Main findings
In spite of a relatively large sample, we did not find any statistically significant difference in psychiatric treatment rates between fixed-term contract workers and unemployed people in the general population of Denmark. Sixteen rate ratios and confidence intervals were estimated; two in the main effects analyses, ten in the stratified analyses and four in the sensitivity analyses. All of the estimated confidence intervals included unity. Moreover, the tests for interaction with age, gender, and education level were not statistically significant.
Strengths, weaknesses and limitations
The study was large enough to address our research questions, which were raised on the basis of some previous studies. The exposure categories of our study were the same as the ones used in the European labor force surveys [
44]. Another advantage was that the participants were drawn from the general working age population of Denmark.
Epidemiologic studies are often associated with substantial publication bias due to multiple testing of outcomes combined with selective reporting of results [
45]. In the present study, the hypotheses and statistical models were completely specified, peer reviewed, and published before we linked the exposure data to the outcome data [
13]. We adhered to the protocol without violations. The study is thereby free from bias due to selective hypothesis-testing. Since the endpoints of the study were ascertained through national registers, which cover all inhabitants of Denmark, we can rule out bias from missing follow-up data. For the same reason, we can also rule out recall-bias. Register data on social security cash benefits, sickness absence benefits and psychiatric treatment prior to baseline enabled us to identify and exclude potentially unhealthy workers and thereby mitigate the possibility of health selection bias. Register data on age, gender, disposable family income and education enabled us to control for and thereby mitigate the possibility of bias from demographic and socio-economic factors.
Smoking [
46,
47] and overweight [
48] have been associated with an increased risk of depression. In the present study, we did not have any person-based data on these lifestyle factors and could therefore not control for them in the analyses. Based on the prevalence of smoking and overweight in another random sample of fixed-term contract workers and unemployed people in Denmark, we have estimated that a failure to control for smoking and overweight in the present study would bias the rate ratio for mental health illnesses among fixed-term contract workers vs. unemployed downward with a factor of 0.96. Which means that a rate ratio at 0.96 without control for smoking and overweight would correspond to a rate ratio at 1.00 with control for smoking and overweight [
13]. It is, therefore, unlikely that the null finding of the present study was due to a failure to control for smoking and overweight.
Immigrants are highly overrepresented among unemployed people in Denmark [
49]. The rates of psychiatric treatment among the immigrants are, however, quite similar to the rates among native Danes. This was shown in a very large register-based Danish population study [
50] in which the incidence rate ratio among first-generation immigrants vs. native Danes was estimated at 0.97 (95% CI: 0.93–1.01) for any psychiatric contact, 0.98 (0.71–1.32) for bipolar affective disorder, 0.81 (0.74–0.89) for affective disorders and 1.05 (0.99–1.12) for anxiety and somatoform disorders. It is therefore unlikely that the null finding of the present study was due to a failure to control for country of birth.
Some of the covariates and inclusion criteria of the study were based on records in national registers, which only were available among the DLFS-participants who had lived in Denmark throughout a one-year period prior to the interview. We therefore had to exclude those DLFS-participants who had immigrated to Denmark within the one-year period preceding the interview (cf. Fig.
2). This group constituted however less than one percent of all participants, wherefore we assess the effect of excluding them to be negligible.
It has been shown that response rates to Danish health questionnaires is affected by calendar time, age, gender, and educational level [
51,
52]. By controlling these factors in the analyses, we aspired to minimize the possible effect of non-participation bias. The present project had, however, not access to data on all of the sampled individuals. We had only access to data for the responders and could therefore not calculate and compare response rates among fixed-term employees and unemployed. Unemployed are probably overrepresented among non-responders. Hence, we cannot rule out the possibility of non-participation bias.
Since the outcomes of our analyses are based on redeemed prescriptions and hospital diagnoses, we need to consider the possibility of detection, prescription, and referral bias. All citizens of Denmark are covered by a tax-funded health insurance, which, among other things, enables them to consult a general practitioner without charge. The general practitioner may in turn refer the patient to a specialist or a hospital for further examinations or treatments. If the patient is referred to a psychiatric specialist or hospital, then the treatment is free of charge. The tax-funded health insurance may be supplemented with private health insurances, which, among other things, cover the costs associated with minor surgeries and psychological therapy. The number of private health insurance holders has increased from 50,000 in 2001 to 1 million in 2008 and 1.9 million in 2017 [
53,
54]. Unemployed people in Denmark do not usually hold a private health insurance; in 2015, approximately 98% of all private health insurances in Denmark were provided by the employers. As the access to psychological treatment is greater among people with than it is among people without a private health insurance, it is possible that our results have been influenced by detection, referral, and prescription bias towards lower rates among the unemployed. On the other hand, the unemployed are able to consult their general practitioner without having to take time off from their job, which may lead to an increased probability of consultation and thereby an increased probability that a mental health problem is detected. Hence, it is also possible that our results are biased towards higher rates among the unemployed.
A major limitation is the measurement of exposure as only point-prevalence self-reported data. In the primary analysis, the exposure category was defined at a single time point (the first interview). To find out if the estimated strength of the association would change if we based the exposure categories on more than one interview, we conducted a sensitivity analysis, in which we only included people who participated in two or more interview rounds and whose exposure was the same in all of their interview rounds. In this sensitivity analysis, the rate ratio for psychotropic drug use among fixed-term contract workers vs. unemployed was estimated at 0.90 (99.5% CI: 0.68—1.20) [cf. Additional file
1, Table S1], which is lower than 0.98 (the rate ratio obtained in the primary analysis). It is possible that those in fixed-term employment could have an earlier high exposure of unemployment, which could explain the lack of significance when comparing the groups. We can therefore not entirely rule out that a more rigorous control for selection processes would lead to the conclusion that fixed-term contract employments are less detrimental for mental health than unemployment.
In the Organization for Economic Co-operation and Development (OECD), mental health problems constitute the most frequent single cause of disability benefits, and in Denmark, they account for almost half of all new applications for disability retirement [
56]. Unemployment is a significant and important risk factor for mental ill health. From this viewpoint, rate ratios of mental ill health between fixed-term contract workers and unemployed should be of interest in political discussions about the pros and cons of a labor market with a high vs. low proportion of temporary jobs, especially if an increased labor market flexibility is seen as a means of reducing unemployment rates. Contrary to our expectations, the present study did not find any statistically significant differences in psychiatric treatment rates between fixed-term contract workers and unemployed in the general population of Denmark, and can therefore not reject the proposition that fixed-term employment may be as detrimental to an individual’s mental health as unemployment itself [cf. 9]. Our null finding thereby suggests that an increased proportion of insecure jobs (measured as fixed-term contracts) may lead to an increased prevalence of mental ill health. The confidence intervals around the estimated rate ratios of the present study are, however, a bit too wide to allow any firm conclusions on this interesting issue.