Study design, participants and data collection

This study was based on cross-sectional data from the Swedish Longitudinal Occupational Health in Healthcare Survey (LOHHCS) [5]. The LOHHCS cohort consists of a representative sample of physicians in Sweden based on the Swedish Occupational Register (as of 2018) [20], managed by the Swedish government agency Statistics Sweden [20]. In 2021, a total of 7200 physicians were invited based on a stratified random sampling from 12 strata, comprising six administrative healthcare regions and two work site conditions (i.e., primary care facility or hospital) in Sweden. In total, 501 participants were originally excluded from the LOHHCS study due to inclusion criteria violation (i.e., no active clinical duty during the past 12 months), resulting in 6699 participants. During February and May 2021, Statistics Sweden collected and processed questionnaire data from 2761 valid respondents (response rate = 41.2%). For details regarding the LOHHCS study population and data collection, please see a previous publication by Hagqvist et al. [5]. In this study, the analytical sample was further restricted by age criteria (i.e., age < 70), removing 127 participants. Additionally, questionnaire data submission failure or missing values in exposure or outcome variables resulted in the exclusion of 602 participants (Fig. 1). In total, the sample of the present study consisted of 2032 participants.

The present study was approved by the Swedish Ethical Review Authority (2020–06613).

Fig. 1MeasurementsJob demand-control-support

Job demands and support were assessed using the core items from the validated Copenhagen Psychosocial Questionnaire III (COPSOQ III) [21], scored on a Likert basis from 1 to 5 (i.e., 1 = Always/to a very large extent, 5 = Never/to a very small extent). The job demands and support items were reversibly coded to reflect high job demands and high support for high scores (i.e., Always/to a very large extent = 5, Never/to a very small extent = 1). In total, job demands were measured with four items (addressing the distribution of workload, time to complete work tasks, work pace, and quantity of work tasks, respectively). Cronbach’s alpha (α) was 0.816 for job demands, indicating high internal consistency. Social support was measured with one item for support from peers, and one item for support from managers, respectively.

To capture healthcare-specific job control, two core items from the COPSOQ III [21] were combined with one validated item regarding professional autonomy [22], and three additional questionnaire items addressing physician’s ability to impact their work. These additional three items were added by the research group for contextual purposes, based on methodological reasoning of the job control aspect for physicians in line with the general J-DCS conceptualization [9]. In total, the following six items addressed job control:

COPSOQ III items [21].

i)

Workplace information regarding decisions, changes, and plans.

ii)

Workplace facilitation of efficient work.

Professional autonomy item [22].

iii)

Workplace facilitation of clinical decision-making.

Additional questionnaire items addressing physician’s ability to:

iv)

Impact the number of patient consultations per day.

v)

Impact the time frame of each consultation.

vi)

Impact the time for administrative work and documentation.

All job control items were answered on a Likert basis from 1 to 5 (i.e., 1 = Always/to a very large extent, 5 = Never/to a very small extent) and reversibly coded (i.e., high scores reflected high job control). Exploratory factor analysis (EFA) was conducted on all items according to guidelines (i.e., retaining factor loadings ≥ 0.40) [23]. The EFA rendered two factors, each including three items of the job control aspect, defined as workplace control (comprising items i-iii above) and task-level control (comprising items iv-vi above). Cronbach’s α was 0.709 and 0.869 for workplace control and task-level control, respectively, indicating sufficient internal consistency for both job control variables. Please see Supplementary Material for a detailed overview of the EFA output, including factor loadings.

Burnout

The Burnout Assessment Tool (BAT), defined by Schaufeli et al. [24], was used to measure burnout risk. The BAT contains 23 items across four categories of symptoms of burnout (i.e., exhaustion, mental distance, emotional impairment, and cognitive impairment) [24]. The BAT provides high reliability and validity and has been psychometrically tested in several countries in close proximity to Sweden [25]. In addition, it provides a cut-off value for high burnout risk based on a score across all symptoms of burnout [26]. All 23 items of BAT were answered on a Likert basis (i.e., 1 = No, never, 5 = Yes, most of the time). Cronbach’s α for all 23 items of BAT was 0.948, indicating high internal consistency. binary outcome variable representing high burnout risk was created i.e., coded as 1 for mean BAT-score ≥ 3.02, indicating a high burnout risk, and 0 for no burnout risk, according to recent recommendations by Schaufeli et al. [27].

Sociodemographic and occupational variables

Sociodemographic variables comprised sex (men and women), age (divided into quartiles and as a continuous measure) and having a partner and/or kids (yes or no). Occupational characteristics comprised the variables sector (public- or private sector), rank (physicians in training (i.e., junior-, intern- and resident physicians or equivalent), specialist physicians and consultants, respectively), worksite (primary care facility, hospital, or other), years of clinical work experience (< 5, 5–10, 11–15, > 15), average working hours per week (< 30, 30–40, 41–50, > 50), and exposure to regular shift work (yes or no). Significant group differences are shown in Table 1.

Table 1 Descriptive statistics and scores of Job demand-control-support and the Burnout Assessment Tool Across Sociodemographic and Occupational Characteristics of Physicians in SwedenData analysis

Descriptive characteristics and scores of exposure variables (i.e., J-DCS) and outcome (i.e., BAT) were calculated and presented across all sociodemographic and occupational groups using appropriate measures of central tendency and dispersion. We evaluated the Pearson’s correlation coefficient (i.e., > 0.7 or < -0.7) to identify possible multicollinearity [28] between the exposure and outcome variables and found no strong evidence of multicollinearity (please see Supplementary Material). To investigate group differences regarding high burnout risk, univariate analysis was performed using the Chi-square test for categorical variables (i.e., sex, partner, kids, sector, rank, working experience, work site, working hours, shift work) and the Student’s t-test for the continuous variable age. The relevant background variables that exhibited statistical significance in the univariate analysis were included in the binary multivariable regression models.

Binary logistic regression models were used to examine the associations between J-DCS (continuous exposure variables) and risk of high burnout (categorical outcome variable). In order to select the background variables, we adopted stepwise logistic regression, which included the following steps. Firstly, the crude analysis included each J-DCS measure in a separate single-variable logistic regression model. Secondly, each model was adjusted for relevant background variables that exhibited significance in the univariate analysis. Thirdly, all J-DCS measures were included into one adjusted multivariable logistic regression model. Lastly, interaction analysis was performed to investigate a potential moderation of the association between job demands and high burnout risk. This was done by introducing interaction terms (not z-standardized) between job demands and each of the job control and support variables (i.e., one at a time) to the adjusted multivariable model (i.e., job demands x workplace control, job demands x peer support, etc.). Similarly, a three-way interaction term between job demands and the significant job control and support variables (i.e., job demands x workplace control x peer support) was added separately to investigate a possible three-way moderation of the association between job demands and high burnout risk. Results from the regression models were presented as odds ratios (OR) and 95% confidence intervals (CI), and all statistical tests were two-sided, where a p-value of ≤ 0.05 was defined as statistically significant. All analyses were performed in STATA version 17.0 (StataCorp, College Station, TX, USA) [29].