Obesity is an important and increasing public health problem, being a risk factor for overall mortality and major chronic diseases, including cardiovascular diseases, diabetes mellitus, cancer and musculoskeletal disorders.1^–3 Many studies have focused on the effect of obesity on mortality. However, consideration of only fatal events may underestimate the consequences of obesity.4 5 For example, several studies have found an effect of obesity on disability in the elderly, and the association appears to be comparable to or even greater than the association between obesity and mortality.6 7

In working age adults, obesity may result in productivity losses through reduced workforce participation, increased work limitations, obesity related sick leave and work disability.8 9 An increased body mass index (BMI), which is the most common proxy measure for overweight and obesity, has also been linked to greater costs to employee health plans10^–12 and to workers’ compensation claims.13 Certain types of work related injuries may be associated with increased weight, such as musculoskeletal problems in the lower back,14 15 knee16 17 and hip.18^–20

Although permanent occupational disability is a great burden to the individual and extremely costly for society in countries with an established welfare system providing disability pensions, only a few studies have addressed the role of obesity as a risk factor for work disability.4 21^–25 In contrast to workers’ compensation, which usually only covers cases which are closely work related, work disability pensions in Germany usually encompass cases of reduced work capacity irrespective of the underlying cause. Evidence from the few studies available suggests that risk of work disability increases with BMI, and that in working age adults, overweight and obesity may be more strongly related to the onset of work disability than to all-cause mortality.24 It has been reported that overweight and obese employees experience a two- to threefold risk of work disability compared to their normal weight peers.22 This association may be even stronger for specific outcomes, such as musculoskeletal disorders or cardiovascular disease.23 24 26 27 Similarly, the strength of the association might also depend on smoking status and physical activity.28 However, little is known regarding these specific aspects and as both obesity and work disability have emerged as important public health problems during recent years, we studied the relationship between BMI and work disability in a large cohort of construction workers followed over a 15-year period.

METHODS

Study population

The study cohort at baseline comprised 19 421 male employees aged 25–59 years from the German construction industry who were working as bricklayers (n = 6204), painters (n = 2947), labourers (n = 2874), plumbers (n = 2804), carpenters (n = 2594) or plasterers (n = 1998). They participated in a routine occupational health examination conducted by the Institution for Statutory Accident Insurance and Prevention in the Building Trade in Württemberg (a region with about 5.4 million people in the south of Germany) between 1 August 1986 and 31 December 1992. This occupational health surveillance is based on legislation on health and safety at work and regular examinations are offered to all construction workers. Although participation is voluntary, over 75% of all invited employees underwent the medical examination during the period of recruitment and were eligible for follow-up. The participants were representative of the underlying population of all construction workers with respect to age, nationality and occupation. The study was approved by the ethics committees of the medical faculty of the university clinics of Heidelberg and Ulm and by the Baden-Württemberg state ministry of social affairs.

Data collection

Baseline examination

The health examination at baseline was part of the routine occupational health surveillance and includes taking a medical and occupational history, information on lifestyle factors, a physical examination (including measurements of height and weight), functional measurements (such as electrocardiograms, lung function, audiometry and visual acuity), as well as blood and serum analysis, which were carried out according to a standardised protocol.

Follow-up

Information on the date and cause of work disability was obtained from the German pension fund in March 2006, while the cut-off date for follow-up was June 2005. This time lag is needed because some cases of disability pension are approved in retrospect. The mean duration of follow-up per person was 10.8 years over the entire cohort. The cause of work disability was coded according to the International Classification of Diseases (ICD-9) by trained medical officers from the pension fund. In case of missing data, which occurred mainly due to some workers leaving the population covered in the regional pension fund, we also included information from the previous follow-up investigations in 1992–199429 and 1998–2000.30 Data on employment, work disability or retirement due to old age as well as vital status and date of death are held in regional offices of the German pension fund. This information is available as it is directly linked to payments to and from the pension fund. However, these data have to be deleted because of confidentiality if the subject moves to another region of Germany or abroad. In cases of migration after previous follow-up investigations, follow-up information from those follow-up investigations can still be used. The pension register provided information regarding vital status and whether the individual was still working, had retired due to age, was unemployed or in rehabilitation, or whether a disability pension (permanent or temporary) had been granted. In case of transient or multiple temporary disability pensions, the first occurrence of disability was taken as the endpoint for the analysis.

The criteria for being work disabled and receiving a disability pension are being revised, but during the follow-up time a disability pension was granted in Germany when earning capacity had been permanently reduced by at least 50% because of injury, illness or impairment – irrespective of whether the injury was caused by work or not – and when the worker was unable to undertake another adequate occupation. A patient applies for a disability pension at the local insurance office, which requires a health certificate from the applicant’s primary physician. In addition, a physician employed by the pension insurance institution examines the applicant and judges whether the patient meets the criteria for a disability pension or whether a rehabilitation measure might be appropriate first.

Statistical analysis

For this analysis, we had to exclude 415 men with missing, incomplete or invalid measures of height or weight at baseline (2.1%). Another 2131 men (11.0%) who had either moved to a different region or had changed employment were also excluded. The very strict confidentiality rules in Germany did not allow us to follow these people further. We used information from earlier follow-up investigations for 1427 workers (7.3%) with missing data in the current follow-up. Hence, the final study population for this analysis comprised 16 875 construction workers who could be successfully linked with the pension register.

For this analysis, BMI (kg/m²) was categorised as lower than 20.0, 20.0–22.4, 22.5–24.9, 25.0–27.4, 27.5–29.9, 30.0–34.9, and 35.0 or higher. These seven categories allow a detailed examination of the association between BMI and work disability across a wide range of body mass values without a priori assumptions about the shape of the dose–response curve. The majority of cut points were chosen according to the BMI classification scheme proposed by the World Health Organization and the US National Institute for Health, which distinguishes between underweight (BMI <18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25.0–29.9 kg/m²), mild or class I obesity (30.0–34.9 kg/m²), moderate or class II obesity (35.0–39.9 kg/m²), and severe or class III obesity (BMI ⩾40.0 kg/m²).31 32 However, in this analysis, we choose 20.0 rather than 18.5 as lowest cut point, because very few construction workers had a BMI <18.5 and because this cut point may be less suited for physically hard working men. Furthermore, we combined the categories of class II and class III obesity as there were very few construction workers with a BMI ⩾40.0 kg/m².

Techniques of survival analysis were employed to assess the impact of BMI on the occurrence of work disability. We defined the onset of being work disabled as the point of time from which a disability pension was granted, irrespective of the date of ascertainment of work disability. A person was denoted as being censored if they were known not to have received work disability compensation according to pension fund records, or their pension fund insurance had been terminated because of other reasons, such as receipt of a retirement pension, their 65th birthday, death or change to another insurer, or (in analyses of cause-specific disabilities only) work disability due to another cause.

Relative work disability hazards according to the level of BMI were calculated using Cox’s proportional hazards model. We separated the follow-up into earlier (up to 3 years) and later periods after checking the proportional hazards assumption with log-log survival plots.

This categorisation was also made based on the consideration that later periods might be less affected by potential bias from possible modification of baseline BMI by pre-existing disease. We also considered age, nationality, smoking status and alcohol consumption at baseline as covariates in the multivariable analysis. For age, a linear and a quadratic age term were simultaneously entered into the model, while index variables were created for the other, categorical variables. With respect to nationality, the construction workers were grouped as German, former Yugoslavian, Italian, Turkish and other. We included nationality as covariate to account for potential confounding due to a “healthy migrant effect” as indicated by a lower work disability among foreign study participants.

Further baseline data were available on comorbidity at baseline such as chronic respiratory diseases (ICD-9: 491–506) and cardiovascular disease (ICD-9: 390–459) (both as recorded by the physician). As chronic respiratory diseases and most cardiovascular diseases (namely hypertension, ischaemic heart disease and stroke) may represent intermediate steps in the causal pathway between obesity and work disability, they were not included as confounders in the multivariate modelling. All comorbidities were determined by reviewing all of the diagnoses ascertained by the physician.

Regarding smoking status we distinguished between never, former or current smokers with the latter subdivided into smokers of up to 10, 10–19, or at least 20 cigarettes per day, and pipe, cigar or other smokers. Alcohol consumption was ascertained by average daily amount of alcohol which was calculated from the frequency and type of beverages consumed. For this analysis, the following categories were used: none, 1–30, 31–60, 61–90, and more than 90 g alcohol per day.

Information regarding smoking and alcohol consumption status was missing in 16% and 9% of all subjects, respectively. Thus, we used multiple imputation to fill in the pertinent missing baseline data for smoking status and alcohol consumption according to the subject’s age and nationality. The results of the regression analysis based on the imputed data were consistent with results using either only complete cases or adding index variables to indicate subjects with missing information. To prevent statistical drawbacks caused by categorisation, BMI was also entered as a continuous variable in supplementary regression models using fractional polynomials as described by Royston et al and Sauerbrei et al.33 34

Additional analyses were carried out in specific age groups (25–39, 40–49, 50–59 years), in subgroups according to the presence or absence of defined types of comorbidity and smoking status and with respect to cause-specific work disability. Within these subgroup specific analyses, we used broader BMI categories to prevent too imprecise effect estimates (<25, 25.0–29.9, 30.0–34.9 and ⩾35.0 kg/m²).

All statistical analyses were performed with the SAS statistical software package, release 9.1 (SAS Institute, Cary, North Carolina, USA).

RESULTS

Baseline characteristics

Characteristics of the study population (overall and grouped by BMI) are shown in table 1. Of all 16 875 men included, 5863 (35%) were in the middle and upper range of normal BMI (20.0–24.9 kg/m²), 317 (2%) were underweight or in the lower range of normal weight (BMI <20 kg/m²), 8131 (48%) were overweight (25.0–29.9 kg/m²) and 2564 (15%) were obese (BMI ⩾30.0 kg/m²). Bricklayers were the largest occupational group in our sample with over 30% of the participants. The mean age of the study population at baseline was 41.9 years, and approximately 76% of the cohort members were of German nationality. Smoking was very common in our study with 58% being current smokers.

Construction workers with higher BMI levels were on average older and more often non-smokers than subjects with lower BMI. At 44%, almost half of the study population had musculoskeletal disorders at baseline, mostly back problems. We observed an increased prevalence of cardiovascular disease (total 31%), liver/bile/pancreas disease (13%) and diabetes mellitus (5%) at baseline with increasing BMI, while the prevalence of chronic respiratory diseases and gastrointestinal disorders as well as the proportion of healthy persons without any recorded comorbidity (9%) strongly decreased with increasing BMI.

All-cause work disability according to BMI

There were 3064 incident cases of work disability during the mean follow-up time of 10.8 years per person. Table 2 presents the association between BMI and all-cause work disability. BMI in the middle normal weight range 20–22.4 kg/m² was used as the reference category. Crude analysis revealed a J-shaped association between BMI and all-cause work disability with a very strong, almost threefold risk increase in obese men with moderate to severe obesity (BMI ⩾35 kg/m²). This pattern changed substantially after adjusting for age, which led to a U-shaped relationship with a moderately but significantly reduced risk associated with overweight and mild obesity (25.0– 34.9 kg/m²). Further adjustment for nationality, smoking status and alcohol consumption did not materially alter the observed associations.

There were 515 cases of occupational disability (17% of all cases) during the first 3 years of the follow-up period. Risk of short term work disability was possibly somewhat higher in lean subjects after adjustment for all covariates, whereas the increase of work disability in the highest BMI class was small and not statistically significant.

After exclusion of the first 3 years of follow-up, the results were very similar compared to the analysis without left truncation. After adjustment for age, nationality, smoking status and alcohol consumption, we again observed a U-shaped relationship between BMI and all-cause work disability, with a lower risk at moderate overweight and an increased risk at moderate to severe obesity (BMI ⩾35 kg/m²). This U-shaped association was also observed when BMI was entered as a continuous variable in supplementary survival regression analysis using fractional polynomials (data not shown).

The increased risk for severely obese participants (⩾35 kg/m²) was only observed among men without diabetes mellitus or cardiovascular disease and not among men with these disorders. We also stratified for the presence or absence of gastrointestinal diseases or chronic respiratory diseases at baseline because these diseases might cause weight loss and therefore could influence the outcome. The risk increase in moderately to severely obese persons (⩾35 kg/m²) seemed to be restricted to men without these diseases, but confidence intervals for hazard ratios in obese men with these disorders were broad and still compatible with both a possible risk increase or decrease. Additional analyses with stratification by the presence or absence of musculoskeletal diseases, the most frequent comorbidity in this group, did not indicate any relevant confounding or interaction by this condition (data not shown).

Work disability according to BMI and cause of disability

The frequencies of cause-specific work disability are shown in table 3. Information on the cause of disability could be obtained for 91% of all cases of work disability. With 1279 (46.2%) cases, musculoskeletal disorders represented the most common cause of work disability, with half of them being due to back problems. The second most common cause was cardiovascular diseases (17.1%), followed by mental disorders (8.9%) and cancer (8.1%).

After exclusion of the first 3 years of follow-up, the risk of work disability due to osteoarthritis increased very strongly with BMI (p value for trend <0.001) (fig 1A). Compared to men with a BMI <25.0 kg/m², the relative hazard of work disability due to osteoarthritis was 1.53 (95% CI 1.12 to 2.10), 1.78 (1.21 to 2.60) and 3.65 (2.01 to 6.62) for BMI ranges 25.0–29.9, 30.0–34.9 and ⩾35 kg/m², respectively. By contrast, there was no association between BMI and work disability due to back problems (fig 1B).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Work disability due to musculoskeletal disorders (A: osteoarthritis; B: back problems) according to body mass index after exclusion of the first 3 years of follow-up. Hazard ratios (log scale) are adjusted for age, nationality, smoking status and alcohol consumption. A body mass index (BMI) of 20.0 kg/m² was used as the reference. Results of dose–response modelling by fractional polynomials29 30 are shown. Dashed lines represent 95% confidence limits.

With respect to work disability due to cardiovascular diseases, we also observed a strong positive association with increased BMI (p value for trend <0.001), which is shown in fig 2. Compared to normal weight men with a BMI <25.0 kg/m², the relative hazard (95% CI) of work disability due to cardiovascular diseases was 1.12 (0.88 to 1.44), 1.38 (1.01 to 1.87) and 3.73 (2.39 to 5.83) for BMI ranges 25.0–29.9, 30.0–34.9 and ⩾35 kg/m², respectively.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Work disability due to cardiovascular disease according to body mass index after exclusion of the first 3 years of follow-up. Hazard ratios (log scale) are adjusted for age, nationality, smoking status and alcohol consumption. A body mass index (BMI) of 20.0 kg/m² was used as the reference. Results of dose–response modelling by fractional polynomials29 30 are shown. Dashed lines represent 95% confidence limits.

In contrast, we found a strong inverse association between BMI and work disability due to cancer (p value for trend = 0.003) with relative hazards (95% CI) of 0.60 (0.44 to 0.82), 0.51 (0.31 to 0.84) and 0.81 (0.30 to 2.23) for BMI ranges 25.0–29.9, 30.0–34.9 and ⩾35 kg/m², respectively (fig 3). This inverse association was particularly strong with respect to work disability due to lung cancer, but even after exclusion of lung cancer an inverse association remained with respect to work disability due to cancer (data not shown).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Work disability due to cancer according to body mass index after exclusion of the first 3 years of follow-up. Hazard ratios (log scale) are adjusted for age, nationality, smoking status and alcohol consumption. A body mass index (BMI) of 20.0 kg/m² was used as the reference. Results of dose–response modelling by fractional polynomials29 30 are shown. Dashed lines represent 95% confidence limits.

Considering work disability due to other diseases, we found an inverse J-shaped association with BMI, with strongly increased risk in lean subjects and only a slight non-significant increase in obese men, whereas risk was lowest for men with a BMI between 25.0 and 30.0 kg/m². The excess work disability among men with BMI <25.0 kg/m² was mainly due to accidents and mental and gastrointestinal diseases.

Further age specific analysis did not provide any major modification of the association between BMI and all-cause mortality. No significant interaction was revealed by investigation of interaction terms between the aforementioned four BMI categories and the age groups 25–39, 40–49 and 50–59 years. Stratified analysis by occupation revealed that the relationship between BMI and work disability seemed to be strongest among painters and weakest among plumbers (data not shown), but the differences were only marginal and mostly non-significant. Likewise, no interaction was found by age, nationality, smoking and alcohol consumption.

DISCUSSION

We found a J-shaped association between BMI and all-cause work disability in crude analysis with lowest risk at the normal BMI range (20.0–22.5 kg/m²). After controlling for age as confounding factor and other covariates an U-shaped association emerged, with a nadir at 25–27.4 kg/m². Occurrence of work disability in the overweight and moderately obese range (BMI 25.0–35.0 kg/m²) was decreased by approximately 10–20% compared with men in the normal BMI ranges <25 kg/m². Exclusion of the first 3 years of follow-up did not materially change this pattern. The overall pattern was found to result from the combination of strongly divergent associations of BMI with disabilities from different causes. In comparison to the association of BMI and mortality in the same cohort, where we also observed lower rates among overweight than among normal weight men,35 36 relative risk of work disability was somewhat higher in all overweight and obesity classes.

Cause-specific analysis revealed that musculoskeletal disorders, especially back problems, were the most common cause for being work disabled in this cohort of physically hard working men. This finding is consistent with expectations, given that construction workers are exposed to heavy physical and static work such as carrying and lifting of loads. In contrast to findings from a population-based cohort study from Norway,21 BMI was unrelated to disability retirement due to back problems in our study, which may point to a different meaning of BMI in a cohort of physically hard working men such as ours. By contrast, the strong association of BMI with disability due to osteoarthritis is consistent with findings from a Finnish cohort23 and in line with other epidemiological studies that have observed a positive association between BMI and osteoarthritis, especially osteoarthritis of the knee.16 17 In construction workers, the impact of overweight and obesity on osteoarthritis may possibly be further aggravated by work in kneeling positions.

The positive association between BMI and disability from cardiovascular disease is likewise consistent with expectations from epidemiological studies2 3 5 and similar to the relationship between BMI and cardiovascular mortality.36 However, in our cohort a major increase in disability risk was confined to levels of BMI above 35 kg/m², which were seen in a minority of 2% of men only. Therefore, and because disability due to cardiovascular diseases was found to be relatively less common than in other occupational settings,37 the positive association between BMI and disability due to cardiovascular diseases had only a small impact on the association between BMI and disability from all causes in our cohort.

In contrast to findings for other chronic diseases, we found a clear inverse association between BMI and work disability due to cancer, which was nearly identical to the association with cancer mortality.36 This association was especially strong for disability due to lung cancer, which could be partly explained by smoking. Smoking, the key risk factor for lung cancer, is known to be inversely related to BMI. Despite the high prevalence of smoking, cancer accounted for less than 10% of causes of work disability. Nevertheless, the decrease in risk of disability with BMI, which was seen over the all but the highest BMI levels, along with a similar pattern seen for disability due to mental and other diseases, was strong enough to make the overall BMI–disability association U-shaped.

An important limitation of our study, which is shared with many other pertinent studies, is that only BMI was available to define overweight and obesity. Although BMI shows a high correlation with body fat, it is uninformative with respect to the distribution of body fat and fat free mass. Furthermore, BMI is influenced by other factors, including muscle mass, which might be of particular relevance in this cohort of physically hard working men. Furthermore, only a single measurement of BMI at baseline was taken into account, thereby neglecting possible changes in weight over time, which may have led to some underestimation of BMI–disability associations.

Although we controlled for major potential confounders including smoking and alcohol consumption, there remains a potential for residual confounding. In particular, smoking and alcohol consumption were self-reported and referred to a single point of time only. Other important factors such as social and labour market related factors which are known to affect disability risk38 39 were not available. However, potential for confounding by such factors seems to be limited, given the relative homogeneity of the cohort with respect to such factors.

As in any occupational cohort, the potential impact of a healthy worker effect and a healthy worker survivor effect have to be considered. In this occupational group, which includes a large proportion of foreign workers, a potential healthy migrant effect deserves additional attention. While this phenomenon would be expected to have an impact on external validity and on absolute rates of disability, a major impact on observed associations between BMI and disability risk seems less obvious and less likely.

In this study, receipt of a work disability pension rather than the occurrence of disability was used as the endpoint. Although the occurrence of disability and receiving a work disability pension are likely to be closely correlated in the very elaborate German social security system, the correlation is certainly not perfect. However, from a public health point of view, leaving the workforce and receiving a disability pension is by itself a very important endpoint whose determinants deserve careful investigation.

Our study also has particular strengths, including the size and relative homogeneity of the study population and the length and completeness of follow-up. The large number of workers with work disabilities allowed us to assess the BMI–disability association in great detail, in particular with respect to dose–response relationships. Our analyses point to major peculiarities in those dose–response relationships in our large cohort of physically hard working men for whom results from epidemiological studies conducted in other settings may be of limited relevance.

Therefore, our results may be helpful for the proper interpretation of BMI levels and associated health risks in occupational groups characterised by heavy physical demands. In fact, our study suggests that the standards of “normal weight” or “optimal weight” might be quite different in these groups compared to people in mostly sedentary occupations, which has clear implications for health counselling in these groups.

In conclusion, our study indicates a U-shaped association between BMI and risk of work disability in this cohort of construction workers. BMI levels commonly considered to reflect moderate overweight are not associated with an increased risk of work disabilities in this cohort, but obesity increases the risk of work disability due to osteoarthritis and cardiovascular diseases.