Mental illness rates among employees with fixed-term versus permanent employment contracts: a Danish cohort study

Main findings

We found that the rate ratios for use of psychotropic drugs and psychiatric hospital treatment due to mood, anxiety or stress-related disease, in the Danish labor force, were statistically significantly higher among employees with fixed-term vs. permanent employment contracts. The tests for interactions with age, gender and education level were not statistically significant.

Results in relation to previous research

We found four relevant studies that estimated longitudinal associations between fixed-term vs permanent employment and indicators of mental ill health, one from Germany (Demiral et al. 2022), one from Sweden (Hammarström et al. 2011) and two from Finland (Virtanen et al. 2008; Ervasti et al. 2014).

The German study dealt with employees in employments subject to social security payments (Demiral et al. 2022) aged 31–60 years—representing 80% of all people working in that age range (n = 2009). Odds ratios for depressive symptoms as a function of fixed-term employment contract (yes vs. no) were 2.20 (95% CI 0.80–6.06) among men and 1.42 (0.61–3.32) among women. The analyses were adjusted for baseline (2012) age, partnership status and socioeconomic position. The study population of the Swedish study (Hammarström et al. 2011) consisted of all ninth-grade graduates of the calendar year 1981, in Luleå, who held temporary and/or permanent employment contracts between the age of 30 and 42 years (n = 660). Questionnaire data were collected at the age of 30 and 42 years. Odds ratios at the age of 42, for the contrast “temporary employment for a total time of more than 10 months” versus “permanent employment during the whole 12-year period” were estimated at 1.90 (95% CI 1.33–2.71) for psychological distress and 1.79 (95% CI 1.04–3.08) for depressive symptoms. The analyses were controlled for gender, self-rated health, psychological distress and depressive symptoms at age 30.

One of the Finnish studies (Virtanen et al. 2008) examined associations between temporary employment and redeemed prescriptions for antidepressant medication (1998–2002) among 17,071 men and 48,137 women employed municipalities in Finland. After adjustment for age, socioeconomic status (SES), and calendar year, the odds ratio for the contrast “fixed-term > 6 months vs. permanent employment” was estimated at 1.18 (95% CI 1.03–1.37) for antidepressant use in men and 0.99 (95% CI 0.93–1.06) in women. The corresponding odds ratios for the contrast “fixed-term <  = 6 months vs. permanent employment” were estimated at 1.43 (95% CI 1.19–1.73) in men and 1.18 (95% CI 1.09–1.28) in women, and for the contrast “subsidized temporary work vs. permanent employment” they were estimated at 1.57 (95% CI 1.23–2.02) in men and 1.38 (95% CI 1.20–1.59) in women. The association between type of employment contract and use of antidepressants was statistically significantly weaker among women than it was among men (p = 0.007). The association was, moreover, weaker among men with high SES than it was among men with low SES (p = 0.033).

The other Finnish study (Ervasti et al. 2014) examined associations (2005–2011) between temporary vs. permanent employment and sickness absence due to medically certified depressive disorders (ICD-10 codes F32–F34) among 107,828 Finnish public sector employees. The concerned rate ratio was estimated at 1.02 (95% CI 0.97–1.08). The analysis was adjusted for age, gender, level of education, chronic somatic disease and history of work disability due to mental or behavioral disorder (ICD-10 codes F00–F99). No significant interaction with gender, age, or education was observed (p > 0.25).

The associations between fixed-term contracts and mental ill health that were observed in the present study aligns well with the findings of the Swedish study (Hammarström et al. 2011) and the first of Finnish studies (Virtanen et al. 2008). The German study’s relatively small population size combined with low prevalence of fixed-term contracts might explain its insignificant findings (Demiral et al. 2022). A possible explanation for the null-finding observed in the second of the Finnish studies (Ervasti et al. 2014) is that it did not estimate rate ratios for depressive disorders but for sickness absence due to depressive disorders. Some workers with depressive disorders may call in sick while others may continue to work, and it has been shown that temporary employees, due to job insecurity, tend to have higher rates of sickness presence than permanent employees do (Virtanen et al. 2003; Reuter et al. 2019). It might also be that different levels of employment protection in fixed-term contract and in permanent contracts across countries could lead to country dependent associations between contract type and health (Voßemer et al. 2018).

Methodological considerations

Our study has several strengths. The study was quite large and the statistical power was high enough to investigate main effects of having a fixed-term versus permanent employment contract. Bias from missing follow-up data was substantially reduced, since the endpoints of the study were ascertained through national registers that cover all inhabitants of the target population. Within-study selection bias was eliminated, since all hypotheses, significance criteria, endpoints and statistical methods were completely defined and published before we looked at any relation between the exposure and outcome data of the study (Hannerz et al. 2021).

The major drawback of the study is that it is observational and thereby has a weaker design than a randomized controlled trial, which is the golden standard in determining causality. Another weakness is the low response rate, which means that we cannot rule out the possibility of non-response bias. We believe, however, that any such bias has been mitigated by the many control variables that were included in the analyses. Individual participant data were available on a large variety of socioeconomic and occupational factors, which enabled us to control the analyses for a series of possible confounders and health selection effects such as age, gender, education, industry, nighttime work, unemployment benefits and income. Control for unemployment is relevant in order to take selection into part-time work into account. Control for income is important, as it has been found that effects of insecurity in employment can be alleviated by increased wage levels (Böckerman et al. 2011).

It has been shown that the risk of developing depression is associated with smoking habits (Pasco et al. 2008; Korhonen et al. 2007) and body mass index (Luppino et al. 2010). In the present study, we did not have any individual participant data on smoking habits and body mass index, and could therefore not include these factors as control variables in the analyses. We had, however, access to collateral data, which we have used to estimate age, gender and education standardized prevalence of smoking, overweight, and obesity among 20–59 year-old employees in Denmark, by type of employment contract (Hannerz et al. 2021). The estimated prevalence among people with fixed-term contracts were very similar to those among people with permanent contracts. It is therefore unlikely that the results of the present study have been influenced by differential prevalence of smoking, overweight and obesity.

We have not conducted any validation study of self-reported information on employment contract. We believe, however, that most (if not all) employees know if they have a permanent or temporary employment contract and that the question that was used to obtain the information in the present study is very easy to understand and difficult to misinterpret. Moreover, the question is not sensitive and it is not subject to recall bias. It should, however, be noted that our analysis do not account for time-variant unobservable characteristics that may have an impact on the results. It is, for example, possible that a person with fixed-term employment at baseline will become permanently employed or unemployed during the 5 year follow-up period. It is also possible that a person with permanent employment at baseline will become unemployed or shift to fixed-term contract position. Such transitions are probably associated with a bias toward unity.

In the present study, we used rate ratios of hospital treatment and redeemed prescriptions of drugs as proxy measures for underlying morbidity ratios. Hence, we need to consider the possibility of detection, prescription, and referral bias. In Denmark, all citizens are covered by a tax-funded health insurance, which enable them to consult a general practitioner and to receive psychiatric treatment free of charge, whenever it is needed. Since fixed-term and permanent employees have equal access to general practitioners as well as psychiatric hospitals and specialists, we do not think that the present study is subject any detection, prescription or referral bias of practical importance.

Psychiatric treatment is a rare event; hence, insufficient statistical power restricted the study of that outcome to a main effect only model. Psychotropic drugs include a few types that are used for disorders not expected to be associated with stressors like fixed-time contracts, e.g., psychostimulants and antidementia drugs. However, only a few promille of the cases were due to such drugs.

The underlying research hypothesis of the present project was that objective job insecurity may act as a stressor that increases a person’s vulnerability to mental ill health, without further specification. From this viewpoint, it may seem natural to include all types of mental disorders in the case definition of psychiatric hospital treatment. We chose, however, to exclude the vast majority (87%) of the diagnoses listed in the chapter on “mental and behavioral disorders” of the ICD-10 classification, and to focus solely on diagnoses that are labeled as mood, anxiety or stress-related disorders. We excluded F00–F09 “Organic mental disorders” because of their etiology in cerebral disease or brain injury, which make them quite irrelevant to the context of the present study; F60–F69 “Personality disorders”, F70–F79 “Mental retardation”, F80–F89 “Disorders of psychological development” and F90–F98 “Behavioral and emotional disorders with onset usually occurring in childhood and adolescence” because such disorders typically develop well before the entering of the labor market; somatoform disorders, firstly, because of an extraordinarily long expected duration between the onset of the complaints and the diagnosis (Herzog et al. 2018) and, secondly, because the labeling of such disorders as mental illnesses is controversial (Rief and Isaac 2007; Kroenke 2007); F10–F19 “Mental and behavioral disorders due to psychoactive substance use”, F42 “Obsessive–compulsive disorder” and F50–F59 “Behavioral syndromes associated with physiological disturbances and physical factors” because we wanted to keep our case definition simple and easy to communicate, which would not have been the case if we had included these diverse sets of behavioral disorders; and F20–F29 “Schizophrenia, schizotypal and delusional disorders” because they are associated with an extraordinarily high heritability (Hilker et al. 2018) and a low labor market attachment (Marwaha and Johnson 2004; Rinaldi et al. 2010), which make them quite irrelevant to the context of the present study. Here, it should be noted that the last mentioned category contains F25 “Schizoaffective disorders” and that manic, bipolar and depressive schizoaffective disorders thereby were excluded from our case definition.

We chose to base our case definition on diagnostic standard groupings at the two- or three-character level rather than on an ad hoc collection of four-digit level sub-categories, for several reasons. Firstly, because we wanted to decrease the probability that relevant cases were missed. Secondly, because the probability of misclassifications, i.e., false positive and false negative diagnoses, are likely to be higher at the four-character level than they are at the two and three-character level (Jensen et al. 2010). Thirdly, because a wider diagnostic category is less sensitive to random variation than a narrower diagnostic category.

Since (i) participants who received social security cash benefits, sickness absence benefits, psychotropic medicines or psychiatric hospital treatment within a 1-year period prior to baseline were excluded from the analysis and (ii) most of the mental disorders that are likely to depend on factors occurring before adulthood were excluded from the case definition, we do not believe that the study is subject to reverse causality bias of practical importance. The case definition included, however, phobic anxiety disorders (ICD-10: F40), which often manifest themselves already in childhood or adolescence (Kessler et al. 2007; Solmi et al. 2022). It is possible that some of the cases of phobic anxiety that were observed in the present study existed already at the start of the follow-up. It is also possible that that some people may be unable to obtain or hold a permanent employment position due to phobic anxiety disorders. Hence a possibility of reversed causation. To explore this possibility, we conducted a post hoc sensitivity analysis in which we excluded phobic anxiety disorders from the case definition. All other details of the analysis were the same as in the primary analysis of the psychiatric hospital treatments. In this post hoc sensitivity analysis, the concerned rate ratio was estimated at 1.42 (99.5% CI 1.06–1.91).

To further explore the possibility of reversed causation in the analysis of the hospital treatment data, we conducted a post hoc sensitivity analysis in which we extended the required period of “no psychiatric hospital treatment” from one to five-year prior to the baseline interview. All other details of the analysis were the same as in the primary analysis of the psychiatric hospital treatments. In this post hoc sensitivity analysis, the concerned rate ratio was estimated at 1.34 (99.5% CI 0.98–1.82).

In the analysis of psychotropic drugs, we aimed at estimating the association between our exposure variable and redeemed prescriptions for psychotropic medicine, and with such an aim, it made sense to include all types of psychotropic medicine in the case definition.

Two types of health selection bias need to be considered in the interpretation of the results. The first one concerns the possibility of bias due to health selection into a fixed-term or permanent employment position. It is, for example, possible that some people are unable to obtain or hold a permanent employment position due to lingering mental health problems. The second type of bias concerns health selection into the analysis. In our primary analysis, we included only DLFS participants with no social security cash benefits, no sickness absence benefits, no redeemed prescriptions for psychotropic medicines and no psychiatric hospital treatment during a whole year prior to the baseline interview. It was, moreover, required that they were full-time employees at the time of the interview. The purpose of the rigorous inclusion criteria was to counter potential bias from health selection into a fixed-term employment position. The consequence of the rigorous inclusion criteria is that the subset of fixed-term contract employees that was included in the primary analysis was far from representative of the full set of DLFS participants with a fixed-term contract position at baseline. It goes without saying that those who were permanently employed at baseline are more likely to have been permanently employed also prior to baseline and vice versa. Hence, if there are any health risks associated with not having a permanent employment contract then the fixed-term employees who were included in the primary analysis are likely to be more privilege and less vulnerable to the consequences of not having a permanent employment than the ones who were excluded. Seen from this perspective, selecting away cases 1 year—and especially 5 years—prior to baseline can be regarded as a very conservative approach underestimating possible effects of fixed-term contracts on depressive symptoms, as effects can have occurred before the follow-up period.

To shed some light on these health selection effects, we conducted two sensitivity analyses. In one of the analyses, we (i) removed the requirement of not receiving sickness benefits or social security cash benefits during a 1-year period prior to the baseline interview and (ii) removed all control variables except for gender, age, and education. We kept, however, the requirement of full-time employment at baseline and no redeemed prescriptions for psychotropic medicines and no psychiatric hospital treatment during a whole year prior to the baseline interview. The purpose of this analysis was to obtain an unbiased estimate of the rate ratio of psychotropic drug use between “a representative set of the DLFS participants with a full-time fixed-term contract position” and “a representative set of the DLFS participants with full-time permanent employment” after standardization for gender, age and education. The rate ratio in this analysis was estimated at 1.31 (99.5% CI 1.21–1.42). In another sensitivity analysis, we extended the required period of “no redeemed prescriptions for psychotropic medicines and no psychiatric hospital treatment” from 1 to 5 years prior to the baseline interview (on top of the rigorous inclusion criteria and potentially over-adjusted confounder control of the primary analysis). In this sensitivity analysis, the rate ratio of psychotropic drug use between employees with fixed-term vs. permanent employment contracts was estimated at 1.05 (99.5% CI 0.90—1.23). Further details about our pre-specified sensitivity analyses are given in the appendix.

Generalizability

The results of the present paper should be seen in the light of specific conditions at the Danish labor market, which have been labeled flexicurity, a certain combination of low employment contract protection and generous compensation levels regarding unemployment benefits (Bredgaard and Madsen 2018; Madsen 2006, 2013). Effects of fixed-term contracts on health might be dependent on the welfare states’ employment protection regarding fixed-term and permanent contracts (Voßemer et al. 2018). This means that experienced job insecurity in fixed-term and permanent contracts could vary considerably between welfare state regimes making inference of study results across countries difficult.

Conclusions

We know very little about the possible effects of contract type and mental health across countries. Increased cooperation between labor market and health researchers could contribute to shed more light into this question. The present study supports the hypothesis that employment in a fixed-term rather than permanent contract position is associated with an increased risk of developing mental health problems in Denmark. The results in themselves do not warrant specific interventions regarding fixed-term contracts as they can range from (a) restrictions in establishing fixed-term contracts over (b) improvements in working conditions for this group of workers to (c) specific health-related interventions. We can, however, conclude that the results of the study lend support to the necessity of the EU council directive 1999/70/EC of 28 June 1999 concerning the framework agreement on fixed-term work (The Council of the European Union 1999). The notably higher RR within transport and storage and to a lesser extent construction industries might warrant a particular focus on possible preventive efforts in these industries.

留言 (0)

沒有登入
gif