Depressive symptoms as a predictor of 5-year subsequent work nonparticipation due to long-term sickness absence, unemployment and early retirement in a cohort of 2,413 employees in Germany

doi:10.21203/rs.3.rs-2586559/v1

Download PDF

Research Article

Depressive symptoms as a predictor of 5-year subsequent work nonparticipation due to long-term sickness absence, unemployment and early retirement in a cohort of 2,413 employees in Germany

https://doi.org/10.21203/rs.3.rs-2586559/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 03 Nov, 2023

Read the published version in BMC Public Health →

You are reading this latest preprint version

Background: We examined the impact of depressive symptoms on events – and duration thereof – of work nonparticipation (long-term sickness absence, unemployment, early retirement).
Methods: We used a 5-year cohort of the Study on Mental Health at Work (S-MGA) based on a random sample of employees subject to social contributions aged 31–60 years in 2012 (N =2413). Depressive symptoms were assessed at baseline through questionnaires, while employment history was recorded in follow-up interviews. Effects of depressive symptoms on work nonparticipation were examined in two-part models, with events analysed by logistic regressions and their duration by generalized linear models (GLMs).

Results: Medium to severe depressive symptoms increased the risk of events of work nonparticipation (males OR=3.22, 95% CI=1.90–5.45; females OR: 1.92, 1.29–2.87), particularly events of long-term sickness absence in both genders and events of unemployment among males. Mild depressive symptoms increased the risk of events of work nonparticipation (males OR=1.59, 1.19–2.11; females OR = 1.42, 1.10–1.84). Among those experiencing one or more events, the duration of total work nonparticipation was twice as high among males [EXP(B) =2.06; 95% CI: 1.53–2.78] and about one third higher [EXP(B) =1.38; 95% CI: 1.05–1.83] among females with medium to severe depressive symptoms.

Conclusions: The present study takes into account the occurrence of work nonparticipation but also its duration, which are both critical for assessing societal consequences of depressive symptoms. It is important to consider mild depressive symptoms and examine several types of work nonparticipation.

(4-6) longitudinal study

depression

sickness absence

employment

disability pension

work participation

In the last two decades, the proportion of sickness absence due to mental health reasons has increased in Germany [1]. The proportion of disability pensions due to mental disorders has shown an increase from 19–43% over 22 years [2]. Approximately half of all days of absence due to sickness are accounted for by periods of absence that last less than six weeks [3]. These data indicate that mental health can have an impact in terms of individual costs (income), direct costs (health costs) and indirect costs (taxes, support payments) for the social security system [4]. Mental health conditions are possible causes for short- and long-term work nonparticipation. Sickness absences are temporary, while early retirement and disability pension (DP) are – with a few exceptions – permanent [5]. In addition to sickness absence and early retirement before statutory retirement, there are other types of work nonparticipation, such as unemployment [6]. Both temporary and permanent states of work nonparticipation are associated with productivity losses for the economy [4].

The attribution of mental health reasons to sickness absence and disability pensions is the result of a decision-making process taking clinical, administrative and regulatory aspects into account [7, 8] Hence, administrative statistics shed light on the predictive role of mental health for work participation only to some degree. Moreover, the increasing number of sickness absences and disability pensions due to mental health that occurred in recent decades is contrasted by a rather stable prevalence of mental disorders in Europe [9, 10]. Epidemiologic evidence on the possible causal relationship between mental health conditions and work (non)participation can be obtained from longitudinal studies examining data on depressive disorders or depressive symptoms as predictor variables and sickness absence, disability pension, early retirement, and unemployment as outcomes. Although not exhaustive, this list of outcomes reflects important aspects of work nonparticipation. The opposite of the latter, i.e., work participation, has been conceptualized as “[…] as the capability and/or opportunity to participate in the workforce, fulfilling one’s work role” [11]. An assessment of the possible impact of health conditions on work nonparticipation should take into account important types of nonparticipation, such as those aforementioned. In addition, the gradual impairment of the work functioning of employees who participate partially in work [11] must also be considered.

Mental health conditions include depressive disorders as one of the most disabling mental disorders [10]. Usually, a diagnostic interview is the basis for ascribing an individual to a discrete depression category [12, 13], whereas depression as a continuum is traditionally assessed by using items and scales administered via paper and pencil questionnaires [14, 15].

The evidence on depressive symptoms as a predictor of work (non)participation based on representative populations is limited. Some studies have indicated an association between depressive symptoms and sickness absence. The Household Income and Labour Dynamics (HILDA) Survey in Australia [16] showed that an increasing level of depressive symptoms was associated with higher incidence rate ratios for short- and long-term sickness absence. A similar result was observed in the Danish National Working Environment Survey (DANES), which linked self-reported depressive symptoms at baseline to 60-week follow-up data on long-term sickness (≥ 4 weeks) from a nationwide register [17]. However, even long-term sickness absence with a minimum of 8 weeks [18] is only one type of exit from paid work. Depressive symptoms seem to have differential effects on outcomes such as sickness absence, unemployment and disability pension. For example, Porru et al. investigated some of these different types of work nonparticipation [19] using the Survey of Health, Ageing and Retirement in Europe (SHARE). They used a cut-off score for depressive symptoms to predict the risk for disability benefits and unemployment. Both males and females above the said cut-off had a hazard rate of HR = 2.46 (95% CI: 1.68–3.60) for disability benefits. The effect of depressive symptoms on unemployment was not significant, but it was more likely for males [HR = 1.55 (95%-CI: 0.94–2.57)] than for females [(HR = 1.14; 95%-CI: 0.73–1.76)] for temporary and permanent exit from paid work by unemployment. Rijn et al. conducted a systematic review of poor health and exit from paid work [20]. They showed differential effects on indicators of poor (mental) health on disability pension, unemployment and early retirement, with mental health presenting the strongest impact on disability pension and unemployment. With few exceptions [19], little is known about these differential effects of depressive symptoms on different types of work nonparticipation.

Nonparticipation can be recorded both as an event and as a quantity, i.e., the duration of such events. Most longitudinal studies have dealt with events as dichotomous outcomes only. Event-related analysis is important for understanding the general causal relationship between depressive symptoms and different outcome events, such as permanent exits from paid employment. The occurrence of an event does not capture the duration of events allowing for the assessment of the associated productivity loss. There have been only a few attempts to model these time aspects, e.g., [21–25].

The present paper addresses events of absence and exit from paid employment along with the duration thereof. Based on the aforementioned literature [19, 25], we do not assume a homogeneous impact of depressive symptoms on different outcomes of long-term absences and exits from paid work and we do not assume equivalent effects for males and females.

Study Context

From 2011 to 2017, the statutory retirement age has increased from 65 years to 65 years plus 6 months in Germany. Old-age pensions before the age of 63 have become very rare. Old-age pensions at the age of 63 or 64 without deductions have been restricted to people with an employment history of ≥ 45 years. Other groups of employees have been entitled to early old-age pensions only in exchange for deductions in pension level. People with an employment history of less than 35 years have lost any access to old-age pensions before the statutory retirement age [26].

In case of sickness absence, the employer provides full sick pay for a period of 6 weeks, followed by lower sickness benefits granted by the health insurance. Such sickness benefits expire after 78 weeks [27].

Rehabilitation to regain the capacity to work is mainly provided by the German statutory pension insurance. Employees are entitled to disability pension if no capacity to work is left [28].

The German context described above is comparable to other western or Nordic EU countries. For instance, in Denmark many contracts establish full sick pay at least for a certain period of time. Sickness benefits are restricted to a maximum of 22 weeks in a nine-month period. In certain situations, this period can be extended [29]. In Denmark, the statutory pension age has been increased from 65 years in 2019 to 67 years in 2022 [30].

Population

We analysed data from the Study on Mental Health at Work (German: S-MGA), waves 1 and 2 as baseline and follow-up. This study was initiated by the Federal Institute of Occupational Safety and Health in collaboration with the Institute of Employment Research (IAB). The latter institute integrates data from all employees in Germany subject to social security contributions within a register of the Integrated Employment Biographies (IEB), which constitutes more than 80% of the working population in Germany. Employees subject to social insurance contributions as of 31 December 2010 were registered in the IEB; civil servants, self-employed individuals and freelancers were excluded by this definition. The age range in the S-MGA includes birth cohorts born between 1951 and 1980. To facilitate face-to-face interviews, a two-stage cluster sampling procedure was applied with a random sample of 206 municipalities in Germany proportionately stratified by region and population, followed by a random gross sample of 13,590 addresses drawn within selected municipalities [31]. Altogether, 4511 individuals participated in a computer-assisted personal interview (CAPI). The follow-up of the data collection took place in 2017. The comparison of the basic population (IEB) and the cohort for the analysis showed only moderate differences [32] in sociodemographic characteristics. Individuals who had retired between the time of sampling and the baseline interview were disregarded for the current analysis (Fig. 1).

All information was obtained in the respondents’ home, mainly through a personal interview (CAPI), whereas sensitive information such as depressive symptoms was gathered through paper questionnaires (PAPI) completed without the interviewer being present.

Independent variables

We categorized depressive symptoms using the Patient Health Questionnaire (PHQ-9) [33] into three categories: ‘medium to severe’ (≥ 10) [34], ‘mild’ (5–9), and ‘none’ (0–4). The PHQ-9 yields a sum score based on the following questions [35]: ‘Over the last 2 weeks, how often have you been bothered by any of the following problems?` The nine items in the scale are ‘Little interest or pleasure in doing things’, ‘Feeling down, depressed or hopeless’ ‘Difficulty falling asleep or sleeping or increased sleep’, ‘Tiredness or feeling unable to have energy’, ‘Decreased appetite or excessive need to eat’, ‘Bad opinion of yourself’, ‘Difficulty concentrating on something’, ‘Slowed speech/movement or restlessness (“fidgety’)’, ‘Thoughts that you would rather be dead or want to self-inflict pain’. The response options are ‘Not at all’ (0), ‘Several days’ (1), ‘More than half the days’ (2) and ‘Nearly every day’ (3). The score for depressive symptoms is computed as the sum of all available items. Based on previous analyses performed by authors of this study [36], the distribution of the PHQ values is positively skewed. This means that there are many low values and few high values. The reason for this is that at the time of sampling all individuals were employees subject to social security contributions, who are more likely to be well integrated into the labour market and less limited in functioning than it would be the case for individuals participating in a clinical trial.

Outcomes

At follow-up, individuals were asked about their employment biography [37] since the baseline assessment in 2011/12. They reported the month and year of the start of a) early retirement or disability pension, b) unemployment and c) long-term sickness absence (≥ 6 weeks) from work and their duration during follow-up. In the follow-up, we censored all observations at the month of statutory retirement age.

Covariates

Covariates were gender, age, skill level and employment status. Due to the limited sample size, we needed to keep the number of parameters in the regression models low and decided not to include any interactions by gender. Instead, we used gender as stratification variable [38, 39], also based on previous studies (see Introduction) suggesting gender-related differences in the examined associations. Age was categorized into three decades. The skill level of the job was based on the International Standard Classification of Occupations (ISCO 08) of the International Labour Office [40] and categorized into four groups: unskilled workers (ISCO groups 9), skilled workers (4–8), semiprofessionals (3), and professionals (1,2). The most frequent group (‘skilled workers’) was used as a reference group. The status “in paid employment at baseline interview” (acronym NPE) was coded as a dichotomous variable with the categories ‘no’ and ‘yes’.

Statistical analysis

The current analysis is based on two-part models (Min and Agresti 2002; Smyth 1996). In the following, we give reason for this approach. The number of months as an outcome is semicontinuous data. The number of months takes exclusively nonnegative integer values, and it is zero-inflated; i.e., a large stack of data points assemble at 0 as the lower bound. The distribution is a mixture of 0 und nonzero values and a positively skewed distribution of nonzero values. This type of data places special demands on the statistics.

Transformation techniques cannot produce a normal distribution. This must also be taken into account when using regression techniques. A common regression analysis or multiple regression analysis is bound to assumptions such as the normal distribution of regression residuals. This assumption is not met for the number of months. Linear regression would result in an inadequate model and biased estimates of confidence intervals and p values. Moreover, homoskedasticity is an important prerequisite for linear modelling, which requires equal variances in the target value for the different values of the predictor. This is not the case with the number of months as the data: higher mean values and thus higher variances are present for increasing values of predictor values.

The likelihood function $l$ of a two-part model [41, 42] with a variance parameter σ has the form:

$$l\left({{\beta }}_{1},{{\beta }}_{2},\sigma \right)=\left[\prod _{\left\{i:{y}_{i}=0\right\}}P({y}_{i}=0|{{x}}_{1,i}^{ },{{\beta }}_{1})\right]\left[\prod _{\left\{i:{y}_{i>0}\right\}}P({y}_{i}>0|{{x}}_{1,i}^{ },{{\beta }}_{1}\left)f\right({y}_{i}|{{x}}_{2,i}^{ },{{\beta }}_{2},\sigma )\right]$$

${{\beta }}_{1}$ and ${{\beta }}_{2}$ are two sets of regression parameters to be estimated in the model. $f$ is a density function specified in the concrete case. This likelihood function has only one global maximum; it can be decomposed into two parts, and each part can be maximized separately. First, the logistic regression part with parameters ${{\beta }}_{1}$ is used to model zero and nonzero values. Second, conditional on all nonzero values, generalized linear models (GLMs) with a gamma distribution and a log-link function are used to model the nonzero values with parameters ${{\beta }}_{2}$.

In other words, logistic regression is used to model the event-related data. Odds ratios estimate the risk for one or more long-term sickness at paid employment, unemployment and exit from paid employment by early retirement between baseline assessment and follow-up. Conditional on all nonzero events, the regression analysis of the (total) amount of nonparticipation aims at the number of months expected for these events. The statistical effect of a generalized linear model with a logarithmic link function is given by exp(β), which is similar to an OR derived from a logistic regression. Exp(β) and OR operate as multiplier. The predictive value is calculated by multiplying the exponential function of the intercept (exp(β₀)) by the exponential functions of the regression coefficients. These exp(β) indicate the factor by which the number of months is increased.

The results are stratified by gender and adjusted for covariates as possible confounders. The statistical analysis is validated by comparing the GLM expected values for males and females within three categories of the PHQ with the observed values of these groups, i.e., the number of months lost due to these events. Persons who have no long-term absences of at least one month are not taken into account when estimating the monthly figures. All statistical analyses were performed using IBM® SPSS® Statistics version 27.01.0.

Description of the analysed population:

There were 4511 employees who participated at baseline (Fig. 1). Although all individuals were in employment at the time of sampling, there was a time gap of an average of 13 months between the sampling date and the interview date in which participants could change their employment status. For this reason, 66 participants who had retired in the meantime were excluded from the analysis. In addition, 500 participants from baseline were excluded because their information on the PHQ questionnaire was missing or because they could not be assigned to an ISCO code. For the follow-up study, the largest loss was due to nonparticipation of 1509 respondents from the initial survey and 23 participants in the follow-up for whom an employment biography could not be established. The resulting sample for the analysis consists of 2,413 individuals (n = 1168 males and n = 1245 females). Tables 1 and 2 reports the sample characteristics for both strata, and Fig. 1 displays the flow chart. The total follow-up time was 73079 months for men and 77,913 months for women; the mean follow-up time was 63 months for both genders.

(Fig. 1 about here)

Table 1

Sample characteristics regarding the independent variable and covariates
		Males (n = 1168)		Females (n = 1245)
		N	%	N	%
Age at baseline
	31–40	278	24	300	24
	41–50	487	42	518	42
	51–60	403	34	427	34
Skill level at baseline
	Unskilled (9)	47	4	93	8
	Skilled (4–8)	542	46	456	37
	Semiprofessional (3)	253	22	421	34
	Professional (1,2)	326	28	275	22
(Not) in paid employment at baseline interview ^a
	No	38	3	73	6
	Yes	1130	97	1172	94
Depressive symptoms at baseline
	None (< 5)	777	67	693	56
	Mild (5–9)	320	27	432	35
	Medium to severe (≥ 10)	71	6	120	10
^a, e.g., parenthood, qualification, unemployment at baseline

Table 2

Sample characteristics regarding the dependent variables
		Males (n = 1168)		Females (n = 1245)
		N	%	N	%
Follow-up months total	73079		100	77913	100
Absence > 4 weeks during follow-up, fraction of all months
Total work nonparticipation (TWNP)	5080		7	5738	7
Long term sickness absence (LTSA)	1666		2	2186	3
Unemployment (UE)	1574		2	1530	2
Early retirement (ER)	1709		2	1707	2
Median TWNP, months ^a (by subject)	6			6
Mean TWNP, months ^a (by subject)	13			13

^a Among those experiencing at least one event of work nonparticipation.

Depressive symptoms and work nonparticipation events

Tables 3 and 4 show, for males and females respectively, the crude odds ratios, the odds ratios adjusted for birth cohort, skill level, unemployment at baseline and the corresponding 95% confidence levels for the point estimates. The odds for work nonparticipation (total work nonparticipation; TWNP) were elevated by an OR = 3.22 (95% CI: 1.90– 5.50) for males with medium to severe depressive symptoms. In addition, males with mild symptoms had 1.59 times higher odds (95% CI: 1.19–2.11) than the reference group. Unemployment had the strongest association with medium to severe depressive symptoms among males, with an OR of 5.17 (95% CI 2.58–10.36). Medium to severe symptoms had no significantly elevated effect on early retirement of males, whereas mild depressive symptoms had an effect on early retirement (OR = 1.72; 95% CI: 1.01–2.92). Both mild depressive symptoms (OR = 1.48; 95% CI 1.08–2.02) and medium to severe depressive symptoms (OR = 2.96; 95% CI: 1.76–4.97) had an effect on long-term sickness absence (LTSA) from work. The strongest effect for adjustment by possible confounders was found with the association between early retirement and medium to severe depressive symptoms. Introducing covariates such as birth cohort diminished the unadjusted OR = 1.72 towards OR = 1.20 (95% CI = .47–3.06).

Medium to severe depressive symptoms (OR = 1.92; 95% CI: 1.29–2.87) had weaker effects on TWNP among females than among males. The strongest effect among females was shown with LTSA as an outcome (OR = 2.57; 95% CI: 1.69–3.91). Among females, there was no significant effect of depressive symptoms on unemployment or early retirement.

Table 3

Association between baseline depressive symptoms and work nonparticipation during follow-up among males (n = 1168)
Outcome	Model 1^a					Model 2 ^b
	Depressive symptoms					Depressive symptoms
	No (N = 777)	Mild (5–9) (n = 320)		Medium to severe (≥ 10) (n = 71)		No (N = 777)	Mild (5–9) (n = 320)		Medium to severe (≥ 10) (n = 71)
	OR	OR ^a	95% CI	OR ^a	95% CI	OR	OR ^b	95% CI	OR ^b	95% CI
TWNP	1	1.59	1.21–2.09	3.69	2.24–6.07	1	1.59	1.19–2.11	3.22	1.90–5.45
LTSA	1	1.49	1.10–2.02	3.06	1.85–5.06	1	1.48	1.08–2.02	2.96	1.76–4.97
UE	1	1.79	1.09–2.93	6.10	3.28–11.34	1	1.67	0.98–2.87	5.17	2.58–10.36
ER	1	1.64	1.02–2.66	1.74	.75–4.01	1	1.72	1.01–2.92	1.20	.47–3.06

TWNP: Total work nonparticipation; LTSA: Long-term sickness absence; UE: Unemployment; ER: Early retirement.

^a Crude associations.

^b Adjusted for age, ISCO-skill level, participation at work at baseline.

Table 4

Association between baseline depressive symptoms and work nonparticipation during follow-up among females (n = 1245)
Outcome	Model 1 ^a					Model 2 ^b
	Depressive symptoms					Depressive symptoms
	No (N = 690)	Mild (5–9) (n = 435)		Medium to severe (≥ 10) (n = 120)		No (N = 690)	Mild (5–9) (n = 435)		Medium to severe (≥ 10) (n = 120)
	OR	OR ^a	95% CI	OR ^a	95% CI	OR	OR ^b	95% CI	OR ^b	95% CI
TWNP	1	1.42	1.10–1.82	1.95	1.32–2.88	1	1.42	1.10–1.84	1.92	1.29–2.87
LTSA	1	1.59	1.23–2.11	2.52	1.67–3.80	1	1.56	1.18–2.07	2.57	1.69–3.91
UE	1	1.29	.81–1.87	1.24	.64–2.39	1	1.31	.85–2.02	1.13	.57–2.25
ER	1	1.09	.68–1.74	1.01	.46–2.19	1	1.13	.68–1.89	1.05	.46–2.40

TWNP: Total work nonparticipation; LTSA: Long-term sickness absence; UE: Unemployment; ER: Early Retirement.

^a Crude associations.

^b Adjusted for age, ISCO-skill level, participation at work at baseline.

Depressive symptoms and number of months with work nonparticipation

Tables 5 and 6 shows the factor with which mild depressive symptoms and medium to severe depressive symptoms increase duration of work nonparticipation if the number of months given in the first column (constant) is used as the starting value. Regarding TWNP, the effect of mild depressive symptoms was significant for males [EXP(B) = 1.4; 95% CI: 1.15–1.70] but not for females. The duration of LTSA was extended for both males and females in both depressive symptoms groups. The duration of TWNP was doubled for males with medium to severe depressive symptoms [EXP(B) = 2.06; 95% CI: 1.53–2.78] and more than a third higher [EXP(B) = 1.38; 95% CI: 1.05–1.83] for females. For females, the medium to severe depressive symptoms did not significantly change the duration of early retirement and unemployment. For males, the number of months with early retirement was not affected. However, unemployment events were longer when medium to severe symptoms were present.

Table 5

Baseline depressive symptoms and months of work nonparticipation during follow-up ^a. N = 384 males.
		Depressive symptoms
		Mild (5–9)		Medium to severe (≥ 10)
Outcome	const.	EXP B ^b	95% CI	EXP B ^b	95% CI	N
TWNP	5.39	1.40	1.15–1.70	2.06	1.53–2.78	384
LTSA	5.32	1.24	1.02–1.51	1.78	1.34–2.37	264
UE	6.27	1.51	.98–2.34	1.94	1.14–3.29	88
ER	12.82	1.20	.82–1.77	1.31	.62–2.80	83
TWNP: Total work nonparticipation; LTSA: Long-term sickness absence; UE: Unemployment; ER: Early retirement;
^a among males experiencing work nonparticipation. The results of generalized linear models (GLMs)
^b adjusted for age, ISCO-skill level, participation at work at baseline

Table 6

Baseline depressive symptoms and months of work nonparticipation during follow-up ^a. N = 458 females
		Depressive symptoms
		Mild (5–9)		Medium to severe (≥ 10)
Outcome	const.	EXP B ^b	95% CI	EXP B ^b	95% CI	N
TWNP	7.19	1.13	.94–1.37	1.38	1.05–1.83	458
LTSA	5.85	1.30	1.07–1.57	1.49	1.15–1.94	313
UE	6.20	1.39	.93–2.08	1.50	.76–2.97	112
ER	45.18	.76	.50–1.16	1.47	.72–3.02	85
TWNP: Total work nonparticipation; LTSA: Long-term sickness absence; UE: Unemployment; ER: Early retirement;
^a among females experiencing work nonparticipation. The results of generalized linear models (GLMs)
^b adjusted for age, ISCO-skill level, participation at work at baseline

Validation

We validated the regression results in Tables 5 and 6 by comparing the expected and observed durations of (total) work nonparticipation (Table 7). The GLM parameters in Tables 5 and 6 were used to calculate the expected months for each PHQ category. These GLMs estimated 5676 months of work nonparticipation for the total sample of males and 7662 months for females. The actual numbers of months of nonparticipation were 5080 months for females and 5738 months for females and thus lower than the estimates. These estimated deviations were larger for females than for males and larger for those with medium to severe symptoms compared to the others.

Table 7

Total work nonparticipation (TWNP): Expected months vs. observed months
Males
Depressive symptoms	Cases of work nonparticipation, n	Expected months of work nonparticipation for each case ^a	Expected sum of months of work nonparticipation (observed n*expected months for each case)	Observed sum of months of work nonparticipation
None (< 5)	219	11.89	2604	2390
Mild (5–9)	123	16.60	2042	1817
Medium to severe (≥ 10)	42	24.52	1030	873
Total	$\sum$384		∑5676	∑5080
Females
Depressive symptoms	Cases of work non participation, n	Expected months of work nonparticipation for each case ^a	Expected sum of months of work nonparticipation (observed n*expected months for each case)	Observed sum of months of work nonparticipation
None (< 5)	225	15.22	3425	2593
Mild (5–9)	175	17,24	3017	2276
Medium to severe (≥ 10)	58	21.03	1220	869
Total	$\sum$458		$\sum$7662	∑5738

^a Based on generalized linear models (GLMs) on TWNP, both adjusted for birth cohort, skill level, and participation at work at baseline (Table 3).

The results indicate that depressive symptoms had an effect on work nonparticipation in general and on long-term illness in particular among both genders and on unemployment among males. Regarding medium to severe depressive symptoms, the effects were more pronounced for males than for females. The effects of medium to severe depressive symptoms on unemployment and early retirement were not significant among females, although the magnitude of the coefficients suggests a potential impact. A notable finding is that, after adjusting for the confounders, the risk of early retirement falls sharply among males with moderate to severe depressive symptomatology, presumably because of the small number of participants. Older males who came close to the age limit for retirement at the time of follow-up may have already retired before. This would lead to a selection of males among the remaining employees, who would then no longer present a higher risk of early retirement due to depressive symptoms. Overall, early retirement shows the smallest effects or no effects compared to the other outcomes and very wide confidence intervals for the observed estimates. The results, stratified by gender, show that depressive symptoms did not affect long-term absences in the same way for males and females, but that for females, the effects on long-term illnesses – in existing employment – dominated in particular, while for males, the strongest effects were to be found in unemployment. A look at prevalence values of depression [9] initially pointed to higher prevalence in females and, at first glance, conveyed a stronger concern for females. The present analyses paint a different picture: both for the event-related analyses and for the prediction of the duration, stronger effects were found among males. Reports from pension insurance in Germany [2] suggest a connection between depressive symptoms and early retirement. The dose-response relationship found between depressive symptoms and duration of early retirement supports such an association. However, the latter was not observed in the analysis on events of early retirement. Further investigations in individual occupational groups could help clarify this.

Strengths

The methodology of the present study, the data collection and the validation procedure contribute to the validity of the findings and the possible conclusions. The analysis takes advantage of the longitudinal study design: depressive symptoms was recorded at baseline, and the employment biography was followed up the subsequent five years. For an aetiological interpretation of the associations, the temporal order of the exposure and outcome variables was established.

The prevalence of depressive symptoms was captured in a representative sample. The comparison of sample characteristics with the population indicated no sign of selection bias [31] at baseline. Additional strengths result from the choice of the PHQ as a survey instrument for depressive symptoms. The instrument is well established as a tested procedure both in Germany and in international representative studies. With a cut-off value of > 9 [34], it offers a case definition, and in addition, the sum score offers a quantitative measure of depressive symptoms. From the perspective of prevention, it is important to recognize depressive symptoms before the manifestation of a disorder. Therefore, we included the category of mild depressive symptoms. Our results show that mild depressive symptoms can already be used for the prediction of TWNP and LTSA.

The query of the employment biography with events of unemployment and early retirement for the five years after the baseline survey was carried out with a resolution of one month. Sick leave events were recorded separately. If persons are only asked to recall the events themselves during a follow-up, they are likely to have greater recall errors, whereas this procedure records events more validly [43]. It allows not only the recording of an event but also the quantification of the total duration of this type of event. It is therefore a starting point for quantifying costs due to depressive symptoms, which is beyond the scope of this article. A special feature of the present analyses is the validation of the quantification by comparing the predicted number of months of nonparticipation with the number of months actually recorded. The results not only quantify the prediction errors but also provide particular evidence of gender-specific prediction differences. This may also be due to gender-specific response behaviours in a survey on depressive symptoms, in which males tend to be more reluctant to describe depressive symptoms. We believe that the validation provides further support to the quality of our predictions.

Weaknesses

First, participation in the cohort is low (18%). Based on comparisons with the sampling frame of the study, participation in the cohort was independent of gender, while it was associated with age (20% among those aged 55–60, 15% among those aged 31–36) and skill level (24% among professionals, managers and 14% among unskilled workers) [32]. We believe that differences in attrition due to skill level and age do not introduce major biases in our risk estimates, considering that differences were relatively small and that age and skill level were adjusted for in all analyses.

Second, a weakness of the analyses might be that the study addresses only longer work nonparticipation events, as the study deals with events lasting months, not weeks or days. Additionally, only sickness absences of at least 6 weeks were asked about. According to reports by the Federal Ministry of Health, more than half of all recorded sick days are from events with a duration < 42 days. The present study only addressed events lasting one month (regarding sickness absence, at least 6 weeks), which illustrates a strong underrecording of the time volume. Furthermore, we do not know whether depressive symptoms have the same effect on short-term absences as with LTSA. The German insurance system comprises a large number of health insurance companies, making use of their data difficult. This means that a check using national register data with events of different lengths is not possible.

Comparison with other studies

A comparison of our results with those obtained in previous research is difficult due to the strong heterogeneity of earlier studies. This is not only due to the different study contexts, but also due to the use of different scales to assess depressive symptoms and the selection of indicators of work nonparticipation. For example, in the cited studies from Denmark, two different instruments were used to assess depressive symptoms [44, 45]. The most comprehensive selection of indicators of work nonparticipation was offered in Pedersen's study [25], which also included a large-size sample of 11967 participants. In line with the latter study from Denmark, our study suggests that depressive symptoms have an effect on total work nonparticipation. The Danish study found a consistent shorter work life expectancy among people with depression and an equivalent longer time in nonwork states.

The effects of depressive symptoms on disability pensions that were shown in the studies by Bültmann [18] and Porru [19] could not be investigated in the present study because the number of cases with DP was too small in the S-MGA cohort. However, when replacing DP with any form of early retirement – as we did in the present study – we found that age, i.e., the time (or nearness) to statutory retirement, was a much stronger predictor than depressive symptoms. In addition, the different pathways that can lead to termination of employment must be taken into account. The German Ageing Survey (DEAS) and the socioeconomic panel (SOEP) point to unemployment as a pathway into retirement [6, 46]. This effect may also be reflected in the present results, which show strong effects for unemployment, especially among males, and thus might dilute the effects for early retirement.

Based on the available data, the authors of the present study find it difficult to provide a simple explanation for the observed gender differences. One possible explanation could be that males tend to identify more strongly with their work role than females do, while the latter tend to identify more with their family obligations [47]. The results in SHARE [19], such as the present research, seem to show a stronger effect of depressive symptoms on unemployment among males than among females. However, the results of SHARE are based on 11 different European countries, on older people and on a different method of recording unemployment.

Similar to studies in Denmark [17, 25] and Australia [16], the present findings show the effects of depressive symptoms on LTSA and unemployment. The findings show, however, that the attribution of work nonparticipation due to depression is not only the result of clinical and administrative acts [7, 8] within health care; depressive symptoms are actually predictive of subsequent failures within a timeframe of five years. The prediction refers, on the one hand, to the risk for the occurrence of work nonparticipation and, on the other hand, to the number of months used to quantify the volume of default. Both results are relevant for an assessment of the economic consequences of depressive symptoms on participation in working life. However, when assessing the economic consequences of depressive symptoms, it is not helpful to limit the analysis to only one type of work nonparticipation.

The different operationalisations of nonparticipation and the different paths to this state support, in our view, an approach that does not single out aspects such as LTSA or unemployment but compares the different outcomes. The use of two-part models facilitates the prediction of events, provides a method to quantify work nonparticipation and thus provides a bridge to quantify the societal costs of depressive symptoms.

From a practical and clinical point of view, the study indicates that the risk for – and amount of – work nonparticipation can be increased not only for long-term sickness absence but also for male unemployment. Additionally, subclinical levels of depression can pose a risk factor for work nonparticipation. The early detection of possible cases of depression could contribute to higher participation in the labour market. The study does not indicate, at least among females, that depressive symptoms are a major risk factor for early pensioning; however, larger studies are needed to distinguish between various types of pensions prior to statutory pension age.

The present study is one of the first to indicate that depressive symptoms are an important predictor for total work nonparticipation among males and females. Regarding LTSA, unemployment and pension, the results are supported by only a few other studies. Studies quantifying the effects of depressive symptoms on work nonparticipation are largely lacking. Plenty of administrative data on diagnoses of cases of sickness absence and disability pension are available, but we do not know to what extent these diagnoses reflect the overall effects of depressive symptoms and sickness absence and disability pension in populations. We also do not know what role depressive symptoms play for other types of pension or for unemployment. To understand the impacts of depressive symptoms, approaches quantifying effects on participation above calculations of rate ratios (or hazard ratios, odds ratios) are needed. The present study offers such an approach by taking into account not only the prevalence of nonwork events but also the duration of these events.

Ethical approval and consent to participate

Institutional Review Board Statement: This study was conducted according to the guidelines of the Declaration of Helsinki and was approved by the Ethics Committee of the Federal Institute for Occupational Safety and Health (BAuA; approval number 006_2016_Müller, 6 March 2016).

Informed Consent Statement: Informed consent was obtained from all subjects involved in the study. All employees in the sample were contacted by mail, and the interviews at baseline were only conducted after each respondent gave their informed oral consent. Written consent was given for the willingness to participate at follow-up.

Consent for publication

Not applicable

Availability of data and material

A scientific use file (SUF) containing both wave 1 and wave 2 of the cohort is available at the Research Data Centre of the Federal Institute of Occupational Safety and Health. Acknowledgements: The longitudinal Study on Mental Health at Work (S-MGA) was conducted in collaboration with the Institute for Employment Research (IAB). The S-MGA was based on samples from statistics of the Federal Employment Agency (BA), which have been merged with Integrated Employment Biographies by the IAB. A scientific use file of the first wave and second wave data is available at the BAuA (DOI: 10.48697/smga.w1w2.suf.).

Competing interests

The authors declare no conflicts of interest.

Funding

The 2012 and 2017 rounds of the S-MGA study were funded internally by the BAuA (project no. F 2250 and F 2478). This paper was written as part of the project “Mental health and work participation” at the BAuA (internally funded project no. F 2462).

Author’s contributions

Conceptualization, U.R., N.K. and H.B.; methodology, U.R., N.K. and H. B; software, U.R. and N.K.; validation, U.R.; formal analysis, U.R., N.K., and D.P.; resources, U.R. and H.B.; data curation, U.R. and D.P.; writing/original draft preparation, U. R, H.B. D.P. and N.K. project administration, U.R.; funding acquisition, U.R. All authors have read and agreed to the published version of the manuscript.

Acknowledgements

The Study on Mental Health at Work (S-MGA) was conducted in collaboration with the Institute for Employment Research (IAB). The S-MGA was based on samples from statistics of the Federal Employment Agency (BA), which were merged into Integrated Employment Biographies by the IAB.

We thank the participating employees and Infas (Institute of Applied Social Sciences) for collecting the data.

We thank the reviewers for their thoughtful comments. We also thank Springer Nature Author Services and Paul Conway, University of Copenhagen, for the language check.

Knieps F, Pfaff H. BKK Gesundheitsreport 2019. Berlin: MWV Medizinisch Wissenschaftliche Verlagsgesellschaft; 2020.
Deutsche Rentenversicherung Bund. Rentenversicherung in Zeitreihen. Berlin: DRV; 2019.
Bundesgesundheitsministerium. Arbeitsunfähigkeit: Fälle und Tage nach Falldauer 2019 - Ergebnisse der Krankheitsartenstatistik der gesetzlichen Krankenversicherung. In.: BMG; 2019.
Schofield DJ, Shrestha RN, Percival R, Passey ME, Callander EJ, Kelly SJ. The personal and national costs of mental health conditions: impacts on income, taxes, government support payments due to lost labour force participation. BMC Psychiatry. 2011;11(1):72.
Pedersen J, Solovieva S, Thorsen SV, Andersen MF, Bultmann U. Expected Labor Market Affiliation: A New Method Illustrated by Estimating the Impact of Perceived Stress on Time in Work, Sickness Absence and Unemployment of 37,605 Danish Employees.Int J Environ Res Public Health2021, 18(9).
Buchholz S, Rinklake A, Blossfeld H-P. Reversing early retirement in Germany a longitudinal analysis of the effects of recent pension reforms on the timing of the transition to retirement and on pension incomes.Comparative population studies2013, 38(4).
Johansen K, Andersen JS, Mikkelsen S, Lynge E. Decision making and co-operation between stakeholders within the process of sick leave. A case study in a Danish municipality. J Interprof Care. 2011;25(1):59–65.
Maeland S, Werner EL, Rosendal M, Jonsdottir IH, Magnussen LH, Lie SA, Ursin H, Eriksen HR. Sick-leave decisions for patients with severe subjective health complaints presenting in primary care: a cross-sectional study in Norway, Sweden, and Denmark. Scand J Prim Health Care. 2013;31(4):227–34.
Bretschneider J, Janitza S, Jacobi F, Thom J, Hapke U, Kurth T, Maske UE. Time trends in depression prevalence and health-related correlates: results from population-based surveys in Germany 1997–1999 vs. 2009–2012. BMC Psychiatry. 2018;18(1):1–13.
Wittchen HU, Jacobi F, Rehm J, Gustavsson A, Svensson M, Jönsson B, Olesen J, Allgulander C, Alonso J, Faravelli C, et al. The size and burden of mental disorders and other disorders of the brain in Europe 2010. Eur neuropsychopharmacology: J Eur Coll Neuropsychopharmacol. 2011;21(9):655–79.
Lagerveld SE, Bültmann U, Franche RL, van Dijk FJH, Vlasveld MC, van der Feltz-Cornelis CM, Bruinvels DJ, Huijs JJJM, Blonk RWB, van der Klink JJL, et al. Factors associated with work participation and work functioning in depressed workers: A systematic review. J Occup Rehabil. 2010;20(3):275–92.
Wittchen HU. Reliability and validity studies of the WHO–Composite International Diagnostic Interview (CIDI): a critical review. J Psychiatr Res. 1994;28(1):57–84.
Wittchen H-U, Pfister H. DIA-X-interviews: manual für screening-Verfahren und interview; Interviewheft. 1997.
Beck AT, Ward CH, Mendelson M, Mock J, Erbaugh J. An inventory for measuring depression. Arch Gen Psychiatry. 1961;4(6):561–71.
Hautzinger M, Keller F, Kühner C. Beck depressions-inventar (BDI-II). Harcourt Test Services; 2006.
Wooden M, Bubonya M. Cobb-Clark DJSJoW, Environment, Health: Sickness absence and mental health: evidence from a nationally representative longitudinal survey. 2016(3):201–208.
Thorsen SV, Rugulies R, Hjarsbech PU, Bjorner JB. The predictive value of mental health for long-term sickness absence: the Major Depression Inventory (MDI) and the Mental Health Inventory (MHI-5) compared. BMC Med Res Methodol. 2013;13:115–5.
Bültmann U, Rugulies R, Lund T, Christensen KB, Labriola M, Burr H. Depressive symptoms and the risk of long-term sickness absence. Soc Psychiatry Psychiatr Epidemiol. 2006;41(11):875–80.
Porru F, Burdorf A, Robroek SJW. The impact of depressive symptoms on exit from paid employment in Europe: a longitudinal study with 4 years follow-up. Eur J Public Health. 2019;29(1):134–9.
van Rijn RM, Robroek SJ, Brouwer S, Burdorf A. Influence of poor health on exit from paid employment: a systematic review. Occup Environ Med. 2014;71(4):295–301.
Kausto J, Pentti J, Oksanen T, Virta LJ, Virtanen M, Kivimäki M. Vahtera JJSJoW, Environment, Health: Length of sickness absence and sustained return-to-work in mental disorders and musculoskeletal diseases: a cohort study of public sector employees. 2017(4):358–366.
Lexis MAS, Jansen NWH, van Amelsvoort LGPM, van den Brandt PA, Kant I. Depressive Complaints as a Predictor of Sickness Absence Among the Working Population. 2009, 51(8):887–895.
Nieuwenhuijsen K, Verbeek JHAM, de Boer AGEM, Blonk RWB, van Dijk FJH. Predicting the duration of sickness absence for patients with common mental disorders in occupational health care. Scand J Work Environ Health. 2006;32(1):67–74.
Thorsen SV, Pedersen J, Flyvholm M-A, Kristiansen J, Rugulies R, Bültmann U. Perceived stress and sickness absence: a prospective study of 17,795 employees in Denmark. Int Arch Occup Environ Health. 2019;92(6):821–8.
Pedersen J, Thorsen SV, Andersen MF, Hanvold TN, Schlünssen V, Bültmann U. Impact of depressive symptoms on worklife expectancy: a longitudinal study on Danish employees. Occup Environ Med. 2019;76(11):838.
Altersrenten. [https://www.bmas.de/DE/Soziales/Rente-und-Altersvorsorge/Leistungen-Gesetzliche-Rentenversicherung/Altersrenten/altersrenten.html;jsessionid=6F1B83F9E3159FD3D39CC39E861BCF54.delivery1-replication]
Krankengeld. [https://www.bundesgesundheitsministerium.de/krankengeld.html]
Rehabilitation. [https://www.bundesgesundheitsministerium.de/rehabilitation.html]
Sick leave and sickness benefits in Denmark. [https://www.oresunddirekt.se/en/working-in-denmark/working-in-denmark/sick-leave-and-sickness-benefits-in-denmark]
Altersgrenzen, Alterserwerbstätigkeit. [https://www.bpb.de/themen/soziale-lage/rentenpolitik/292781/altersgrenzen-und-alterserwerbstaetigkeit/]
Rose U, Schiel S, Schröder H, Kleudgen M, Tophoven S, Rauch A, Freude G, Müller G. The study on mental health at work: design and sampling. Scand J Public Healt. 2017;45(6):584–94.
d’Errico A, Burr H, Pattloch D, Kersten N, Rose U. Working conditions as risk factors for early exit from work—in a cohort of 2351 employees in Germany. Int Arch Occup Environ Health. 2021;94(1):117–38.
Löwe B, Spitzer RL, Zipfel S, Herzog W. PHQ-D. Gesundheitsfragebogen für Patienten. Vol. 2. Karlsruhe:Pfizer; 2002:pp. 5–7.
Manea L, Gilbody S, McMillan D. A diagnostic meta-analysis of the Patient Health Questionnaire-9 (PHQ-9) algorithm scoring method as a screen for depression. Gen Hosp Psychiatry. 2015;37(1):67–75.
Spitzer RL, Kroenke K, Williams JB, Group PHQPCS, Group PHQPCS. Validation and utility of a self-report version of PRIME-MD: the PHQ primary care study. JAMA. 1999;282(18):1737–44.
Rose U, Müller G, Burr H, Schulz A, Freude G. Arbeit und Mentale Gesundheit. Ergebnisse aus einer Repräsentativerhebung der Erwerbstätigen in Deutschland F2250. In. Dortmund: BAuA; 2016.
Börsch-Supan A, Brandt M, Hunkler C, Kneip T, Korbmacher J, Malter F, Schaan B, Stuck S, Zuber S. Data Resource Profile: The Survey of Health, Ageing and Retirement in Europe (SHARE). Int J Epidemiol. 2013;42(4):992–1001.
Niedhammer I, Saurel-Cubizolles M-J, Piciotti M, Bonenfant S. How is sex considered in recent epidemiological publications on occupational risks? Occup Environ Med. 2000;57(8):521–7.
Smaardijk VR, Maas AH, Lodder P, Kop WJ, Mommersteeg PM. Sex and gender-stratified risks of psychological factors for adverse clinical outcomes in patients with ischemic heart disease: A systematic review and meta-analysis. Int J Cardiol. 2020;302:21–9.
ILO. International Standard Classification of Occupations 2008 (ISCO-08): Structure, group definitions and correspondence tables. Geneva, Switzerland: International Labour Office; 2012.
Min Y, Agresti A. Modeling nonnegative data with clumping at zero: a survey. J Iran Stat Soc. 2002;1(1):7–33.
Smyth GK. Regression analysis of quantity data with exact zeros. In: Proceedings of the second Australia–Japan workshop on stochastic models in engineering, technology and management: 1996: Citeseer; 1996: 572–580.
Wahrendorf M, Marr A, Antoni M, Pesch B, Jöckel K-H, Lunau T, Moebus S, Arendt M, Brüning T, Behrens T, et al. Agreement of Self-Reported and Administrative Data on Employment Histories in a German Cohort Study: A Sequence Analysis. Eur J Popul. 2018;35(2):329–46.
Berwick DM, Murphy JM, Goldman PA, Ware JE Jr, Barsky AJ, Weinstein MC. Performance of a five-item mental health screening test.Medical care1991:169–176.
Bech P, Rasmussen N-A, Olsen LR, Noerholm V, Abildgaard W. The sensitivity and specificity of the Major Depression Inventory, using the Present State Examination as the index of diagnostic validity. J Affect Disord. 2001;66(2–3):159–64.
Engstler H. Erwerbsbeteiligung und der Übergang in den Ruhestand. Der Alterssurvey – Aktuelles auf einen Blick Ausgewählte Ergebnisse 2005.
Wiley MG. Gender, work, and stress: The potential impact of role-identity salience and commitment. Sociol Q. 1991;32(4):495–510.

No competing interests reported.

Download PDF

Journal Publication

published 03 Nov, 2023

Read the published version in BMC Public Health →

Editorial decision: Major revision
24 Apr, 2023
Reviews received at journal
28 Mar, 2023
Reviewers agreed at journal
21 Mar, 2023
Reviewers invited by journal
18 Mar, 2023
Editor assigned by journal
17 Mar, 2023
Editor invited by journal
24 Feb, 2023
Submission checks completed at journal
24 Feb, 2023
First submitted to journal
14 Feb, 2023

You are reading this latest preprint version

Depressive symptoms as a predictor of 5-year subsequent work nonparticipation due to long-term sickness absence, unemployment and early retirement in a cohort of 2,413 employees in Germany

Status:

Journal Publication

Version 1

Abstract

Figures

Background

Methods

Statistical analysis

Results

Discussion

Conclusions

Declarations

References

Additional Declarations

Status:

Journal Publication

Version 1