Predictors of working days lost due to sickness absence and disability pension

The population of this study was derived from the prospective Finnish Helsinki Health Study. The study is described in more detail elsewhere (Lahelma et al. 2013). The study recruited the employees aged 40–60 years of the City of Helsinki, the largest single employer in Finland. In the present analysis, we limited the study population to the participants aged 55 years at the baseline survey conducted during the years of 2000–2002 (response 73% for women and 61% for men, Laaksonen et al. 2008) or those aged 55 years at the first follow-up survey conducted in 2007 (Fig. 1). We included participants who provided informed consent for register linkages and remained employed by the same employer. The age of 55 years was selected to follow-up people for their working years lost before their statutory retirement age. Additionally, a large number of early exits take place after the age of 55. The final study population consisted of 1630 participants, of whom 81% were women, which reflects the female-dominated role of public-sector work in Finland. The ethics committees of the health authorities of the City of Helsinki, and the Department of Public Health, University of Helsinki approved the study.

Information on all sickness absence days was obtained from the personnel register of the City of Helsinki. The register comprises all self-certified short and medically certified longer SA days. The number of calendar days lost due to SA was calculated by the cumulative number of SA days between ages 55 and 65. Information on disability pension was obtained from the Finnish Centre for Pensions’ register. The number of calendar days lost due to DP was calculated from the start and end date of each DP award. Temporary DP has a fixed end date, while in continuing DP, the end date usually refers to the date on which DP is transferred to legal old age retirement. In Finland, disability pension can also be awarded as partial retirement, which requires part-time work. For the partial DP, we used half weights (i.e., a year on a partial retirement equals to 0.5 years lost).

Baseline characteristics included gender, education (1. elementary school, 2. middle school, 3. vocational school or equivalent, 4. college degree, and 5. university degree), occupational class (1. managers and professionals [e.g., managers, administrators, teachers, and doctors], 2. semi-professionals [e.g., nurses, technicians, and foremen], 3. routine non-manual employees [e.g., clerical workers and healthcare assistants], and 4. manual workers [e.g., transport workers, cleaning workers, and canteen workers]) (Aittomäki 2008; Leinonen et al. 2011), smoking status (never, past, and current), binge drinking (6 units or more at the time in the past 30 days), body weight and height, leisure-time physical activity, work-related physical and psychosocial factors, self-reported long-standing illness, number of pain sites, and common mental disorders.

Body mass index (BMI) was calculated using self-reported weight and height and grouped into three levels: normal (BMI < 24.9 kg/m²), overweight (BMI 25.0–29.9 kg/m²), and obesity (BMI ≥ 30 kg/m²). Only seven participants had BMI < 18.5 kg/m² and we included them in normal weight category. Excluding these seven participants from the analyses did not change the results. Information on the participants’ average weekly hours of leisure-time physical activity was gathered using four questions on: (1) walking, (2) brisk walking, (3) jogging, and (4) running. The number of hours per week for each activity ranged from 0 to 4 h. A metabolic equivalent (MET) index was calculated for each participant by multiplying the MET values of each activity intensity by the time spent on them, and summing them up (Ainsworth et al. 2011; Lallukka et al. 2016). The intensity of leisure-time physical activity was classified into three groups: low (< 14 MET-hours per week), moderate (≥ 14 MET-hours of moderate intensity activities with no vigorous activities), and vigorous activity (≥ 14 MET-hours, including some vigorous activity such as jogging or running). For the assessment of physical workload factors (8 items), working with a computer and mouse (2 items), and environmental factors (8 items), the 18-item inventory developed at the Finnish Institute of Occupational Health (Piirainen et al. 2003) was used. The workload factors included (1) heavy lifting, or pulling or pushing heavy loads; (2) back rotations; (3) awkward working positions; (4) repetitive movements; (5) vibration; (6) standing; (7); walking; and (8) sitting. Five items of Karasek’s Job Content Questionnaire (Karasek et al. 1998) were used to assess job demands and seven items to assess job control. Demand and control scales were dichotomized at the median and job strain was defined as reporting high job demands coupled with low job control.

Common mental disorder was assessed using the General Health Questionnaire (GHQ-12). We used a cut-point 3 or higher scores to define presence of a mental health problem (Goldberg et al. 1997). Data on the presence of acute/subacute (less than 3 months) or chronic (longer than 3 months) pain in the head/face, neck/shoulders, lower back, upper limbs, lower limbs, abdominal area, and any other location were collected by the self-administered questionnaires. We classified the number of pain sites into three groups (none, one, and two or more), and defined multisite pain as pain in two or more sites. Long-standing illness was grouped into illness limiting and not limiting work or daily activities.

Missing data ranged from missing information on occupation for six participants to missing one or more items of leisure-time physical activity for 92 participants (5.6%). Fourteen percent of the participants had missing data on one or more independent variables. We imputed missing data using the method of multiple imputation by chained equations (van Buuren et al. 1999) and created 10 datasets. As a sensitivity analysis, we also ran complete case analysis for both genders combined. Spearman's rank correlation coefficient was used to estimate the correlation between two categorical variables and the variance inflation factor (VIF) was used to assess multicollinearity between independent variables. We first reported the observed average number of days lost by the baseline characteristics and then identified the predictors of days lost using negative binomial regression model. The model estimates rate as the number of days lost (a count) per unit time. Observation time for total number of days lost was calculated by the time between the age of 55 and statutory pension, death, or the age of 65. After accounting for the effects of the predictors, the variance of the number of days lost was larger than the mean, and the dispersion statistic was above one. Therefore, for this overdispersed count data, negative binomial regression was a better model than Poisson regression.

First, gender-specific or gender-adjusted incidence rate ratios (IRR) were estimated (base model) and then adjusted IRRs (full model) were estimated for both genders combined and for women due to a high number of women in the sample. The full model adjusted for gender, education, occupational class, smoking, body mass index, leisure-time physical activity, binge drinking, heavy lifting, back rotations, awkward working positions, long-standing illness, common mental disorder, and number of pain sites. For men, because of small sample size, the non-significant characteristics were removed from the full model one at a time until all characteristics were associated with the number of days lost with P value ≤ 0.2. The predicted number of days lost were calculated by user-written mimrgns command in Stata (StataCorp, College Station, TX, USA). The predicted number of days lost was calculated from the fitted full model, while keeping all other covariates at their means (using “atmeans” option) (Williams 2012).

The number of days lost did not differ between men and women (Table 1). The number of days lost in the participants with college or university degree was half of that of those with elementary or middle school education. Compared with managers or professionals, the number of days lost was higher in other occupational groups, particularly manual workers. Past and current smoking, overweight and obesity, binge drinking, heavy lifting, back rotations, awkward working positions, job strain, long-standing illness limiting work or daily activities, common mental disorder, and number of pain sites increased the number of days lost (Table 1). Leisure-time physical activity, particularly moderate level, reduced the number of days lost. Job control was associated with lower rate of days lost (IRR = 0.63, CI 0.56–0.71 for 1-unit increase), whereas job demands were not associated with the number of days lost ((IRR = 1.03, CI 0.93–1.14). In gender-specific analyses, the findings in women were similar to those of both gender-combined. In men, smoking, obesity, leisure-time physical activity, binge drinking, and awkward working positions were not statistically significantly associated with the number of days lost.

Figure 2 presents the predicted number of days lost due to both SA and DP as well as the number of days lost due to only DP by background characteristics. Gender, occupational class, overweight and obesity, vigorous leisure-time physical activity, back rotations, and long-standing illness not limiting work or daily activities were not associated with the number of days lost in the model adjusted for all covariates (Table 2 and Fig. 1). Past and current smoking, binge drinking, heavy lifting, awkward working positions, long-standing illness limiting work or daily activities, common mental disorder, and number of pain sites increased the number of days lost, while high level of education and moderate level of leisure-time physical activity reduced the number of days lost. The number of days lost was also slightly lower in participants with job strain. Job demands and job control were not associated with the number of days lost. In subgroup analysis, job strain was associated with lower number of days lost among managers, professionals or semi-professionals, but not among routine non-manual workers or manual workers. In women, the associations were similar to those of both gender-combined, except that manual workers had significantly higher number of days lost compared with managers or professionals (Table 2). In men, current smoking, overweight, back rotations, long-standing illness limiting work or daily activities, and multisite pain increased the number of days lost (Table S2).

Table 3 shows the predicted number of days lost based on the full model according to the participants’ baseline characteristics. Long-standing illness limiting work or daily activities increased the number of days lost by 261 days (95% CI 191–332), common mental disorder by 118 days (95% CI 68–169), and multisite pain by 101 days (95% CI 47–155). The increase in the number of days lost for smoking, binge drinking, heavy lifting, and awkward working positions ranged between 46 and 75 days. On the other hand, moderate level of leisure-time physical activity reduced the number of days lost by 59 days (95% CI 15–102).

This study utilized data from an occupational cohort with good response rates, highly reliable and complete register-based work disability measures (Laaksonen et al. 2008; Roos et al. 2013), and a long follow-up time to estimate working days lost among midlife employees until the desired retirement age. Moreover, we were able to distinguish between SA and DP days, and combine all SA spells of different lengths for a more comprehensive picture of days lost as compared to previous studies that have mostly missed particularly a large number of short SA spells. We also had a large set of different potentially modifiable work and health-related predictors of working days lost, to identify potential groups for intervention and prevention. Based on earlier studies, we included the determinants of DP and SA in the full models; however, it cannot be ruled out that we have not missed some potential predictors of working days lost. Finally, including a homogenous group of people of the same age working for the same employer helps produce more reliable results. Despite these strengths, the study had some limitations. First, the assessments of weight and height, leisure-time physical activity, smoking, drinking, occupational physical workload factors, common mental disorder, and long-standing illness were self-reported; however, differences in their predictive power for SA have been shown to be small in this cohort (Korpela et al. 2013). Second, we did not have information on sickness absence for the employees who changed their employer during the follow-up period. However, only a small number of the participants aged 55 or older changed their employer. Third, the number of men was small, and thus, the results for men should be interpreted with caution.