Design
A cross-sectional design was used for the current study, for which two data sources were used: 1) questionnaire data, and 2) register data of work disability assessments. The period of inclusion started in July 2011 and ended in February 2012. Data were collected when study participants approached the maximum term of 24-month sick leave and applied for a work disability benefit at the Dutch Social Security Agency (SSA). In the Netherlands, the SSA is responsible for the assessment of work disability of workers on long-term sick leave. The assessment of functional abilities at 24-month sick leave (the maximum period allowed by law) is done by an insurance physician. If applicable, based on the physician’s report, a labour expert calculates the loss of former wages earned. In 2009, 65% of Dutch cancer survivors who applied for a work disability benefit was granted a full work disability benefit [24]. This implies a wage loss of ≥ 80% of former wages earned.
Questionnaires were sent to the participants at their home address. Upon receipt, data of the questionnaires were linked to SSA data. The study was approved by the Medical Ethics Committee of the VU University Medical Center.
Study population
The study population consisted of sick-listed employed workers (hereafter designated as workers) who were registered at the SSA. They were aged between 18 and 64 years. All workers had a reported diagnosis of cancer, and were approaching a sick leave term of 24 months. Diagnosis had to be confirmed within the first six months of sick leave. Workers were excluded if they received active chemotherapy and/or radiotherapy treatment, if they had a previous diagnosis of cancer but applied for a work disability benefit due to another somatic or psychiatric disorder, if they were self-employed, if they were applying for a revision of a previous work disability assessment, or if they were employed in a so-called sheltered workplace.
Study procedure
Potentially eligible participants of our study were selected at the head office of the SSA. During the period of inclusion, the list of new work disability benefit applications was checked by one author (KBG or PvM) every week. Based on this list of social security numbers and corresponding documents, we selected the sick-listed workers with a diagnosis of cancer, as reported in the attached medical records. After starting the selection, in case of doubt, cases were included based on consensus. Potentially eligible participants received a questionnaire, an informed consent form, and information stating the aim and background of the study. A postage-paid return envelope (to the Research Center for Insurance Medicine at the EMGO + Institute at the VU University Medical Center) and an introductory letter, by the chief medical officer of the SSA, were added. This letter stated the independency of the researchers and stressed that participation would be of no influence on the outcome of the work disability assessment. Participants had to complete the informed consent form by hand and affix a signature. On receipt of the signed informed consent form and the questionnaire, we linked the latter with personal data (i.e., family name, address, birth date), as collected at the SSA head office, and entered these data in a secured database. The chief medical officer of the SSA gave permission to access the SSA’s registry data. A reminder was sent after two weeks. Also, a reminder was sent in case of a missing signature on the informed consent form. Questionnaires of respondents lacking a completed form were destroyed. All respondents received a gift voucher.
Workers who reported to receive chemotherapy and/or radiotherapy and workers of whom the main reason for application was not cancer-related, were excluded. They were sent a letter explaining the reason of exclusion. The questionnaires were checked for completeness and, if necessary, respondents were contacted to supply missing data.
Variables
The independent and dependent variables were collected through questionnaires as used in earlier studies on cancer survivorship and return to work [1, 11, 25–27].
Independent variables
Socio-demographics
The following socio-demographic characteristics were determined: (a) age (in years), (b) gender (male; female), (c) marital status (single; married/living with partner; divorced/widowed), (d) number of children, (e) principal wage earner (yes; no), (f) educational level (no education/primary school/lower vocational education; secondary school; vocational education/upper secondary school; upper vocational education/university), (g) nationality (Dutch; non-Dutch).
Health determinants
The following health characteristics were assessed: (a) tumor type, (b) extensive disease (negative lymph nodes; positive lymph nodes; metastasis), (c) treatment modalities (surgery; radiotherapy; chemotherapy; hormone therapy; bone marrow transplant; immunotherapy), (d) being free of disease (yes; no; don’t know), (e) comorbidity (number of additional diseases). Physical symptom burden was measured using (f) the physical dimension score of the Sickness Impact Profile (SIP), covering three scales, i.e., Body Care and Movement, Ambulation, and Mobility [28]. Also, (g) fatigue, (h) depressive mood, and (i) global health were measured using the Functional Assessment of Chronic Illness Therapy-Fatigue Scale (FACIT-F) [29], the Center for Epidemiologic Studies Depression Scale (CES-D) [30], and the European Organisation for Research and Treatment of Cancer Quality of Life Questionnaire-C30 (EORTC-QLQ-C30) [31], respectively.
Work-related determinants
The following characteristics of the previous job held were determined: (a) type of job (white collar; civil servant; blue collar; health care worker), (b) job tenure (in years), (c) working hours (hours/week), (d) shift work (yes; no), (e) managerial tasks (yes; no), (f) number of supervised co-workers, (g) work demands (psychological; physical; both), (h) company size (number of employees), and (i) work ability expectations (same; increase; decrease; don’t know). Related to the present (j) work status (working; not working), (k) the actual number of working hours were determined. Finally, with the first three items of the Work Ability Index (WAI) [1] current work ability compared to life time best, (m) current work ability related to physical work demands and (n) current work ability related to psychological work demands, were measured [32].
Dependent variable
The primary outcome variable was the level of work disability after 24 months of sick leave. This was operationalised by dichotomising the results of the work disability assessments, the entitlement for work disability compensation, as performed by the SSA. In the Netherlands, the level of work disability is assigned to one out of four categories, depending on wage loss or sustainable absence of functional abilities. If functional abilities are assessed present, wage loss can be either (1) less than 35%, (2) in between 35 to 80%, or (3) over 80% of former wages earned. The compensation granted can be none, partial, or complete, respectively. If a person has no labour capacities (sustainable absence of functional abilities) the claimant is granted (4) a compensation by the Benefit Act for the fully and sustained work disabled. The participants with a wage loss of less than 80% were grouped together, as well as those with a wage loss equal to or more than 80% and those with a permanent and sustainable work disability. Herewith, workers assessed as still being able to earn an income were distinguished from those unable to earn an income, i.e., incomplete versus complete work disability.
Statistical analysis
The following variables were binominal: gender, nationality, work status, principal wage earner, shift work, managerial tasks, and treatment modalities. A number of variables was dichotomized: age, job tenure, working hours per week in previous job, the number of supervised co-workers, working hours per week in present job, scores of the SIP, FACIT-F, EORTC-QLQ-C30, and WAI, using the median as a cut-off point. For the CES-D, the variable was dichotomized at a score of 16, the predetermined cut-off point most often used for likely cases of clinical depression [30]. Categorical variables were marital status, number of children, education, type of job, work demands, company size, comorbidity, tumor type, extensive disease, and being free of disease.
The association between independent variables and the binominal level of work disability at 24 months (wage loss <80%; ≥80%) were analysed with univariate and multivariate methods. For univariate analysis, a Chi-square test was performed using a cut-off for p-values of 0.20. The remaining significant independent variables of univariate analysis were then tested for multicollinearity and accepted in a logistic regression model if correlation coefficients were ≥ -0.6 and ≤0.6 [33]. Next, for each category of variables, i.e., socio-demographics, work-related characteristics and health characteristics, multiple logistic regression analysis was performed, using a backward stepwise method. For each category of variables, this resulted in a logistic regression model presenting variables associated with work disability at 24 months of sick leave. Next, using the results of these three backward stepwise models a final model was built. In this final model, variables of the three categories i.e. socio-demographics, health determinants and work-related determinants were added in consecutive order. This resulted in a final model, presenting variables associated with work disability at 24 months of sick leave, controlling in a hierarchical way for socio-demographics, health determinants and work-related determinants. The association for each independent variable and the level of work disability at 24 months was calculated using odds ratios (OR). In the logistic regression analyses, a cut-off for p-values of 0.1 (Wald statistics) for independent variables was chosen. The Hosmer-Lemeshow test was used to assess the goodness of fit. All analyses were performed using SPSS 20 [34].