Emotional support predicts more sickness absence and poorer self assessed work ability: a two-year prospective cohort study

The aim of this study was to investigate prospectively, over a two-year follow-up in a general middle-aged working Swedish population, the effect of psychosocial work environment, psychosocial resources and previous back pain on self-reported sickness absence and self assessed prognosis of work ability.

The Life Conditions, Stress, and Health (LSH) study [25‐28] is a longitudinal study of socioeconomic differences in health in a population of men and women (non-patients) drawn from the county of Östergötland in the southeast of Sweden. It was based on a random sample of 1007 men and women between 45-69 years of age in 2003, as stratified by sex and age, who belonged to any of the catchment areas of ten primary health care centres in the region (response rate 62%). Data collection at baseline (2003 to 2004) and at follow-up (2005 to 2006) included data self-reported by postal questionnaire. Exclusion criteria were serious disease and difficulties in understanding the language. Follow-up data were collected from a total of 795 men and women (response rate 79%).

The eligible individuals for the present study were individuals who were employed both at baseline and follow-up. At baseline, 534 of the respondents in the survey were employed. 33 respondents were sick-listed at baseline and therefore excluded. Of the 501 baseline respondents, 409 returned the follow-up questionnaire (82% response rate). Another 68 individuals were no longer employed at the time of the follow-up, resulting in a final study sample of 341 individuals (prospective cohort). At follow-up the proportion of individuals with a manual occupation was higher in non respondents than in respondents (p = 0.002).

The reliability of the psychosocial scales was estimated by Cronbach's α internal consistency coefficient. The prospective impact of each psychosocial instrument and previous back pain at baseline was analysed with the number of sickness absence spells or days at follow-up using a Poisson regression model with a scale parameter to account for over-dispersion [40]. The analyses were adjusted successively for the following variables: 1) Model I for age and sex; 2) Model II adding lifestyle factors (BMI, smoking, and alcohol consumption), SES, and disease; 3) Model III all factors in the previous steps and adding past sickness absence. The prospective association between each psychosocial factor and previous back pain at baseline and a work ability rating of 'poor' at follow-up was analysed using a logistic regression in Models I to III described above.

A varimax-rotated principal components factor analysis [41] was performed among psychosocial factors (job strain, overcommitment, social support at work, emotional support, self-esteem, coping, sense of coherence, and perceived control) in order to derive composite dimensions of psychosocial factors, and thereby reduce the risk of reporting statistical significance due to multiple testing [42]. Finally the independent effect of three factors, derived from the factor analysis of all psychosocial instruments: work environment, emotional support and psychological resources as well as previous back pain, were analysed, simultaneously, as determinants of sickness absence (or, as the case may be poor work ability) with multivariate Poisson (or logistic) regression fully adjusted for covariates as above.

The correlation between the number of sickness absence spells and days, both at baseline and follow-up, was estimated by Spearman correlation coefficient. A significance level of 5% was considered to be statistically significant. SPSS 17 was used for factor analysis and SAS 9.1 for Poisson and logistic regressions.

The data had zero to 29 missing values for each variable, with the exception of self assessed work ability, which had 63 missing values (around 20% of participants). To retain all participants, imputed values were generated for the missing data from self assessed work ability. The 63 non responders for work ability had less sickness absence, for both spells and days, than the 278 respondents (all p < 0.01). Non response on self assessed work ability was coded as good work ability and statistical analyses were conducted using the imputed dataset [43]. A complete case statistical analysis of work ability was performed, and gave similar results to that using imputed values reported in the tables.

Descriptive statistics for the prospective cohort are provided in Table 1. Mean number of sickness absence spells and days at baseline and follow-up differed little over the two-year study (Table 1). The Spearman correlation coefficient between sickness absence at baseline and follow-up was 0.41 for number of spells, and 0.38 for number of days.

Most of the sick leave spells were short-term as 87/121 (72%) of the individuals sick-listed within the last 12 months before baseline, and 114/142 (80%) of the individuals sick-listed within the last 12 months before follow-up, had a total of 1-14 sick leave days.

The descriptive statistics for the different psychosocial factors are presented in Table 2.

Three factors of psychosocial scales were identified in the factor analysis: I. work environment (including job strain, overcommitment, social support at work), II. emotional support and III. psychological resources (including 1. sense of coherence, 2. mastery, 3. self-esteem, and 4. perceived control). Together, the three derived components captured 64.2% of the total variance.

In model I adjusted for age and sex, several factors emerged as predictors of more sickness absence spells (Table 3) and days (Table 4): occupation (non manual vs. self-employed), alcohol consumption (low vs. medium intake), disease, past sickness absence, high job strain, high emotional support, and low self-esteem. Previous back pain and low decision latitude predicted more sickness absence days but not spells.

In Model II, adjusted also for effects of demography, lifestyle factors and disease, high emotional support was the only significant predictor of more sickness absence spells and days (data not shown) and high perceived control predicted an increased number of sickness absence days. In Model III, adjusting also for previous sickness absence caused little change in most estimates.

The multivariate analysis of previous back pain and derived components of factor analysis confirmed that high levels of emotional support were related to a significant increased number of sick-leave spells (RR = 1.36; 95% CI = 1.11-1.67; p = 0.004) and days (RR = 1.68; 95% CI = 1.22-2.31; p = 0.002) at follow-up (Table 5).

In Model I adjusted for age and sex, several factors emerged as predictors of poor self assessed work ability (Table 6): occupation (manual vs. non manual), high BMI, past sickness absence, previous back pain, high job strain, low decision latitude, low self-esteem, low coping, low sense of coherence, and low perceived control.

In Model II, previous back pain, high job strain, low decision latitude, low self-esteem, and low coping were predictors of poor self assessed work ability (data not shown). In Model III, adjusting for past sickness absence caused little change in most estimates (see Table 6).

The multivariate logistic regression analysis of derived components of factor analysis (Table 5) showed that previous back pain (OR 2.56; 1.13-5.81; p = 0.02), high emotional support (OR 1.58; 1.02-2.46; p = 0.04), and low psychological resources 0.62 (0.44-0.89; p = 0.008) were related to poorer self assessed work ability at follow up, while the impact of work environment on work ability was no longer statistically significant (OR 1.39; 0.98-1.98; p = 0.07).

A high level of emotional support was associated with both a higher frequency and longer duration of sickness absence. Such an association could be surprising since social support is considered protective against the development of depression in those exposed to life events [44]. Prospective studies, which control for baseline health status, consistently show increased risk of death among persons having few social relationships [45]. The association between high emotional support and increased risk of sickness absence is not surprising if such absence is seen as the effect of an "illness behaviour" rather than illness itself. High level of confiding/emotional support may encourage empowerment, security, and perceptions of control, which legitimize taking leave from work when ill [23]. Our findings confirm and extend previous findings in sickness absence, where high levels of confiding/emotional support were associated with higher frequency of short-term and long-term sickness absence [22]. Notably the definition of long-term sickness absence is >7 days in the Whitehall II cohort and >2 weeks in our study. Both definitions are in correspondence with the social insurance system in the two countries.

Findings of an association between increased job strain and more periods of sickness absence and days when adjusted for age and sex is in line with earlier research. The component of the demand-control model associated with lower sickness absence was "high decision latitude". This confirms the result of previous research, in particular that decision latitude [14, 18] appeared to be a more important risk factors for sickness absence, than psychological demands and social support at work [19]. The association between job strain and sickness absence was reduced and was no longer statistically significant in the model adjusted for health behaviour and SES. The finding that previous back pain was related to the duration of sickness absence when adjusted for age and sex is consistent with previous research [46, 47]. Poor psychosocial work conditions and physical workload are important risk factors for musculoskeletal pain [48]. The association between previous back pain and duration of sickness absence was no longer statistically significant in the model adjusted for health behaviour and SES. In both cases loss of effects after adjustment for SES can be an effect of over adjustment because of the strong relation between occupation, back pain and psychosocial factors at workplace [10, 15].

Knowledge of predictors for work ability is important for disability prevention, since work ability is an important predictor of duration of sickness absence [49], and return to work [50, 51].

In this study work ability was related to a broader array of determinants than sickness absence. Work ability is the self-perceived relation between work demands and individual resources, defined as health and functional ability, education and competence, values and attitudes [52, 53]. The association of psychosocial factors at work (high job strain and low decision latitude) and poor self assessed prognosis of work ability in the model adjusted for age, sex, lifestyle factors, SES, disease and past sickness absence (Table 6) is consistent with previous research [54, 55]. However in the multivariate logistic regression adjusted additionally for previous back pain, emotional support and psychological resources (Table 5), the association between psychosocial work environment and poor work ability was no longer statistically significant.

Previous back pain was related to reduced self assessed prognosis of work ability in all models proposed (Tables 5 and 6). Back disorders constitute one of the most common causes behind long-term sickness absence and disability pension in Sweden [46, 47]. Persistent musculoskeletal pain has been shown to be a predictor of reduced work ability [56]. If activity aggravates the pain (such as with heavy physical work load), and the individual avoids or reduces his activities, then pain may lead to disability. Cognitive function, and overall health were related to work ability in patients with chronic musculoskeletal pain [57].

Low individual psychological resources (coping and self-esteem) were related with self assessed prognosis of poor work ability (Tables 5 and 6). Coping and self-esteem are closely related to self-efficacy [58]. Self-efficacy, which is defined as confidence in being able to carry out a set of defined activities [59], has been highlighted in the literature as playing an important role for work ability and in the process of returning to work [60, 61].

Just as could be seen for measures of sickness absence, high emotional support was related to poor prognosis of work ability in the multivariate logistic regression in Table 5. In this analysis, work ability is adjusted for age, sex, lifestyle factors, SES, disease, past sickness absence, previous back pain, work environment and psychological resources. In the analysis presented in Table 6, which was only adjusted for age, sex, lifestyle factors, SES, disease and past sickness absence, emotional support was not related to work ability. The results presented in Table 5 suggest that availability of emotional support provided by a person close's community outside work (family, friends, acquaintances), increases self-appraisal and boosts self-esteem, encourages to be absent from work.

A further question is how this applies to the perception of future work ability. It is possible to expect that absence in response to e.g. perceived strain at work, would actually reduce the risk of future inability to work. It is also possible that there is an element of reverse causality: that workers with a perception of decreased work ability may elicit more emotional support. This should be investigated further.

Several studies have shown that self-reported sickness absence is highly correlated with administrative information on such absence and have concluded that self-reported data are sufficiently valid measures for its correct assessment [62, 63]. Furthermore, self-reported data provide information on the entire period of sickness absence, including also the first week of sickness absence, when no sickness absence certificate is needed.

In the analysis of psychosocial resources and previous back pain, all multivariate analyses were adjusted for SES (socioeconomic status measured as occupation), as SES might cause both workplace exposures and poor health [64]. The model that includes SES might be over-adjusted because of the strong relation between occupation and psychosocial factors in the workplace [10, 15]. The true relation between job strain, previous back pain, and sickness absence is probably between the unadjusted (Model I), and adjusted rate ratios (Model III).

The strength of the LSH study is its longitudinal design and a randomised sampling strategy. Several known and potential determinants, such as smoking, alcohol consumption, high BMI, and health status at baseline [10, 54, 65, 66], were controlled in the analysis, which therefore limited confounding bias. Health behaviour (i.e. smoking) may be part of the causal pathway linking exposures to psychosocial factors at work and sickness absence and adjustment for these factors might reduce the true effect of the psychosocial work environment on sickness absence. The analysis was adjusted for past sickness absence, as previous research has shown that sickness absence is a strong predictor of future absence [29].

A limitation was the relatively small size of the study population and the subsequent low statistical power, leading to a possible non identification of true effects. Another limitation resulted from loss of a number of participants because of non-response and because some individuals were no longer gainfully employed at follow-up. There was a pattern of non-response correlated with occupation that tended towards a healthy worker effect selection, again leading to a possible underestimation of true effects. The baseline work ability data is not available and thus it was not possible to study the association between work ability at baseline and work ability at follow-up. The potential limits of self reported measures in terms of common method variance or shared response biases may lead to an overestimation of associations between exposure and outcome variables. Negative affect could be mediating the effect of personality on absenteeism [67] and could affect the response, but this was not controlled for in the model. A final limitation is that these data do not provide information about how the exposure variables (work environment, emotional support, psychological resources, back/neck pain) developed between baseline and follow-up