Leadership at work and risk of treatment for depressive and anxiety disorders in Denmark. A nationwide prospective study with register-based follow up

Positive leadership behaviours at work are associated with worker well-being and performance. However there is less knowledge about whether exposure to low levels of positive leadership behaviours increase workers’ risk of clinical mental disorders. We investigated whether low levels of positive leadership


Introduction
There is an increasing discussion about work environment conditions as possible contributing factors in the aetiology of common mental disorders, in particular depression.In 2022, the World Health Organization (WHO) published their "Guidelines on mental health at work" (World Health Organisation, 2022), emphasizing the importance of working conditions for workers' mental health and arguing for using the workplace as an arena for primary, secondary and tertiary prevention.
The epidemiological evidence for adverse working conditions as a causal risk factor in the aetiology of mental disorders is, however, debated.Whereas a recent meta-umbrella review concluded that there is convincing evidence for a causal contribution of at least one psychosocial work environment factor (job strain, i.e., the combination of high job demands and low job control) for risk of onset of depressive disorders (Arango et al., 2021), another recent comprehensive review stated that methodological limitations in the literature preclude drawing any firm conclusion (Mikkelsen et al., 2021).Among other things, the authors of the latter review expressed concern about the use of self-administered rating scales for measuring mental disorders.Further, the authors highlighted that studies investigating associations between work environment factors and mental disorders usually lacked adjustment for important confounders, including adverse life events and childhood adversities.
Among risk factors in the work environment, leadership behaviours have been only seldom examined in large-scale prospective epidemiological studies.This is surprising, considering that leadership behaviours are not confined to a specific job type or sector and are amendable to change through interventions (Lacerenza et al., 2017).
Even though there is some consensus in the literature that leadership at work can be associated with the mental health of the workers (Inceoglu et al., 2018;Montano et al., 2017;Nielsen and Taris, 2019), studies are mainly small-scale and cross-sectional without clinical measures of mental health (Montano et al., 2017).To our knowledge, only two epidemiological studies have examined so far the prospective association between a broad measure of leadership behaviours and risk of depressive disorders, both using purchase of antidepressant medication as the outcome measure (Bonde et al., 2009;Madsen et al., 2014).In both studies, leadership behaviours were not related to risk of depressive disorders.
In the present study, we used a broad measure of positive leadership behaviours, including support and other types of leadership behaviours that provide workers with resources such as role clarity and decision latitude.
The aim of this study was to investigate, in a large-scale nationwide prospective cohort study with comprehensive adjustment for a wide range of potential confounders, whether exposure to low levels of positive leadership behaviours were associated with an increased risk of onset of depressive and anxiety disorders (measured by treatment registered in national health register data).To identify groups with a particularly elevated risk, or other factors that might mitigate or enhance effects of low levels of positive leadership behaviours, we further analysed effect modification by several covariates, including sex, age, educational attainment, cohabitation, young children at home, job sector, history of other psychiatric diagnoses, clinical levels of selfreported depressive and anxiety symptoms at baseline, a history of severe life events in the year preceding the baseline assessment and different childhood adversities.

Participants
The study population was derived from the Work Environment and Health in Denmark survey (WEHD), a biennial nationwide survey of the Danish workforce that included a workplace sample and a random national sample (Johnsen et al., 2019).The survey had four waves, the first taking place in 2012.We included all 88,076 unique responders; if an individual participated in more than one wave, we used the first response.The response rate was 51.8 %.We excluded participants that were not in paid work at baseline (n = 5713) and who either reported that they did not have a leader or did not reply to one or more of the questions regarding leadership (n = 4768).Of the remaining 77,595 participants, we excluded 699 participants who had missing values on register-based variables (sex, age, educational level, cohabitation, job type or sector).Lastly, we excluded 17,151 participants, who were registered with the outcome, i.e., hospital or psychotropic treatment for depressive or anxiety disorders before baseline.The final sample consisted of 59,743 participants.A flowchart of the exclusion process is provided in the supplementary online material (Supplementary Figure 1).
According to Danish law, by participating in a survey individuals give consent to link the survey data to Danish register data.Therefore, the Danish National Center of Ethics did not request collection of further consent and ethical approval for this study.

Leadership behaviours
The leadership behaviours were measured with an index consisting of eight positive leadership behaviours related to providing different resources at work.The items assessed leadership behaviours related to clarity of organisational goals, adequate decision latitude, professional development, involvement in planning of work, feedback, recognition, social support, recognition, and trust.A previous validation showed an acceptable fit of the scale as an formative measure of leadership behaviours (Sørensen et al., 2022).A complete list of the items is provided in Supplementary Table 1.Every item had six possible response categories.The first five options were "always"=1, "often"=2, "sometimes"=3, "rarely"=4, "never"=5.The sixth option was "I have no leader", which was used to exclude participants without a leader.We combined the eight items into a leadership index that ranged from 8 to 40, with low scores indicating high levels of the positive leadership behaviours and a high score indicating low levels.We divided the index scores in four groups, "high" (score 8-15), "medium high" (16-23), "medium low" (24-31), and "low" (32-40) levels of the leadership behaviours.We chose these cut-off points based on the wording of the response categories, with 16, 24 and 32 points corresponding to the response options "often", "sometimes" and "rarely" on all eight items, respectively.
In the analyses with hospital treatment as the outcome and when stratifying by covariates, number of cases were very low.To increase the statistical power in these analyses, we dichotomised the scale scores and formed two exposure groups using a cut-off score of 24.We also used this dichotomised measure in analyses of effect modification.Furthermore, we tested each item separately, grouping the options "always", "often" and "sometimes" together to form the reference category and the options "rarely" and "never" to form the exposure category.

Combined and separate measures of treatment for depressive and anxiety disorders
To identify cases of depressive and anxiety disorders before baseline and during follow up, we merged the WEHD data on leadership behaviours with data from national health registers on (i) in-and outpatient treatment at psychiatric hospitals and (ii) redeemed psychotropic prescriptions with antidepressants and anxiolytics (Pedersen, 2011;Thygesen et al., 2011).More specifically, we identified hospital treatment with the ICD-10 codes F32 (depressive episode) or F33 (recurrent depressive episode) for depression and F41 (other anxiety disorders) for anxiety disorders.For redeemed prescriptions, we used the Anatomical Therapeutic Chemical (ATC)-codes N06A and N05B for antidepressants and anxiolytics, respectively.
We created a combined outcome variable that indicated any treatment for a depressive or anxiety disorder, as well as variables for the four outcomes separately (antidepressant treatment, anxiolytic treatment, hospital treatment for depressive disorder, hospital treatment for anxiety disorders).In the combined outcome, time to event corresponded to the first time any of the outcomes occurred during follow-up.

Demographic covariates and job characteristics
We included sex, age, educational level, cohabitating partner or spouse, child under age 7 and hospital treatment for other psychiatric disorders before baseline.Job type and whether the participant worked in the public or private sector were used as covariates accounting for job characteristics.These covariates were retrieved from registers.Sex (men, women), cohabitation (yes, no) and child under age 7 (yes, no), public or private sector employment, and treatment for other mental disorder were used as binary variables.Educational level was based on the highest achieved education in the year prior to baseline, with the three levels low, medium, and high, corresponding to primary, secondary education or vocational training, and tertiary education, respectively.For analyses of effect modification, we collapsed the low and medium levels.Age was treated as a continuous variable in all analyses, except when examining effect modification of age, here the variable was dichotomised by the mean (<44.5 vs. ≥44.5 years).Job type was a 10-level variable, categorised according to the Danish version of the International Standard Classification of Occupations (ISCO)-system (International Labour Organization, 2018).Finally, we included two variables pertaining to the WEHD sampling design (Johnsen et al., 2019), one indicating whether the respondents were sampled through the workplace sample or the random sample, the other indicating the year of the WEHD wave.

Adverse life events
We identified adverse life events in the year before baseline in national registers.Adverse life events included a) either diagnosis of a mental disorder, diagnosis of a severe somatic illness or the death of a cohabiting partner/spouse, child(ren) or parent(s), b) serious somatic illness of the participant or c) a divorce or separation, conceptualised as end of cohabitation with a cohabiting spouse/partner.Serious illness of the participants, their partner or parents were identified using the Charlson comorbidity index and included for example cancer, myocardial infarction or liver disease (Charlson et al., 1987;Hude et al., 2005).For participants' children who were older than 18 years at baseline, we used the Charlson Comorbidity Index to identify serious illness.If the participants' child was below age 18 at baseline, we used the 10 most fatal childhood illnesses in Denmark (Rod et al., 2020).
We further identified adverse life events during the two-year followup period.If the participant had an adverse life event during follow up, we counted the time before the life event as time without a life event, and the time after the life event as time with a life event (Zhang et al., 2018).
We also constructed three variables for childhood adversities of the participant based on register data.The first variable indicated whether the parents of the participant did not live together when the participant was 15 years old (as a proxy for family disturbance).The second variable indicated whether the participant's parents either died, were sick with an illness from the Charlson Comorbity Index, or received a psychiatric diagnosis when the participant was aged 0-18.The third variable indicated household income (in quartiles and adjusted for family size and composition) in the year the participant turned 15.For analyses of effect modification, we dichotomised the income variable, using the lowest quartile as the risk group and the other three quartiles as reference croup.
For the three variables of childhood adversities, we did not have the information for all participants, because of the limited retrospective availability of the registers.If the participant was born too early to be covered by the register or the information was not available for other reason (e.g., immigration), the participant was categorised as "unknown" on the three childhood adversities.

Depressive and anxiety symptoms at clinical level at baseline
We measured depressive and anxiety symptoms at baseline by validated self-administered ratings scales.Depressive symptoms were measured with the Major Depression Inventory (MDI) (Bech et al., 2015), while anxiety symptoms were measured with the SCL-ANX4 subscale of the Symptom Check List (Christensen et al., 2005).In accordance with the literature, we considered an MDI-score ≥21 (Bech et al., 2015) and a SCL-ANX4 score ≥3 (Christensen et al., 2005) as indicative of clinically relevant depressive and anxiety symptoms, respectively.In the analyses, we used a categorical variable with three levels for both depressive symptoms and anxiety symptoms; (i) no clinical level, (ii) clinical level, (iii) missing value.

Statistical method
We used Cox proportional regression with chronological time in days as the underlying time scale to analyse the association between exposure to low level of positive leadership behaviours at baseline and treatment for depressive or anxiety disorders during follow-up.We analysed treatment both as a combined variable (combining hospital treatment and psychotropic drug treatment for both depressive and anxiety disorders) and as four separate variables.We censored for the outcome, emigration, death, or end of follow up, whichever came first.Each participant contributed with follow-up time up until two years after filling in the baseline questionnaire.We made an exception for the 4th wave of WEHD (the 2018 wave), which had a shorter time of follow-up since data collection ended on the 1st of February 2019.
In the first model, we adjusted for sex, age, educational level, cohabitation, children under age 7, treatment for a mental disorder other than depressive or anxiety disorders at a hospital before baseline, job type, sector, the three variables for childhood adversities, life event before baseline, life event during follow-up, sampling method and WEHD wave.In a second model, we further adjusted for self-reported clinical levels of depressive symptoms or anxiety symptoms at baseline.
When analysing effect modification by the covariates, we used the combined measure as the outcome, and used the p-value of the multiplicative interaction term as indicator for presence or absence of effect modification.In the interaction analyses with baseline symptoms of anxiety or depression, and the three variables of childhood adversities we removed the participants with missing values on the relevant variable before conducting interaction analysis.
Information on childhood adversities, was not available for the large majority of participants above age 44.5, therefore when we investigated effect modification by age we did not include the variables on childhood adversities in the models in any of the strata to make the results comparable.

Results
During a mean of 1.70 years of follow-up, we identified 999 cases in the combined measure of depressive or anxiety disorders (96.6 cases per 10,000 person-years).Most of the cases were due to redeemed prescriptions of antidepressants, followed by prescriptions of anxiolytics.Cases due to hospital treatment were rare.
Table 1 shows the characteristics of the participants at baseline and cases per 10,000 person-years during follow-up for each characteristic.The sample included 50.0 % women; the average age was 44.5 years, and 44.2 % had a high educational level.All main job groups (at the ISCO-10 levels) were represented in the study sample, with 41.2 % working in the public sector.The number of cases was higher among women and among participants with low educational level.Adverse life events before baseline were weakly associated, whereas life events during follow-up were strongly associated with an increased likelihood of becoming a case.With regard to childhood adversities, participants were more likely to become cases when they had experienced an adverse life event at age 0 to 18 and when they had lived in a household with low income at the age of 15.
Fig. 1 shows the hazard ratios (HR) and 95 % confidence intervals (95 % CI) for the prospective association between the groups exposed to low levels of positive leadership behaviours at baseline and risk of treatment for depressive and anxiety disorders during follow-up.Compared to high levels of positive leadership behaviours, lower levels were associated with a statistically significant higher risk for the combined outcome of treatment for depressive and anxiety disorders.As the levels of the positive leadership behaviours decreased, the risk of the combined outcome increased in a linear exposure-response fashion (p for trend: p < 0.001).In model 1, the group with medium-low levels of positive leadership behaviours displayed a HR of 1.43 (95 % CI: 1.20-1.70)and the group with low levels displayed a HR of 2.05 (95 % CI: 1.56-2.69)compared to the group with high levels.In model 2, where we further adjusted for self-administered clinical levels of depression and anxiety at baseline, the estimates attenuated to 1.14 (95 % CI: 0.95-1.36)and 1.36 (95 % CI: 1.03-1.83),respectively.
Fig. 1 also shows the estimates for the separate outcomes, that is, antidepressant treatment, anxiolytic treatment, depression treatment at a hospital and anxiety treatment at a hospital.In the most adjusted model (model 2), the HRs in the group with low levels of positive leadership behaviours were similar to the estimates in the main analyses, albeit not statistically significant for antidepressant treatment (HR: 1.38 95 % CI: 0.98-1.93)and anxiolytic treatment (HR:1.33 95 % CI: 0.80-2.21).There were only very few cases with hospital treatment for either depression or anxiety disorders, resulting in imprecise estimates with wide 95 % CIs.When we further examined the adjustments with adverse life events and childhood adversities variables in more detail, we observed that gradually adjusting for childhood adversities and life events before baseline had virtually no effects on the estimates.Further adjusting for life events during follow-up slightly attenuated the estimates (See details in Supplementary Table 2) Table 2 gives an overview of the results of the analyses investigating effect modification.Only age showed a statistically significant effect modification.In the adjusted model, the HR for the association between low levels of positive leadership behaviours (dichotomised variable) and risk of the combined outcome of treatment for depressive and anxiety disorders was 1.25 (95 % CI: 1.04-1.50)for participants ≥44.5 years and 1.69 (95 % CI: 1.40-2.04)for participants <44.5 years, respectively.For further details see Supplementary Table 3.
The analyses for each leadership behaviours item are shown in Table 3.The associations with risk of the combined outcome were strongest for "help and support from the leader" and "getting sufficient authority in relation to responsibility" and weakest for "being involved in planning of work" and "leader explains the company's objective".

Discussion
We found prospective associations between low levels of positive leadership behaviours and an increased risk of treatment for depressive or anxiety disorders.The associations followed a linear exposureresponse trend, such that lower levels of positive leadership behaviours were associated with a higher risk of the outcome.Thus, the results of the present study adds to the evidence that leadership behaviours at work may have a role in the aetiology of depressive and anxiety disorders.When splitting the data in two strata a large majority of participants above age 44.5 had missing on the childhood conditions, therefore the three variables pertaining to childhood conditions were left out of the analysis on effect modification by age in both strata to make the strata comparable.** The analyses were done on a smaller subsample, where participants with missing/unknown where excluded.
The strongest association was found for antidepressant treatment.The point estimates were similar for anxiolytics and antidepressants, but the association was weaker for the former.Due to the small number of cases, the point estimates for hospital treatment of depression and anxiety were highly imprecise, precluding any interpretation.
Age was the only statistically significant effect modifier in the study, and the results suggested that the association between low levels of positive leadership behaviours and the combined outcome was stronger in the younger than in the older participants.This result could perhaps be explained by younger workers being more dependent on their leaders, due to less experience and seniority at the workplace.This hypothesis should be examined in future studies.
The risk of the different outcomes decreased when adjusting for selfreported clinical levels of depressive and anxiety symptoms at baseline.The association remained statistically significant only for the combined outcome.Depressive symptoms at baseline may be a confounder, as they may lead both to an over-reporting of negative leadership behaviours and an increased risk of depressive and anxiety disorders and thus cause inflated estimates (Mikkelsen et al., 2021).However, the analyses of effect modification showed that the association between low levels of positive leadership behaviours and the outcome was weaker in the group with clinical levels of depression or anxiety.Further, we were concerned that adjusting for depressive symptoms might introduce over adjustment bias, if baseline depressive symptoms were not a confounder but a mediator of the association between leadership behaviours and subsequent depressive disorder (Madsen et al., 2017).Therefore, we consider model 1 (i.e., the model without adjustment for self-reported clinical levels of depressive and anxiety symptoms at baseline) as the main model but also provided, in the interest of completeness, a model 2, with adjustment for baseline symptoms.

Comparison with previous studies
Two recent systematic reviews reported associations between various types of leadership behaviours and workers' well-being and mental health (Inceoglu et al., 2018;Montano et al., 2017).However, the quality of evidence in these reviews was low, as the majority of the studies reviewed were based on a cross-sectional design, included small samples from selected jobs or sectors or convenience samples, or used self-report measures to assess mental health problems (Inceoglu et al., 2018;Montano et al., 2017;Nielsen and Taris, 2019).The present study overcomes some of these challenges by virtue of its large sample size, its prospective design and the use of register-based clinical outcomes.
To our knowledge, only two studies have previously analysed the association between a broad measure of leadership behaviours and clinical measures of depressive disorders, both employing prescribed antidepressant medication as the outcome measure (Bonde et al., 2009;Madsen et al., 2014).In contrast to our results, these studies did not report any associations between leadership behaviours and risk of antidepressant treatment.A possible explanation is that our study used a different measure of leadership behaviours, which only partly overlapped with the measure adopted in the other studies.Another

Table 3
The association between low levels on single items and risk of combined outcome for treatment for depression and anxiety.Model 2: Like model 1, but further adjusted for self-reported depressive symptoms at clinical level at baseline, and self-reported anxiety symptoms at clinical level at baseline.
explanation could be the higher statistical power of our study, as our study included 59,743 participants with 999 cases, compared to 21,129 participants with 574 cases (Bonde et al., 2009) and 9507 participants with 407 cases (Madsen et al., 2014) in the two previous studies, respectively.

Strengths
The strengths of the present study include its prospective design, allowing us to exclude participants with a history of treatment for depression or anxiety at baseline.This supports internal validity in terms of the temporal sequence of the exposure and the outcome.The study was conducted in a large national sample of workers in Denmark and was not restricted to specific occupational groups, enabling analyses with adequate statistical power and allowing the generalisation of the results to the Danish workforce.By merging the questionnaire data with register data, we used two different methods for collecting exposure and outcome data, which addresses possible common method bias.As antidepressants and anxiolytics can only be legally purchased by prescription, using register data for the outcome measure means that we have a complete follow-up for all participants with no attrition and that the assessment of the need for treatment was ascertained by a medical professional.
We adjusted for severe life events, both before baseline and during follow-up, as well as for childhood adversities, such as income level in the household, family disruption and serious illness or death of parents while the participant was between the age of 0 and 18.A severe life event may confound the association between experienced leadership behaviours and the risk of depressive and anxiety disorders.For instance, it is conceivable that the experience of a severe life event (e.g., the loss of a significant other) may increase the risk of depressive or anxiety disorders and reduce effective functioning at work; in turn, these may induce conflicts with the supervisor and lead to a more negative self-reporting of leadership behaviours.It is also conceivable that childhood adversities may act as a confounding factor in the association between working conditions and mental health problems, since childhood adversities are known risk factors for mental health disorders (Arango et al., 2021) and might also affect educational attainment and subsequent career opportunities.Notwithstanding this, adjusting for severe life events in the year preceding baseline measurement and for childhood adversities left all estimates virtually unchanged.When we further adjusted for severe life events during follow-up, estimates were only slightly attenuated.Life events during follow-up could be a confounder, if they impact on both leadership behaviour and risk of depressive and anxiety disorders in a similar way as life events before baseline as described above.However, they could also be mediators, if the exposure, i.e., leadership behaviours, have caused the life event during follow-up (e.g., divorce/separation or physical illness) and in this case, adjustment for life events during follow-up would have been an over-adjustment.This is, though, mostly a theoretical concern, as it empirically turned out that adjustment for life events during follow-up only had a marginal effect on the estimate for the association between leadership behaviours and risk of depressive and anxiety disorders.

Limitations
Since Danish health register data are collected for administrative rather than for research purposes, we could only identify treatment for psychotropic drugs and treatment at hospitals but not other types of treatment, e.g., non-pharmaceutical treatment by the general practitioner or psychological treatment in private practice (Thygesen et al., 2011).As it is voluntary for individuals to seek help in the Danish health care system, we could not detect persons with a depressive or anxiety disorder that was not treated (Thielen et al., 2009).Therefore, our sample only includes a sub-group of individuals in Denmark with depressive and anxiety disorders.A recent Danish study indicated that a substantial proportion of individuals with clinically significant depressive symptoms (measured by a self-administered rating scale) are not treated (Weye et al., 2023).It is possible that those who receive treatment, either in a hospital or by being prescribed psychotropic drugs, are more severe cases of depressive and anxiety disorders than those who are not treated.In this case, the results of our study would mainly pertain to individuals with more severe forms of depressive and anxiety disorder.
The measure of leadership behaviours had not been validated before collection of the survey data.We therefore conducted a validation of the scale and found it had satisfactory psychometric properties for a formative scale (Sørensen et al., 2022).Finally, it is a limitation that we only had one measurement point for assessing leadership behaviours and could therefore not take into account cumulative exposure or changes in exposure.

Conclusion
The results of this epidemiological study within a large sample of the Danish workforce suggest that exposure to low levels of positive leadership behaviours is a risk factor for the onset of treatment for depressive and anxiety disorders.

Declaration of competing interest
None.

Table 1
Sample characteristics.
Association between level of leadership behaviours at baseline and risk of treatment for depressive or anxiety disorders at follow-up.Both combined and separate outcomes are displayed.LB abbreviation for Leadership Behaviours.
Model 1: Adjusted for sampling method, sampling wave, sex, age, educational level, cohabitation, child under the age of 7, job type, public/private sector, hospital treatment for other mental health disorders before baseline, life event before baseline, life event during follow up, childhood conditions (family disruptions, income level in household at age 15 and life event related to parents' health or death between age 0 and 18).Model 2: Like model 1, but further adjusted for self-reported depressive symptoms at clinical level at baseline, and self-reported anxiety symptoms at clinical level at baseline.

Table 2
Results from analyses of interaction as an indicator of effect modification of the association between leadership behaviours and treatment for depressive or anxiety disorders by the list of covariates.

often do you get the necessary help and support from your immediate leader?
Model 1: adjusted for sampling method, sampling wave, sex, age, educational level, cohabitation, child under the age of 7, job type, public/private sector, hospital treatment for other mental health disorders before baseline, life event before baseline, life event during follow up, childhood conditions (family disruptions, income level in household at age 15 and life event related to parents' health or death between age 0 and 18).