk of being unemployed compared with the reference group (Table 2). Similar relative risks were observed for the groups with MD (HR range: 1.11–1.34; 95% CI range: 0.88–1.81) compared with the reference group. Short-term unemployment before baseline (< 180 days) was, as in the SPHS cohort, inversely related to subsequent unemployment (data not shown). After adjustments for short-term unemployment before baseline (Model 3) similar results to those of the main model (Model 2) were observed, although the relative risk appeared to increase for the obese group with MD. Notably the obese group without MD had a stable relative risk increase (20–30%) of being unemployed short- and long-term compared with the reference group regardless of adjustments made.

Days of unemployment

For the ULF/SILC cohort results from the main model (Model 2; Table 3, upper part) show no statistically significant differences in average (median) unemployment days between any of the exposure groups and the reference group. There was; however, an indication of more unemployment days on average for the obese group with MD than for the reference group (Table 3, upper part). Further, the results from the main model (Table 3, lower part) showed a higher relative risk of being unemployed more than 30 days for the obese non-disabled group compared with the reference group (RR = 1.19; 95% CI = 1.08–1.32). A similar result was observed between the obese group with MD and the reference group (RR = 1.31; 95% CI = 0.97–1.77). Adjusting further for short-term unemployment before baseline (< 180 days, Model 3) did not materially change the estimates observed from the main model (Table 2).

Main findings

This study followed two population-based cohorts over a long period of time which included vulnerable groups of people with obesity and MD at risk of unemployment. Two different definitions of MD were used and several unemployment risk outcomes were investigated. We observed an overall trend of increased unemployment risk for all the exposure groups compared with normal weight people without MD. The results were similar and robust for both cohorts. Further, no excess unemployment risk of MD and obesity combined was observed beyond the excess unemployment risk of MD or obesity alone. Instead, MD and obesity remained separate risk factors after taking into account healthy worker selection into work as well as confounding by various socio-demographic factors.

No apparent excess burden of co-existing MD and obesity in terms of unemployment risk

People with MD and obesity (double exposed group) were not at higher risk of unemployment than people with MD only or obesity only (single exposed groups). The assumption was that over time people with MD and obesity risk more severe comorbidities, and perhaps other social adversities, that would put them at higher risk of being unemployed compared to having only one of the conditions. An association of poor health and subsequent unemployment has been demonstrated. A study of Finnish middle-aged men with lower occupational status (construction workers) has demonstrated that both various health problems (OR range: 1.97–7.75; 95% CI range: 1.01–39.93) and employment history (OR: 2.13; 95% CI: 1.20–3.80) independently predicted long-term unemployment [35]. Another study found that middle-age people with poor self-rated health, especially with physical disability such as musculoskeletal pain (OR: 1.93; 95% CI: 1.65–2.27), were at an exceptionally great risk for long-term unemployment [36]. We tried to account for potential comorbidities between the double- and single-exposed groups by excluding people with long-term unemployment before follow-up started. Moreover, in Sweden disability benefits are, since 2003, almost exclusively awarded to individuals with permanently reduced work capacity associated with a medical diagnosis. By also excluding people with any form of disability benefits (most often people with MD) this would, at best, help to ensure that any potential long-term unemployment differences between the double- and single-exposed groups were not due to a history of poor health. Thus, in light of the study findings, it is possible that any unemployment differences between the double- and single-exposed groups are best explained by group differences in socio-demographic factors and comorbidities.

Further, people living in Sweden who have longer or permanent employment contracts are protected against job termination due to poor health, but it is possible that potential differences in health between the double- and single-exposed groups could influence unemployment risk for individuals on shorter and temporary work contracts. Unfortunately, more detailed information, such as the employment type (permanent or temporary) and length (full-time or part-time), was unavailable; thus, we can only speculate whether potential health differences between the double- and single-exposed groups could influence subsequent unemployment risk differently depending on the job individuals hold. Further, qualitative information on reasons for being unemployed could also be valuable when investigating how conditions such as MD and obesity with poor health influence future unemployment risk between groups, but this was outside the objective of this quantitative study. Another possibility is that people with MD may experience more severe comorbidities with increasing obesity levels (BMI ≥ 35) that would put them at greater risk of being unemployed. The study samples had few respondents with more severe obesity levels (low statistical power), which could explain why we did not observe any noticeable relative risk difference between the double- and single-exposed groups. Last, it is possible that any excessive health problems potentially experienced by the double-exposed group, compared with the single-exposed groups, are more likely to increase their risk of being permanently excluded from the work force through disability benefits, rather than contributing to a greater risk of unemployment. Such relationships has been more frequently observed in the Nordic countries with highly developed welfare systems [13, 24].

High unemployment rates of people with MD and/or obesity

We found that people with obesity, and people with MD regardless of weight status, are groups who are more likely to be unemployed than people without these conditions. The results were robust and of similar magnitude between the study cohorts despite being different in terms of the MD definition used. The results are in line with previous research investigating other aspects of work participation in similar groups [12, 26, 27, 37, 38]. One study found that people with MD had approximately 2–4 fold higher odds of not participating on the labour market than people without MD during 8 years of follow-up [12]. Another study showed that people with MD were much less likely to be working than those with no or other disabilities (OR: .28, 95% CI: .21–.38; p < .0001) [38]. The association of obesity as a risk factor for subsequent unemployment is poorly understood, and not without contradiction. While a study found a higher risk of unemployment in women (OR: 2.0, 95% CI: 1.2–3.4), but not in men [39], two other studies found no meaningful association of being obese and subsequent risk of unemployment, after taking into account multiple confounding factors [40, 41]. Further, work-related factors such as a work accommodations, including flexible working schedules, and supportive work environments, have been suggested to play a key role in retaining people with MD and/or obesity [26, 29, 32, 37]. Low occupational status has been associated with cumbersome and less-skilled occupations with poorer conditions and non-supportive work environments [42]. People with MD and/or obesity are more likely to hold these types of jobs [37, 43], partly due to lower education and financial problems, possibly because of increased health care expenditures. Although detailed information on work-related factors was unavailable in this study, information on occupational status (besides educational level) was used to account for some of the confounding effect of work-related factors, but it is likely that some unmeasured and residual confounding remains.

The potential confounding of short-term unemployment (defined as less than 180 unemployment days) before baseline, on the associations under study was explored [28, 33, 34]. The results show that short-term unemployment before baseline was higher in the groups with MD than in the groups without it, especially the SPHS cohort, as well as inversely associated to subsequent unemployment in both cohorts. The SPHS is a population-based cohort from the county of Stockholm. Since people with MD are more likely to hold temporary and insecure job contracts [28, 43] than people without MD, they may be more likely to be unemployed in an urbanized setting where the labour market concentrates and where job competition is high. Moreover, the influence of the economic downturn during the study period may have had a greater impact on unemployment for people MD [28]. Nevertheless, based on both the data available in this study and follow-up time, it is very hard to disentangle both the temporality and potential causal mechanisms of short-term unemployment before baseline on chronic conditions such as obesity and MD; thus, we can only speculate about its influence. Most importantly, the overall unemployment risk differences between the groups in both cohorts were of similar magnitude when comparing a model including short-term unemployment before baseline to a model without it (the main model of this paper). Moreover, the results indicate that adjusting for short-term unemployment before baseline may further increase the unemployment risk in the groups with MD (bias away from the null).

Another important aspect to consider is the type, severity and duration of the condition underlying a reported MD in the study participants. Apart from accidents and other traumas, there are many physical, mental, and other chronic health states that may underlie a disabling condition [44]. In a study from New Zealand, it was suggested that the longer people were disabled, the higher was their risk of being unemployed [45]. In a Korean study of people with MD, the probability of being unemployed increased dramatically with increasing MD severity [27]. Moreover, there is some evidence from working age populations showing that unemployment risk increases the longer people have lived with obesity, and this association appears stronger in women [46–48]. It is likely that these factors have a strong influence on the association between MD and obesity with the risk of unemployment over the life-course, although we were unable to investigate the magnitude and direction of such an influence in the current study.

Employer prejudice and discrimination may explain part of the unemployment gap observed in this study between people with MD and/or obesity and those without these conditions [29, 31]. However, it is difficult to separate the impact of discrimination from that of health-related productivity differences between the groups in this study. Prejudice and discrimination may act both on wage and on employment prospects for people with MD and/or obesity, but the impact varies by factors such as gender, and the severity and type of the disabling condition.

In addition to residual confounding, it is possible that other factors influenced the association investigated in this study. It is; however, beyond the scope of the current paper to summarize and discuss all potential pathways to unemployment for people with MD and/or obesity, which represents a task better left for systematic review.

Strengths and limitations

The current study used two large population-based samples of the Swedish working population, which allowed for the identification of rather large groups of people with MD and/or obesity. Further, using different definitions of MD between the cohorts allowed investigation of the robustness of the findings. The study participants were followed up over a reasonably long period of time in a national register with high coverage and almost no loss to follow-up regarding objective unemployment outcomes [22] for people of working age who live in Sweden.

Some limitations need to be discussed. First, non-response is always a problem when using information from population-based surveys. In the ULF/SILC surveys, non-participation has increased in the last decade and reached levels beyond 50% in 2013 [20]. The SPHS surveys had participation rates of approximately 60% in 2002 and 2006. Non-response rates have been higher for people of younger ages, with less education, and who were born outside Sweden. In the current study, younger participants were more often found in the group without MD and with normal weight, whereas participants with lower education and those born outside of Sweden were over-represented in the groups with MD and/or overweight. Even after accounting for these confounding factors in the regression models, it is difficult to exclude residual confounding. Second, information about the underlying comorbidities of MD and obesity, such as cancer, CVD, and diabetes were not included. It is likely that these factors mediate the association of MD and obesity with risk of unemployment, but this lies outside the study objective and would therefore wrongfully attenuate the estimated association of interest. Moreover, these factors certainly influence more permanent pathways out of the workforce for people with MD and/or obesity, for instance through disability benefits [15]. Instead, we tried to account for bias of health selection into unemployment [35, 36] by excluding people who were on disability pension or who had been long-term unemployed for more than, or equal to, 180 days before the start of follow-up. Last, information on aspects of MD is often limited in health surveys, and thus its definition and prevalence differ between countries [1]. In this study, information on mobility status was obtained from self-reports, but with the possibility of using two rather different definitions of MD, which yielded similar results between the study cohorts. Information on BMI was also calculated using self-reported height and weight, and misreporting has been shown to exist with respect to weight and height in population-based surveys [49]. Further, more accurate information on body size and fat mass through measures of bioelectrical impedance and waist circumference could have provided a more accurate obesity prevalence among people with MD [15, 50]. In the current study, the prevalence of severe obesity is most likely underestimated, which may partly explain why we did not observe any clear differences in unemployment rates between the groups with MD.