Setting and participants
Participants were recruited from five inpatient rehabilitation centers for patients with chronic musculoskeletal disorders. In Germany, these services are provided by the German Pension Insurance in order to improve or to restore work ability and to prevent health-related early retirement. We included employed women at the beginning of their three-week rehabilitation programs. Rehabilitation was granted due to work ability restrictions related to musculoskeletal disorders. Data were collected through questionnaires. Age and diagnoses according to the International Classification of Diseases 10th revision (ICD-10) were extracted from the standardized rehabilitation discharge forms. Ethical approval was obtained from the Charité – Universitätsmedizin Berlin (EA1/049/11). Additional approval was gained from the data protection commissioner of the Federal German Pension Insurance.
Measures
Work-Family Conflict Questionnaire
The German version the Work-Family Conflict Questionnaire (WFCQ) is based on the original version by Kelloway, Gottlieb and Barham [14] and consists of 22 items, which can be grouped into four subscales: time-based WIF (five items), strain-based WIF (six items), time-based FIW (five items) and strain-based FIW (six items). The translation procedure was guided by the recommendations of Beaton et al. [17] and involved five steps: first translation, design of a preliminary questionnaire, back translation, consent of a commission of experts and testing of the preliminary questionnaire.
The first translation was done by two people whose first language was the target language (German). Both researchers were familiar with the subject of the WFCQ. Additionally, the first phase was supported by the translation of an English teacher, acting as a naive element. She focused on the general comprehensibility of the items. This first translation procedure resulted in three versions. Throughout the course of designing a preliminary questionnaire, another German speaking researcher was consulted. Taking into account the original version of the WFCQ, all three translations were compared and a first version of the German WFCQ was prepared. Following this, two persons whose first language was the original language (English) independently retranslated the questionnaire. Both persons (Australian, US-American) were not familiar with the WFCQ and had no medical background. Based on all of the translations, all of the translators involved so far (commission of experts) created the final preliminary questionnaire.
This questionnaire version was preliminarily given to 154 female patients in order to test its linguistic comprehensibility. As there were no comments that indicated a further need for revision, this version was used in the current study. The original and the translated items are presented as Additional file 1. All of the items were rated using a four-point scale (1 = never; 4 = almost always). The total scores of the four subscales ranged from 5 to 20 points (time-based WIF and FIW) and from 6 to 25 points (strain-based WIF and FIW), respectively.
Work Ability Index
Work ability was assessed using the German version of the Work Ability Index (WAI) questionnaire [16], a short self-report measure comprising the following subscores: (1) current work ability compared with lifetime best, (2) work ability in relation to the demands of the job, (3) number of current diseases diagnosed by a physician, (4) estimated work impairment due to disease, (5) sick leave during the past year, (6) own prognosis of work ability two years from now, and (7) mental resources.
The test-retest reliability of the WAI was found to be acceptable [18]. Moreover, several studies have confirmed that a poor WAI rating predicts productivity loss at work, retirement intentions, long-term sickness-related absences, early retirement and different indicators of need for rehabilitation [19-25]. Levels of work ability can be categorized as poor (7 to 27 points), moderate (28 to 36 points), good (37 to 43 points) and excellent (44 to 49 points).
Covariates
We considered age, educational level (low vs. high, i.e. an enhanced lower secondary school certificate) and the primary rehabilitation diagnosis (M40–M54 according to the ICD-10 vs. other musculoskeletal diagnoses) as basic socio-demographic and medical data. For additional adjustments, we also assessed the responsibility for young children (at least one child ≤12 years vs. no children or all children >12 years) and the amount of working time (full-time vs. part-time).
Data analyses
Descriptive statistics were used to characterize the recruited sample. In the case of continuous multi-item measures, Cronbach’s alpha was calculated to determine the internal consistency among items. Values >0.7 were regarded as satisfactory [26].
To check the factorial validity of the German WFCQ, a confirmatory factor analysis (CFA) was performed. This CFA tested whether the assumed four-factor model of the original WFCQ fit the data [27]. Several goodness-of-fit statistics were calculated to validate this assumption. First, the ratio of χ2 and degrees of freedom were obtained. Values less than three indicate a reasonable fit of the hypothesized model as compared to a saturated model [28]. Second, we checked the Goodness of Fit Index (GFI), the Comparative Fit Index (CFI), the Incremental Index of Fit (IFI) and the Tucker-Lewis Index (TLI). The four fit indices yield values ranging from zero to one, whereby values close to one are indicative of good fit and those greater than 0.90 or, even better, 0.95, generally indicate satisfactory fit [27]. Third, the root mean square error of approximation (RMSEA) was inspected. The RMSEA informs on the modeling of the covariance structure. Values less than 0.08 are indicative of good fit [27].
To determine the direct and indirect associations of the WFCQ scales and self-reported work ability as measured by the WAI, a path model analysis was performed. We assumed direct effects of both strain-based scales on the WAI and only indirect effects of the time-based scales, which were mediated by the corresponding strain-based scales. Bootstrapping with 2000 repetitions was performed to determine 95% confidence intervals of the direct and indirect effects. Moreover, modification indices were inspected to determine if additional paths would improve the fitting of the data. Goodness-of-fit statistics of the final model were examined as described above.
Finally, analyses of the assumed direct effects of strain-based WIF and FIW with self-reported work ability were complemented by a set of linear regression models. The first model considered both strain-based scales as explanatory variables. Educational level, age and primary rehabilitation diagnosis were added in the second model. The final model also included the amount of working time and the responsibility for young children.
Statistical differences were regarded as significant if the two-sided P value of a test was less than 0.05. AMOS 21 was used for the confirmatory factor analysis. All other calculations were performed with STATA SE 12.1.