Contribution of comorbid conditions to the association between diabetes and disability pensions: a population-based nationwide cohort study.

OBJECTIVES
Using Swedish population-based register data, we examined the extent to which comorbid conditions contribute to the risk of disability pension among people with diabetes.


METHODS
We carried out Cox proportional hazard analyses with comorbid conditions as time-dependent covariates among 14 198 people with newly diagnosed diabetes in 2006, and 39 204 people free from diabetes during the follow-up from 2007-2010. The average follow-up times were 46 and 48 months for those with and without diabetes, respectively.


RESULTS
For those with diabetes only, the incidence of all-cause disability pension was 9.5 per 1000 person-years. The highest incidence of disability pension were for those with: diabetes and depression (23.6); diabetes and musculoskeletal disorder (30.6), and those with diabetes and more than one comorbid condition (36.5). The incidence rate was 5.8 for those without diabetes. Diabetes was associated with a 2.30 times [95% confidence interval (95% CI) 2.09-2.54] higher risk of disability pension (adjusted for sociodemographic factors). This association attenuated by 41% after further adjustment for comorbid chronic conditions. While diabetes was a risk factor for disability pension due to musculoskeletal disorders and diseases of the circulatory system, even after accounting for the above-mentioned conditions, the association between disability pension due to mental disorders and diabetes was diluted after adjustment for mental disorders.


CONCLUSIONS
Although diabetes is an independent risk factor for disability pension, comorbid conditions contribute to this risk to a large degree.

Type 2 diabetes constitutes a considerable burden in Western countries, increasingly also in Latin America, the Caribbean, Southeast Asia, East Asia, and Oceania (1). Diabetes-related disability and the loss of healthy life-years also have a major impact from an economic point of view. While diabetes as the diagnosis on sick leave certificates is rare, individuals with diabetes have about 17 more days of absence from work due to mental disorders than individuals without diabetes. For musculoskeletal disorders and diseases of the circulatory system, the number of annual excess days of absence were 12 and 6, respectively (2). Thus, while preventing or delaying diabetes (ie, primary prevention) is the best way to avoid morbidity and premature mortality (3,4) -due to the high prevalence of comorbid conditions related to diabetessecondary prevention and prevention of permanent loss of working capacity are both of importance.
A systematic review concluded that diabetes was associated with a 30-300% increased risk of early retirement (5). However, most studies included in this review did not take into account comorbid conditions either at baseline or during follow-up. A few earlier studies on the Comorbidity in diabetes and disability pension effect of comorbidity on work disability among people with diabetes suggest that the risk seems to be shaped by comorbid conditions (6)(7)(8). However, no previous study has examined the contribution of time-dependent comorbid conditions, ie, those that occur during follow-up, to the association between diabetes and work disability pension. Because of these shortcomings, the effect of diabetes on disability pension remains unknown.
The purpose of this study was to explore the risk of all-cause and cause-specific disability pension among people with diabetes and compare it to that of individuals without diabetes, while taking into account the timing of comorbid conditions. The specific study questions were: (i) What is the incidence and relative risk of disability pension among people with diabetes compared to those without diabetes? (ii) What is the contribution of time-dependent comorbid conditions to the risk of allcause and diagnosis-specific disability pension among people with diabetes compared to those without diabetes having the corresponding other conditions?

Methods
We conducted a prospective study with a four-year follow-up. The data were from the nationwide population-based Insurance Medicine All-Sweden (IMAS) research database covering all people living in Sweden, from which we drew individuals aged 25- Comparisons to the general population were made on the basis of a 1% random sample of the population, totaling 39 204 individuals who were living in Sweden on 31 December 2006 and had no indication of diabetes during 2003-2010. The Regional Ethical Review Board, Stockholm, Sweden, approved the project.

Measures
Diabetes. Identification of diabetes was based on having at least one of the following: (i) insulin or other diabetes medication [Anatomical Therapeutic Chemical (ATC) code A10] (13); (ii) in-or out-patient record with ICD-10 code E10, E11, E12, E13, or E14 as described previously (2,8). Of those identified as having diabetes, 7% were treated with both A10A (insulin) and A10B (other diabetes medication), 11% with A10A only, 67% with A10B only, and 15% none of these (diabetes diagnosis was based on hospital records only). In the diabetes cohort, we excluded those with an indication of diabetes Disability pension. Data regarding the dates (first occurrence) of all-cause and diagnosis-specific (ICD-10) disability pensions were retrieved from the Social Insurance Agency, over the period of 1 January 2007 to 31 December 2010. With regard to diagnosis-specific analyses, we examined disability pensions due to mental disorders (F00-F99), musculoskeletal disorders (M00-M99), and diseases of the circulatory system (I00-I99).
Time-dependent covariates. Comorbid conditions examined were depression, other psychiatric disorders, cardiovascular disease, hypertension, and musculoskeletal disorders. These conditions are common causes of work disability in Sweden (14), and depression (15,16), cardiovascular diseases (17), and musculoskeletal disorders (18,19), have also been linked with diabetes. Indication of these conditions was based on having either a medication purchase record entry or in-or out-patient hospitalization record entry, or both. The identification of these disorders and data sources are further described in supplementary table A (www.sjweh.fi/data_repository. php). The presence or incidence of comorbid conditions among people with diabetes is likely to influence the Ervasti et al risk of disability pension. The presence or incidence of an illness influences the risk of disability pension also among the general population (20). Thus, information regarding the year of the first indication of an illness (pre-existing at the beginning of the follow-up or incident cases thereafter) was used as a time-dependent covariate. The timing of these conditions was calculated as either being present already at the beginning of the follow-up in 2007, incidence in 2008, incidence in 2009, or incidence in 2010. As we only had the year of the diseases and no exact date, the time of onset was defined as 1 January each year.

Statistical analysis
The absolute risk for disability pension was calculated for those with and without diabetes, presented as incidence rates per 1000 person-years. The relative risks were calculated for those with diabetes and compared to the risks of those without diabetes and presented as hazard ratios (HR) with 95% confidence intervals (95% CI), using competing risks models. A competing risks model is a variant of the Cox proportional hazard model. It is based on the fact that -when at risk of disability pension due to a specific diagnosis -individuals are simultaneously at risk of disability pension due to other causes and death, and these risks are mutually exclusive. We included comorbid conditions as time-dependent covariates in the models, which is considered a more robust method of adjusting for confounding, because it utilizes all available data (21).
The individuals were followed from 1 January 2007 through 31 December 2010 or the year they emigrated, until date of death, or date of disability pension, whichever came first. Disability pensioning (=1) and death (=2) were viewed as strongly dependent on health status   (ie, competing events). Emigration was viewed as independent of disability pension or death. Thus, those who emigrated were censored (=0). Regarding diagnosisspecific disability pension, death and receiving disability pension based on diagnosis other than that used as the outcome were defined as competing events (=2).
With regard to Kaplan-Meier curve presented in figure 1, we tested the proportional hazards assumption by including an interaction of each predictor with the log of follow up time. These interactions with time were all nonsignificant, confirming that the proportional hazards assumption was justified (P>0.09). SAS statistical software, version 9.4, was used for all analyses (SAS Institute, Cary, NC, USA).
The contribution of sociodemographic factors and other chronic conditions was determined by the percentage reduction in the coefficient for diabetes status after the inclusion of the covariate in question, using the formula: 100 × (B Model 1 -B Model 2 ) / (B Model 1 ), where Model 1 is the crude and Model 2 the adjusted model.

Results
Those with diabetes were more often men, had a lower level of education, did not have children <18 years of age living in the household, were born outside Sweden, and were somewhat older than those without diabetes. Differences in type of living area were small. Depression, other mental disorders, cardiovascular disease, hypertension, and musculoskeletal disorders were more prevalent among those with diabetes than those without at the beginning of the follow-up (table 1). Table 2 shows that the four-year onset and incidence estimates of disability pension were higher among those with diabetes than those without. The incidence of allcause disability pension was 9.5 per 1000 person-years for those with diabetes only and 5.8 for those without diabetes. The incidence rates varied with comorbid conditions ranging from 15.0 among those with diabetes and cardiovascular disease to 30.6 among those with diabetes and musculoskeletal disorder. Those with diabetes and more than one comorbid condition had the highest incidence of all-cause disability pension, 36.5 per 1000 person-years. Figure 1 shows the hazard curves of time from onset of diabetes until disability pension per comorbid musculoskeletal disorder, depression, other psychiatric condition, cardiovascular disease, and hypertension with diabetes, and also for those without diabetes. Because the event was defined as disability pension, longer survival time indicated a more favor- Ervasti et al able outcome. At four-year follow-up, 4% of those with diabetes but without any of the other conditions had ended up on disability pension, whereas for those with comorbid conditions, the percentage ending up on disability pension varied between 6% and 14%, depending on comorbid condition (figure 1). As also shown in figure 1, only 2% of those without diabetes ended up on disability pension.

Disability pension with and without diabetes
The mean follow-up time was 46 months (SD 9.5) for those with diabetes and 48 months (SD 5.9) for those without. During the follow-up, 100 (0.7%) of those with diabetes and 433 (1.1%) of those without diabetes emigrated from Sweden. A total of 482 (3.4%) of those with and 307 (0.8%) of those without diabetes died during the follow-up. The crude relative risk for all-cause disability pension was 3.11-fold (95% CI 2.84-3.40, Model 1) for people with diabetes compared to those without. Adjustment for sociodemographic factors attenuated this risk by 26% (HR 2.30, 95% CI 2.09-2.54, Model 2) (table 3).

Contribution of comorbid conditions
Other conditions also contributed to the risk of disability pension: further adjustment for other conditions attenuated the risk of all-cause disability pension by 41% when compared to the sociodemographics-adjusted model (HR  1.63, 95% 1.47, 1.80, Model 3). Specifically, depression (HR 2.93, 95% CI 2.63-3.27) and other mental disorders (HR 2.04, 95% CI 1.83-2.28) constituted a greater risk factor for all-cause disability pension than diabetes.
The results regarding disability pension based on mental disorders, musculoskeletal disorders, and diseases of the circulatory system are shown in supplementary tables B-E (www.sjweh.fi/data_repository. php). The incidence of disability pension due to mental disorders, musculoskeletal disorders and diseases of the circulatory system was at least twice as common among those with diabetes as among those without (4.7 versus 2.1 for mental, 3.6 versus 1.5 for musculoskeletal, and 2.8 versus 0.5 for circulatory diseases) (supplementary table B).
Mental disorders were the strongest predictor of disability pension due to mental disorders (HR 6.81; 95% CI 5.52-8.40 for depression; HR 4.44; 95% CI 3.58-5.49 for other psychiatric disorders); musculoskeletal disorders were the strongest predictor of disability pension due to musculoskeletal disorders (HR 5.34; 95% CI 4.37-6.53); and cardiovascular disease was the strongest predictor of disability pension due to circulatory system disease (HR 7.36, 95% CI 5.32-10.19). Having diabetes was associated with an increased risk of disability pension due to musculoskeletal disorders and circulatory system disease. However, adjustment for mental disorders diluted the effect of diabetes as a risk factor for disability pension due to mental disorders. In the sociodemographics-adjusted model, diabetes was associated with a 2.02-fold (95% CI 1.70-2.41) risk of disability pension due to mental disorders, a 1.59-fold (95% CI 1.32-1.94) risk of disability pension due to musculoskeletal disorders, and a 5.99-fold (95% CI 4.25-8.44) risk of disability pension due to cardiovascular diseases. Adjustment for other conditions attenuated the risk of disability pension due to musculoskeletal disorders by 65% when compared to the sociodemographics-adjusted model. The corresponding decrease in disability pensions due to circulatory system disease was 48%. In the fully-adjusted models, diabetes was associated with a 1.17-fold (95% CI 0.98-1.40) risk of disability pension due to mental disorders, a 1.37fold (95% CI 1.12-1.67) risk of disability pension due to musculoskeletal disorders, and a 3.30-fold (95% CI 2.30-4.72) risk of disability pension due to diseases of the circulatory system.

Discussion
This nation-wide prospective study among the population of Sweden showed that people with diabetes were, in both relative and absolute terms, at a higher risk of all-cause disability pension than those without diabetes during a four-year follow-up. The established socio-demographic risk factors for disability pension (20,22), such as older age, female sex, and lower educational level, explained 26% of this excess risk. Moreover, adjusting for comorbid conditions further attenuated the risk by 41%.
In relative terms, depression and other mental disorders constituted a greater risk factor for all-cause disability pension than diabetes. In diagnosis-specific analyses, we found that diabetes was a risk factor for disability pension due to musculoskeletal disorders and diseases of the circulatory system, even after accounting for the above-mentioned conditions. However, diabetes was not a risk factor for disability pension due to mental disorders after accounting for comorbid other conditions. This indicates that diabetes has an independent effect on the risk of disability pension, and especially on the risk disability pension due to musculoskeletal disorder or circulatory system disease. However, comorbid mental disorder contribute to that risk to a large degree. Earlier studies have found that diabetes was associated with a 37% lower probability of return to work after depression-related absence (23) and a 43% higher probability of recurrent depression-related absence (24). Disability pension due to diseases of the circulatory system among people with diabetes is possibly directly associated with diabetes-related complications, such as coronary heart disease or stroke (17,25). Diabetes-related complications also include disorders of the musculoskeletal system, such as Dupuytren's contracture, carpal tunnel syndrome, neuropathic osteoartropathy, and osteoarthritis (18,26), which may even lead to permanent work disability, ie, disability pension. Finally, pre-existing conditions including diabetes explained why older age was associated with longer sickness absence following musculoskeletal injury (27).
The results corroborate earlier evidence regarding the effect of diabetes and comorbid conditions on the excess risk of work disability (6)(7)(8). The study contributes to existing evidence by using comorbid conditions as time-dependent covariates and treating other diagnoses of disability pension as competing events (in diagnosis-specific analyses). In doing this, we were better able to demonstrate and isolate the direct effect of diabetes on all-cause and diagnosis-specific disability pension.
In addition to sophisticated methods of analysis, this study benefits from the large population-based prospective cohort study design and reliable registerbased measures with high coverage and specificity (28).
Though the validity of disability pension diagnoses was not studied, sick leave diagnosis showed a high validity compared to diagnoses from medical records (29). We also had a longer follow-up time than that reported in previous studies on the role of diabetes and comorbid conditions in work disability (6,7).

Limitations
The limitations include the fact that we did not reach cases of undiagnosed diabetes or those treated by lifestyle intervention only. We may also have missed some individuals with diabetes in 2003-2005 since data on medication purchases was from July 2005 only. However, in Sweden, you can only buy a maximum of three months' supply of medication at a time. Thus, as most people diagnosed with diabetes should have bought medication at least once during July-December, 2005, this purchase would have coded them as having diabetes in 2003-2005 and resulted in their exclusion from the data. The in-and out-patient records for 2003-2005 were from specialized healthcare, and part of diabetes care is carried out in primary healthcare. However, most of the cases managed in primary healthcare should have been identified through medication purchases. We were unable to identify some common complications of diabetes, such as neuropathy and retinopathy. Finally, people with chronic conditions such as diabetes, cardiovascular disease, or hypertension use more healthcare services and are thus more likely to be diagnosed with other conditions as well. This may lead to the overrepresentation of comorbid conditions among people with diabetes and the subsequent overestimation of the role of comorbidity in the association between diabetes and disability pension.
When interpreting these findings, we must acknowledge that, in some cases, diabetes may have caused the comorbid condition. In some cases, both may be a consequence of a common underlying factor. And in other cases, diabetes and the comorbid conditions may be causally unrelated. In all-cause disability pensions, it is possible that disability pension was granted due to diseases other than those we studied, causing unmeasured residual confounding. Thus, we are not able to make causal conclusions on the basis of the current study. Finally, our data lacked information on diabetes severity or diabetes control. However, the fact that the onset of diabetes occurred during the same year for all participants with diabetes somewhat dilutes the effect of the lack of this information.

Concluding remarks
In conclusion, although having diabetes is a risk factor for disability pension, comorbid conditions greatly contribute to this excess risk. Moreover, diabetes as a comorbid condition to mental and musculoskeletal disorders or diseases of the circulatory system constituted a risk factor for disability pension. Thus, comprehensive health examinations and especially the treatment of the mental health of people with diabetes may prevent permanent loss of working capacity.

Conflict of interest and sources of funding
The Academy of Finland (258598, 265174) supports J Ervasti and M Virtanen. J Ervasti is also supported by the Finnish Work Environment Fund (114260). T Lallukka was supported by the Academy of Finland (287488) and the Association for Promotion of Occupational Health. K Alexanderson, P Tinghög and L Kjeldgård received funding from the Swedish Research Council for Health, Working Life and Welfare. E Mittendorfer-Rutz was supported by the Swedish Research Council.