Participants and design
We used data from a clinical cohort following patients receiving a 24-week Multi-modal intervention (MMI) for exhaustion disorder at two healthcare clinics in Stockholm, Sweden. Measurements were taken at five time-points: assessment, pre-treatment (week 0), mid-treatment (week 12), post-treatment (week 24) and a follow-up 12 months after completing treatment. To decrease instrumentation bias, the order of the questionnaires was randomized at each dispatch.
All participants were referred from primary health care centers, general practitioners, and occupational health services. Before entering the cohort, all participants were assessed by multi-professional teams (physician, psychologist and physiotherapist). To be included in the cohort, the patients needed to 1) be diagnosed with exhaustion disorder by the multi-professional team, 2) score > 4.5 on the Shirom-Melamed Burnout Questionnaire [26]; 3) be considered suitable for group treatment, 4) be 18–65 years of age, 5) not abuse alcohol or drugs, 6) not participate in any other MMI, 7) not have severe depression, moderate/high risk of suicide or untreated PTSD. Data collection was registered on Clinicaltrials.gov (Identifier: NCT03360136). Furthermore, this study was approved by the Regional Ethical Review Board in Stockholm, Sweden (Approval Nr. 2016/1834-31/2) and followed the ethical principles of the Declaration of Helsinki. All participants provided informed consent before entering the study. For more information on the inclusion process and the contents of the MMI, see [27].
In total, 1643 persons were assessed. Out of these, 472 did not fulfill the criteria for participation, 15 were included in a rehabilitation program for pain, and 73 were offered a short version of the rehabilitation program for SED (12 or 16 weeks). Of 1083 who were offered to participate, 151 declined, 17 withdrew before the start of MII, and 27 withdrew during MII. Further, 8 participants were excluded since they were on disability pension and could not be granted sick leave, leaving 880 participants in the cohort.
Measures
Outcome
Net days of sick leave in the year following treatment were obtained from registers from the Swedish Social Insurance Agency (SSIA), which is the governmental agency that administers reimbursement for sick leave longer than 14 days. SSIA grants both full-time and part-time sick leave (i.e. 0%, 25%, 50%, 75% or 100%). We obtained information on paid sick leave from SSIA six months prior to treatment to one year following treatment for all participants. To calculate net days of sick leave in the year following treatment, we summed all days of sick pay in the year following the first calendar months after treatment completion. Part-time sick days were treated as fractions when summing (e.g. one day on 25% paid sick leave was counted as 0.25 days).
Exposures
Exhaustion symptoms were measured using the Karolinska Exhaustion Disorder Scale (KEDS; (53). KEDS consists of nine items, covering the Swedish diagnostic criteria for exhaustion disorder, and is rated on a 7-point scale from 0 to 6. The total score is calculated by summing all items giving a score ranging from 0 to 54. KEDS has shown satisfactory internal consistency in earlier evaluations [28] and Cronbach’s alpha for KEDS in the current sample was 0.75.
Insomnia symptoms were measured using the Insomnia Severity Index (ISI), consisting of seven items rated on a five-point numerical scale from 0 to 4, and the items are summed to give a total score ranging from 0 to 28. The ISI has demonstrated adequate internal consistency and sensitivity to changes in insomnia severity [29]. Cronbach’s alpha in the current sample was 0.85.
Psychological flexibility was measured using the Swedish Acceptance and Action Questionnaire (SAAQ-II), consisting of six items rated on a seven-point numerical scale from 1 (Never true) to 7 (Always true). The items are summed to give a total score ranging from 6 to 42, where a low total score signifies a high degree of psychological flexibility. The SAAQ-II has shown adequate convergent validity and internal consistency in psychometric evaluations [30], and Cronbach’s alpha in the current sample was 0.86.
Perfectionistic concerns and perfectionistic strivings were measured using the Clinical Perfectionism Questionnaire (CPQ). The CPQ consists of 12 items rated on a numerical scale from 1 (Never) to 4 (Always). Factor analytic studies have indicated two dimensions of the CPQ, perfectionistic concerns (item 1, 3, 6, 10 and 11) and perfectionistic strivings (items 4,5,7,9 and 12) [31]. We summed the items for each dimension respectively, giving subscale scores ranging from 4 to 20. Cronbach’s alpha was 0.79 for the perfectionistic concerns’ subscale and 0.66 for the perfectionistic strivings' subscale in the current sample.
Self-perceived work ability was measured with the following item: "Let us assume that your work ability when it was at its peak, was rated 10. What number would you give your current work ability?". This item is rated on an 11-point scale ranging from 0 (cannot work at all) to 10 (my work ability at its peak). This question was taken from the Work Ability Index (WAI) and has been shown to correlate well with the entire scale and is considered a reliable measure to follow the development of a person’s work ability [32].
Covariates
Covariates were chosen based on prior knowledge of predictors of sick leave in common mental disorders [33, 34] and we included any factor assumed to potentially cause both exposure and the outcome to control for confounding. All covariates were measured pre-treatment to ensure no covariate was on the causal pathway from the exposures to the outcome.
The covariates included: gender (male, female); age (18–29, 30–41, 42–53, 54–65); civil status (Single, Married/living with partner, Partner not living together); yearly household income in SEK (0-250k, 250k-500k, 500k-1000k, > 1000k); education level (elementary or secondary school, university < 3 years, university > 3 years, other); country of birth (Sweden, Nordic countries, Europe, Outside Europe); type of work (handling of heavy loads, heavy repetitive work, medium-heavy work, light repetitive work, administrative/ computer work); comorbidity (none, psychiatric condition, pain condition, psychiatric and pain conditions); unemployment (yes/no); net days of sick leave six months before treatment (continuous); anxiety and depression symptoms (measured with the Hospital Anxiety and Depression scale treated as two continuous subscales) [35].
Statistical analyses
Baseline characteristics of the sample are presented for the full sample and by quartiles of KEDS at post-treatment (Q1 = 0–17, Q2 = 18–22, Q3 = 23–28, Q4 = 29–51) in Table 1. Means and SDs of the exposure levels at pre- and post-treatment is presented in Table 2. The point-wise amount of net sick leave grated by the SSIA to the participants at each follow-up time-point is presented in Fig. 1.
Table 1
Pre-treatment characteristics for the full sample, and by quartiles of KEDS score at post-treatment
|
|
KEDS quartiles at post-treatment
|
|
Full sample
(n = 880)
|
1st
(n = 230)
|
2nd
(n = 210)
|
3rd
(n = 211)
|
4th
(n = 202)
|
Gender, n (%)
|
|
|
|
|
|
Male
|
122 (13.9)
|
41 (17.8)
|
23 (11.0)
|
27 (12.8)
|
29 (14.4)
|
Age, n (%)
|
|
|
|
|
|
18–29 years
|
70 (8.0)
|
21 (9.1)
|
21 (10.0)
|
14 (6.6)
|
11 (5.4)
|
30–41 years
|
326 (37.0)
|
85 (37.0)
|
76 (36.2)
|
78 (37.0)
|
77 (38.1)
|
42–53 years
|
354 (40.2)
|
87 (37.8)
|
80 (38.1)
|
93 (44.1)
|
83 (41.1)
|
54–65 years
|
130 (14.8)
|
37 (16.1)
|
33 (15.7)
|
26 (12.3)
|
31 (15.3)
|
Civil status, n (%)
|
|
|
|
|
|
Married/living together
|
557 (63.3)
|
144 (62.6)
|
142 (67.6)
|
133 (63.0)
|
118 (58.4)
|
Partner living apart
|
58 (6.6)
|
14 (6.1)
|
12 (5.7)
|
10 (4.7)
|
20 (9.9)
|
Single/Other
|
265 (30.1)
|
72 (31.3)
|
56 (26.7)
|
68 (32.2)
|
64 (31.7)
|
Yearly household income, n (%)
|
|
|
|
|
|
0–250k SEK
|
73 (8.3)
|
16 (7.0)
|
12 (5.7)
|
18 (8.5)
|
26 (12.9)
|
500k − 1000k SEK
|
668 (75.9)
|
176 (76.5)
|
156 (74.3)
|
160 (75.8)
|
155 (76.7)
|
>1000k SEK
|
139 (15.8)
|
38 (16.5)
|
42 (20.0)
|
33 (15.6)
|
21 (10.4)
|
Education level, n (%)
|
|
|
|
|
|
Elementary or secondary school
|
219 (24.9)
|
55 (23.9)
|
49 (23.3)
|
53 (25.1)
|
56 (27.7)
|
University less than 3 years
|
140 (15.9)
|
39 (17.0)
|
36 (17.1)
|
26 (12.3)
|
34 (16.8)
|
University 3 years or more
|
482 (54.8)
|
124 (53.9)
|
118 (56.2)
|
121 (57.3)
|
105 (52.0)
|
Other
|
39 (4.4)
|
12 (5.2)
|
7 (3.3)
|
11 (5.2)
|
7 (3.5)
|
Country of birth, n (%)
|
|
|
|
|
|
Sweden
|
762 (86.6)
|
204 (88.7)
|
185 (88.1)
|
179 (84.8)
|
171 (84.7)
|
The north of Europe
|
23 (2.6)
|
7 (3.0)
|
3 (1.4)
|
6 (2.8)
|
7 (3.5)
|
Europe
|
34 (3.9)
|
4 (1.7)
|
12 (5.7)
|
10 (4.7)
|
7 (3.5)
|
Other country
|
61 (6.9)
|
15 (6.5)
|
10 (4.8)
|
16 (7.6)
|
17 (8.4)
|
Type of work
|
|
|
|
|
|
Handling of heavy loads
|
19 (2.3)
|
6 (2.9)
|
3 (1.5)
|
3 (1.5)
|
7 (3.8)
|
Heavy repetitive work
|
40 (4.9)
|
6 (2.9)
|
10 (5.0)
|
11 (5.6)
|
13 (7.1)
|
Medium-heavy work
|
144 (17.7)
|
32 (15.4)
|
35 (17.5)
|
33 (16.7)
|
41 (22.3)
|
Light repetitive work
|
49 (6.0)
|
13 (6.2)
|
11 (5.5)
|
17 (8.6)
|
7 (3.8)
|
Administrative/ computer work
|
562 (69.0)
|
151 (72.6)
|
141 (70.5)
|
134 (67.7)
|
116 (63.0)
|
Comorbidity
|
|
|
|
|
|
No comorbidity
|
528 (60.0)
|
144 (62.6)
|
122 (58.1)
|
140 (66.4)
|
104 (51.5)
|
Comorbid psychiatric disorder
|
304 (34.5)
|
74 (32.2)
|
75 (35.7)
|
60 (28.4)
|
87 (43.1)
|
Comorbid pain disorder
|
28 (3.2)
|
8 (3.5)
|
8 (3.8)
|
6 (2.8)
|
5 (2.5)
|
Comorbid pain and psychiatric disorder
|
20 (2.3)
|
4 (1.7)
|
5 (2.4)
|
5 (2.4)
|
6 (3.0)
|
Employed
|
786 (93.2)
|
208 (92.9)
|
191 (95.5)
|
185 (93.0)
|
178 (91.3)
|
Net days of sick leave 6 months prior to treatment, mean (SD)
|
91.53 (56.79)
|
81.93 (50.56)
|
88.10 (55.96)
|
96.04 (57.46)
|
98.33 (60.80)
|
Depression symptoms, mean (SD)
|
11.23 (3.68)
|
10.20 (3.52)
|
11.05 (3.48)
|
11.65 (3.75)
|
12.23 (3.67)
|
Anxiety symptoms, mean (SD)
|
11.59 (4.01)
|
10.86 (3.89)
|
11.22 (3.77)
|
11.96 (3.82)
|
12.36 (4.36)
|
Table 2
Pre- and post-treatment values of the exposures
Predictor
|
Pre-treatment
|
Post-treatment,
|
|
No.
|
Mean (SD)
|
No.
|
Mean (SD)
|
Exhaustion symptoms, mean (SD)
|
879
|
34.8 (6.2)
|
853
|
22.7 (8.1)
|
Insomnia symptoms, mean (SD)
|
880
|
16.0 (6.1)
|
853
|
9.8 (5.8)
|
Psychological flexibility, mean (SD)
|
872
|
22.5 (6.9)
|
852
|
17.8 (6.8)
|
Perfectionistic concerns, mean (SD)
|
876
|
14.5 (3.5)
|
855
|
10.9 (3.5)
|
Perfectionistic strivings, mean (SD)
|
876
|
13.2 (3.2)
|
855
|
10.5 (3.0)
|
Self-perceived work ability, mean (SD)
|
877
|
2.6 (2.0)
|
590
|
5.3 (2.1)
|
The association between changes in the exposure variables during treatment and the mean number of sick days in the year following treatment was estimated using Generalized linear models (GLM’s) with a log link function and a Gaussian error distribution. Separate models were built for each exposure. To avoid zero-values in the outcome (to enable the use of the log-link), a constant of 0.0001 was added to the net sick days values of 0.
To model the association between changes in the exposure variables during treatment and the subsequent number of net sick days, we estimated the association between the post-treatment levels of the exposure and the outcome conditional on pre-treatment levels of the exposure [36]. Separate models were fitted for each exposure taking the form:
Where Y denotes the net days of sick leave in the year following treatment. Xpost denotes the exposure variable at post-treatment, Xpre is the exposure variable at pre-treatment, and Cpre is a set of pre-treatment covariates. βXpost represents the association between change in the exposure from pre-treatment to post-treatment and net days of sick leave.
We fitted three models for each exposure, using different sets of pre-treatment covariates. Model 1 included only the exposure of interest measured at pre- and post-treatment. In Model 2 we further controlled for gender, age, civil status, yearly household income in SEK, education level, country of birth, type of work, comorbidity, and unemployment. In Model 3, we further controlled for number of days on sick leave six months prior to treatment, anxiety and depression symptoms at pre-treatment and pre-treatment levels of all exposures.
Table 3 presents the rate ratio (RR) of estimated mean sick days corresponding to a one standard deviation decrease in the exposure variable during treatment, calculated as \({{e}}^{- {{\beta }}_{{X}{p}{o}{s}{t}} {S}{D}\left({X}{p}{r}{e}\right)}\). Estimates are presented for each exposure and for Model 1–3 along with 95% confidence intervals (CI’s).
Table 3
Associations between changes in the exposures during treatment and mean net days of sick leave in the year following treatment
|
n
|
RR a
(95% CI)
|
E-value PE
|
E-value UCL
|
|
Exhaustion symptoms
|
|
Model 1b
Model 2c
Model 3d
|
852
754
742
|
0.65 (0.61; 0.69)
0.70 (0.65; 0.75)
0.70 (0.66; 0.75)
|
2.45
2.21
2.21
|
2.26
2.00
2.00
|
|
Insomnia symptoms
|
|
Model 1b
Model 2c
Model 3d
|
853
751
740
|
0.68 (0.60; 0.77)
0.81 (0.72; 0.91)
0.75 (0.67; 0.84)
|
2.30
1.77
2.00
|
1.92
1.43
1.67
|
|
Psychological flexibility
|
|
Model 1b
Model 2c
Model 3d
|
845
746
739
|
0.69 (0.59; 0.82)
0.74 (0.65; 0.83)
0.73 (0.65; 0.83)
|
2.26
2.04
2.08
|
1.74
1.70
1.70
|
|
Perfectionistic concerns
|
|
Model 1b
Model 2c
Model 3d
|
851
752
743
|
0.79 (0.70; 0.88)
0.76 (0.68; 0.84)
0.78 (0.71; 0.86)
|
1.85
1.96
1.88
|
1.53
1.67
1.60
|
|
Perfectionistic strivings
|
|
Model 1b
Model 2c
Model 3d
|
851
752
743
|
0.76 (0.67; 0.87)
0.77 (0.69; 0.86)
0.77 (0.69; 0.86)
|
1.96
1.92
1.92
|
1.56
1.60
1.60
|
|
Self-perceived work ability
|
|
Model 1b
Model 2c
Model 3d
|
588
533
526
|
0.57 (0.51; 0.64)
0.56 (0.50; 0.62)
0.56 (0.50; 0.63)
|
2.90
2.97
2.97
|
2.50
2.61
2.55
|
|
a calculated as \({e}^{- {\beta }_{Xpost} SD\left(Xpre\right)}\) and interpreted as rate ratio of mean net days of sick leave corresponding to a 1 SD decrease in the exposure variable during treatment. Except for work ability that was reversed and calculated as
\({e}^{ {\beta }_{Xpost} SD\left(Xpre\right)}\).
b Adjusted for the respective exposure variable at pre-treatment
c Further adjusted for pre-treatment levels of gender, age, civil status, yearly income, education level, country of birth, type of work, psychiatric comorbidity, and unemployment.
d Further adjusted for pre-treatment levels of all exposure variables.
PE: point-estimate; RR: rate ratio; UCL; upper confidence limit.
|
The assumption of linearity between the exposure and the mean of log sick days was assessed by plotting the residuals against the predicted values along with a loess function, showing fairly linear associations between most exposures and log sick days. One exception was the association between perfectionistic concerns and log sick days that showed slight signs of non-linearity, in this case, the estimate should be interpreted as the best linear approximation. Robust standard errors were used to account for the heteroskedasticity of the residuals. The assumption of normally distributed residuals was not evaluated since this assumption has a minor impact on the estimation of coefficients [37]. However, to assess our results’ sensitivity to the error distribution specification, we compared our main results to models using quasi-poisson error distributions. In this sensitivity analysis, the net days of the sick leave variable were rounded to the nearest full day. Estimates were very similar to those from the main models (that used gaussian error distributions). The estimates were within 0.04 on RR scale (using the transformation shown above), except for the association between work ability and sick days, where point estimates were 0.1–0.2 larger, indicating stronger associations compared to the main models.
There were no missing data on the outcome variable but some missing data on the exposures and covariates. All analyses were conducted as complete case analyses, the number of available observations for each model is presented in Table 3.
To assess the robustness of the results to unmeasured confounding, we also calculated the E-value for the point estimates, as well as for the confidence limits closest to the null. The E-value measures how much an unmeasured confounder would minimally have to affect both the exposure and the outcome on the risk ratio scale, to reduce the point estimate to the null, or in the case of the confidence limit, shift the confidence interval to include the null [38]. E-values are presented in Table 2.
Analyses were performed using R version 4.1.3.