Methods

Data

Data on health-related behaviours and working conditions were collected by independent cross-sectional baseline surveys in 2000, 2001 and 2002 among middle-aged employees of the City of Helsinki.20 The City of Helsinki is the largest employer in Finland, with nearly 40 000 employees. The personnel register was used to identify employees who in these consecutive years reached the age of 40, 45, 50, 55 or 60, and a self-administered questionnaire was mailed to them (n = 13 346). The overall response rate was 67%.21 22

The questionnaire data were prospectively combined with sickness absence records derived from the City of Helsinki registers using the unique personal identification number assigned to each Finnish citizen. However, the linkage was only possible for those respondents who had given a written consent for the linkage when returning the questionnaire (78% of the respondents). Sickness absence data could not be sought for 16 respondents because they had removed the code from the questionnaire. Thus, this study includes 5470 women and 1464 men, reflecting the gender distribution among the employees of the City of Helsinki.

Our non-response analysis has shown that the baseline data satisfactorily represent the target population. Men, younger people, manual workers, and those with low income are slightly under-represented among the respondents. Associations of these background characteristics with consenting to register linkages are weaker but generally to the same direction than those for responding to the survey. However, the associations of the background characteristics with sickness absence are similar among the respondents and non-respondents, suggesting that associations between the study variables are unlikely to be affected by drop-out.21 22

Measurement of sickness absence

We separately examined self-certified sickness absence spells of 1–3 days and spells of ⩾4 days for which the employer requires a medical certification. The follow-up time was started from the day the respondent returned the questionnaire and continued until the end of 2005 or until the work contract had terminated. All interruptions in working time due to other reasons than own illness were subtracted from the follow-up time. The overall number of person-years in the analyses was 26 904 with a mean follow-up time of 3.9 years.

Health-related behaviours

Distributions of the health-related behaviours included in this study are shown in table 1. Smoking was classified into four categories. Current non-smokers were divided into never smokers and former smokers on the basis of previous regular smoking. Among current smokers those smoking >20 cigarettes per day were classified as heavy smokers. Alcohol use was calculated as the sum of reported weekly drinks of beer, cider, wine and spirits. Abstainers were considered as a separate group. The rest were classified as moderate, high (15–24 weekly drinks in men, 8–16 weekly drinks in women) and very high drinkers (>24 weekly drinks in men, >16 weekly drinks in women) using cut-off points taken from Finnish recommendations.23 24 Physical activity was assessed by asking the average time spent weekly in physical activities of intensity corresponding to walking, vigorous walking to jogging, jogging, and running.25 Each of the four questions had five response alternatives that were weighted by intensity of the activity to calculate metabolic equivalent tasks (METs)26 and the total activity score was divided into quartiles ranging from <11.5 to >37.5 MET hours per week. Dietary habits were measured by compliance with the Finnish national dietary guidelines.23 A summary index of recommended food habits consisting of six items was constructed: eating fresh fruits or berries daily, eating fresh vegetables daily, eating dark bread such as rye bread daily, eating fish at least twice a week, using margarine on bread, and using vegetable oil in cooking and baking. The summary index was classified to closest possible quartiles. Relative weight was measured by body mass index (BMI) and classified as underweight (BMI <20 kg/m²), normal weight (20–24.9 kg/m²), overweight (25–29.9 kg/m²), and obesity (>30 kg/m²).

Working conditions and occupational social class

Psychosocial working conditions were assessed using four questions. Mental demands of work were asked with a single-item question on a four-point scale. Job strain was measured by the Framingham version of the Karasek questionnaire.27 The summary scores for job demands and job control were dichotomised from the median and cross-classified into high strain, active, passive and low strain categories. Job satisfaction was measured using an item from an inventory asking about satisfaction with nine different areas of life, one of which was work. There were seven response alternatives ranging from “very satisfied” to “very dissatisfied”. Problems with workplace climate were asked using a question asking whether bullying existed in the workplace.

Measures of physical working conditions included a question asking physical demands of work on a four-point scale. Furthermore, an 18-item inventory of potentially harmful characteristics of work and working environment developed at the Finnish Institute of Occupational Health was used.28 The items had four response alternatives ranging from “does not exist” to “exists and bothers a lot”. Factor analysis of these items provided three factors which indicate physical work load (six items, Cronbach’s alpha 0.82), hazardous exposures (nine items, Cronbach’s alpha 0.79), and computer work (three items, Cronbach’s alpha 0.80). The items that were included in the physical work load factor were uncomfortable postures, repetitive trunk rotation, repetitive movements, standing, walking and heavy physical work (lifting and carrying). Hazardous exposures factor included items such as exposure to dirt and dust, damp and wetness, noise, solvents and other irritating substances, and problems with temperature or lightning. Computer work, using the computer mouse, and sedentary work were included in the computer work factor.

Occupational social class was divided into four hierarchical groups: managers and professionals, semi-professionals, routine non-manuals, and manual workers. Manual workers and non-manual employees were distinguished using the socioeconomic classification of Statistics Finland,29 and non-manual employees were further divided into three groups according to the occupational classification of the City of Helsinki.

Statistical methods

Sickness absence rates for self-certified and medically confirmed absence spells across the health-related behaviour categories were first calculated for descriptive purposes. The rates are reported per 100 person-years.

Associations between health-related behaviours and sickness absence were then examined using Poisson regression. The number of sickness absence spells during the follow-up period was used as the outcome variable. This outcome effectively uses the information when one individual has several sickness absence spells and it is not dominated by only a few prolonged absence spells. Age-adjusted rate ratios with 95% confidence intervals were calculated comparing other health-related behaviour categories to the category assumed most advantageous for health. We then examined whether these associations can be explained by physical and psychosocial working conditions and occupational social class. First, psychosocial working conditions were adjusted for by including all four variables simultaneously in the models. Similar procedure was then followed for the four measures of physical working conditions and finally for social class. Changes in the associations due to the adjustments were assessed from the point estimates. A central property of a Poisson model is the assumption that the mean of the outcome variable and its variance are equal. We found larger variance than the mean, thus implying extra-Poisson variation. This extra-Poisson variation was taken into account by adjusting the confidence intervals with a scale parameter obtained by dividing the residual deviance by the degrees of freedom. The adjustment does not affect the point estimates but increases standard errors and thus widens the confidence intervals.30

Furthermore, to quantify the sickness absence burden related to each of the health-related behaviours (and their subcategories) we calculated population attributable fractions (PAFs) using the formula presented by Hanley.31 PAF can be defined as the proportion of sickness absence cases avoided if all employees would change their behaviours to conform with those with the lowest risk, all other things remaining equal. This literal definition of PAF involves strong and direct assumption of causality that is unlikely to hold true. Since complex mechanisms are involved health-related behaviour changes may not automatically lead to changes in sickness absence. However, PAF is a useful measure for comparing the relative importance of health-related behaviours as predictors of sickness absence as it takes into account both the risk of sickness absence in each health-related behaviour category as well as the size of these categories.

All analyses were conducted separately for women and men. SAS V.8.02 for Windows (SAS Institute Inc, Cary, North Carolina) was used for the analyses.

Results

Table 1 presents the distribution of the study population by health-related behaviours in person-years and the number of sickness absence spells per 100 person-years at work. The number of sickness absence spells was higher in women than in men, and self-certified sickness absence spells were more common than medically confirmed spells. All those with less healthy behaviours had more sickness absence than those with more healthy behaviours. For example, heavy-smoking men had 151 self-certified sickness absence spells per 100 person-years, while among never smokers the corresponding figure was 76. For the medically confirmed spells there was a graded decrease from 104 spells among heavy smokers to 47 spells among never smokers.

Among men, all health-related behaviours apart from dietary habits were associated with self-certified sickness absence after adjusting for age (table 2). Both heavy and moderate smokers had increased sickness absence rates but for relative weight only obesity (BMI >30) was notably associated with sickness absence. Adjusting for psychosocial working conditions slightly weakened the association of each health-related behaviour with self-certified sickness absence in men. Adjusting for physical working conditions slightly weakened the associations of smoking and obesity with self-certified sickness absence, while adjusting for social class weakened the same associations somewhat more. Furthermore, we calculated PAFs to quantify the sickness absence burden of each behaviour. The PAF tells, in per cent, how much of the sickness absence cases could be avoided if other health-related behaviour categories had the sickness absence risk of the reference category, assuming causality. Judging from the PAFs smoking contributed most to the burden of self-certified sickness absence spells. Moderate smoking, which was more common than heavy smoking, caused the largest burden. Overweight caused more burden than obesity. Those with healthy dietary habits had highest risk for sickness absence after adjustments, resulting in negative PAF.

Among women, smoking, relative weight and alcohol use had the strongest associations with self-certified sickness absence spells, while the other associations were quite weak (table 2). The adjustments had no effects on the found associations. Smoking and relative weight had the largest effects on the sickness absence burden.

Among men, heavy smoking and obesity more than doubled the risk for medically confirmed sickness absence, but the risk was also increased for both very high alcohol use and abstaining, low physical activity, and overweight (table 3). Adjusting for psychosocial working conditions somewhat attenuated the associations of smoking, alcohol use, physical activity and relative weight with medically confirmed sickness absence. Adjustment for physical working conditions in turn attenuated the associations of smoking and relative weight with medically confirmed sickness absence. Adjusting for social class attenuated the associations of smoking and relative weight with medically confirmed sickness absence. Judging from the PAFs high relative weight contributed most to the sickness absence burden among men, followed by smoking and physical activity.

Also among women smoking and relative weight were most strongly associated with medically confirmed sickness absence while the associations of other health-related behaviours with medically confirmed sickness absence were clearly weaker (table 3). Both very high drinkers and abstainers had increased sickness absence rates but the associations were weaker than in men. Heavy smoking and obesity nearly doubled the risk for medically confirmed sickness absence. Adjusting for psychosocial working conditions had no effect on the associations. Adjusting for physical working conditions slightly attenuated the associations of smoking and relative weight with medically confirmed sickness absence, while adjusting for social class had similar but stronger effects. Judging from the PAFs high relative weight contributed most to the sickness absence burden, followed by smoking and physical activity.

Discussion

Health-related behaviours are important targets for workplace interventions and could provide potential means for reducing sickness absence. However, it has been unclear on which health-related behaviours preventive measures should be focused. This study examined the associations of four health-related behaviours and relative weight measured by BMI with subsequent self-certified and medically certified sickness absence. Smoking and relative weight showed the strongest associations, while those of physical activity and alcohol use were weaker. Dietary habits were only modestly associated with sickness absence in women. The found associations were generally stronger in men, and stronger in medically confirmed than in self-certified absence spells.

Few previous studies have examined the associations of several health-related behaviours and sickness absence. A Danish study8 found that smoking and relative weight were more important predictors of receiving sickness absence compensation for eight consecutive weeks than alcohol consumption and physical activity. A US study9 examining a large number of health-related behaviours found that all behaviours included in our study except heavy alcohol use were associated with having ⩾2 absence days per year. Relative weight and smoking had the strongest associations while those of eating habits and physical activity were weaker. The results of these previous studies thus agree with ours in showing the strongest associations for smoking and relative weight.

Of the health-related behaviours studied, smoking contributed most to the burden of self-certified sickness absence. Smoking was associated with self-certified sickness absence almost as strongly as with medically confirmed sickness absence. Although the association with medically confirmed sickness absence may relate to smoking-induced chronic diseases, the association with self-certified sickness absence is more likely to reflect short-term respiratory infections and other minor transient morbidities. In addition to heavy and moderate current smoking, former smoking slightly increased sickness absence although this association reached statistical significance only in women. The finding of increased sickness absence levels among former smokers is consistent with previous studies.4 5 8 Smoking has long-term health-damaging effects that may increase sickness absence even after quitting. Furthermore, those who were former smokers at the time of responding to the questionnaire may have relapsed during the follow-up, ensuing short-term health effects of smoking. Also, health-related selection is possible, as smoking cessation may be more common among those who already have health problems.

High relative weight was the strongest contributor to medically confirmed sickness absence. Obesity largely increased the risk of sickness absence while the effect of overweight was much weaker. This is consistent with a previous more detailed report from these data showing marked increase in sickness absence levels only among those with a BMI >27 kg/m².32 However, the present study showed that because overweight is much more prevalent than obesity, among men, overweight contributed to the sickness absence burden even more than obesity and among women as well it had a notable additional effect.

In previous sickness absence studies the volume of alcohol use has been examined more often than other health-related behaviours.1 2 3 These studies have reported a curvilinear association between alcohol use and sickness absence similar to what is often found in other health outcomes such as coronary heart disease.33 Also in our study, sickness absence levels were lowest among moderate drinkers. While self-certified sickness absence was most common among very high drinkers, medically confirmed sickness absence showed increased rates among very high drinkers and abstainers. In sickness absence research the volume of alcohol use has been the most commonly used measure. It is possible that other measures such as binge drinking and drinking problems show stronger associations with sickness absence than the average number of drinks.3 34

We also examined whether the associations between health-related behaviours and sickness absence could be explained by psychosocial and physical working conditions and social class. Adjusting for psychosocial working conditions had no effects on the associations among women and only very weak effects among men. Physical working conditions explained a part of the association of smoking and high relative weight with sickness absence. Social class had similar but slightly stronger effects, especially for smoking.

Psychosocial14 15 and physical16 working conditions have shown associations with sickness absence but the mechanisms behind these associations are not wholly understood. Health-related behaviours may be one mechanism through which stressful working conditions affect health as well as sickness absence.10 11 In our study, adjusting for working conditions did not markedly weaken the associations between health-related behaviours and sickness absence, which suggests only a weak link between health-related behaviours and working conditions as determinants of sickness absence. Social class had slightly stronger effects than working conditions. Other aspects of social class than work-related exposures may thus be important for the association between health-related behaviours and sickness absence.

Methodological considerations

The strengths of this study include a relatively large sample, reliable register-based data on sickness absence, and a prospective study design. However, information on health-related behaviours was based on self-reports which may have attenuated the findings. For example, alcohol use is typically under-reported and the found associations may be too weak. Similarly, inadequate reporting of working conditions is likely to reduce their explanatory power.

Another limitation of the study is that health-related behaviours and working conditions were assessed only at baseline. It is therefore possible that health-related behaviours and working conditions change during the follow-up time. The average follow-up time was 3.9 years providing sufficient statistical power. However, optimal follow-up time is difficult to determine. It is possible that short-term associations between health-related behaviours and sickness absence would be stronger than long-term associations. Moreover, since only one measurement of health-related behaviours was available, conclusions on causal associations should be made with caution.

Medical certification is required among Finnish municipal employees for absence spells >3 days and this was used as the cut-off point for “short” and “long” absence spells. The cut-off point is relevant for health-related behaviours as, for example, alcohol consumption might be related to short-term sickness absence spells in particular. However, long sickness absence spells have been shown to be a better measure of illness than short absence spells.35 36 We performed additional analyses using 14 days as the cut-off point for long spells, and the results were similar to the lower cut-off point except for low physical activity for which the association with sickness absence was slightly stronger. The number of sickness absence spells has been a commonly used outcome measure in previous longitudinal studies examining determinants of sickness absence.15 18 Other outcome measures such as the total number of sickness absence days could also be used when considering the sickness absence burden related to health behaviours.

Conclusions

Health-related behaviours, most importantly smoking and high relative weight, were strongly associated with sickness absence. These associations were stronger for medically confirmed sickness absence spells, supporting the idea that health problems induced by health-related behaviours are likely to be primary mechanisms linking these behaviours with sickness absence. Psychosocial and physical working conditions as well as social class only slightly explained the associations of smoking and high relative weight with sickness absence. These results suggest that preventive measures targeted on health-related behaviours are important means to reduce sickness absence and prevent work disability among employees in general.