Study setting
At the time of the study public retirement was available in the Netherlands for all people of 65 years and older. Disability retirement is available to patients below 65 years with either physical or mental impairments that impairs their ability to work.
Study design
This study is a secondary analysis of the COPD sTRAITosphere study and uses an observational cross-sectional cohort study design [16]. Data was collected upon referral from the general practitioner to a pulmonologist for a hospital-based outpatient consultation, through a comprehensive diagnostic examination. The Research Ethics Committee of the Radboud University Medical Centre, Netherlands, approved the study. Due to the observational nature of the study and the provision of usual care, written informed consent was waived (ref: 2017/3597).
Study population
Patients were included between April 2013 to December 2018. Included patients were the first time referred to a pulmonologist with a confirmed diagnosis of COPD and were free from exacerbations three months prior to inclusion. The diagnosis of COPD was confirmed using post-bronchodilator spirometry with a Tiffeneau-Penelli index < 70%. Data were collected from three centers in the Netherlands: Radboudumc, Nijmegen, Amphia hospital, Breda, and Bernhoven hospital, Uden. Patients aged 65 years and above were excluded as this was the minimum retirement age in the Netherlands during the study period (Fig. 1). Patients with incomplete data on education and work were also excluded.
Outcomes and covariates
Patients reported at baseline examination if they were currently employed in paid work (“Yes” or “No”). Patients categorized as “No paid work” may therefore be unemployed or receive disability retirement.
The following other data were collected: Age, sex, comorbidities, educational level, spirometry (performed on Masterlab PFT; Vitalaire, Germany, spirometer, using post-bronchodilator (after inhalation of salbutamol 400 μg) spirometry estimated using the Global Lung Initiative equation), blood gas analysis, body mass index (BMI), six-minute walk test (6MWD), steps per day (recorded with activity monitor (Yamex or Dynaport)), exacerbation history during the last year (patients’ recall), checklist individual strength ((CIS), fatigue assessment), medical research council (MRC) score, clinical COPD questionnaire (CCQ) and smoking status. Education level was grouped into short (finished low-level secondary education or lower (International Standard Classification of Education(ISCED) level 0–2)), and medium/high (finished upper secondary education or above (ISCED 3–8)) [17].
The selection of treatable traits was chosen through the availability of evidence-based interventions, as done by van ‘t Hul et al. [16]. Treatable traits include smoking (currently smoking), activity-related dyspnea (MRC grade ≥ 3), frequent exacerbation (≥ 2 exacerbations per year), abnormal BMI (BMI < 21 or > 30 kg/m2), severe fatigue (CIS score ≥ 36 points), poor exercise capacity (6MWD < 70% predicted) and low daily physical activity (< 5000 steps per day) [18,19,20,21,22,23,24].
Statistical analysis
Continues variables when normally distributed were reported using means and 95% confidence interval (CI), and when not normally distributed as medians with interquartile ranges. Normality was assessed using Shapiro-Wilks test, and visually assessed using histogram and Q-Q plot. Categorical variables were reported using counts and percentages. The distribution of treatable traits between workforce connection groups was shown using a bar chart.
Between group comparison in baseline data was done using an independent T-test for continuous data if normally distributed, and Wilcoxon signed rank test if not normally distributed. For categorical data with two groups chi-square test was used and for multiple groups ANOVA was used.
Primary outcome was examined using a between group comparison of workforce connection groups (paid work vs. not paid work) and treatable traits using binomial logistic regression, reporting odds ratios (OR) and 95% confidence intervals (CI). Age, sex, forced expiratory volume in 1 s (FEV1% predicted) and education level were examined together and included in the analysis of individual treatable trait to adjust for relevant confounders [13, 14, 25, 26]. Patients with missing data in individual treatable traits were excluded from the individual analysis. A multiple logistic regression model including all treatable traits with a p-value below 0.2 in univariate analyses was made to explore the importance of the individual treatable traits in relation to each other. Continuous variables were for the purpose of analysis with logistic regression checked for the linearity assumption by plotting the values against the corresponding logit of the model and if they were not linear then changed to relevant categories. Variables were checked for multicollinearity. Logistic regression models were checked for influential outliers defined as a Cook’s distance above 0.5 or a standardized residual larger than three. There were no influential outliers.
Analysis was reported using R version 1.3.1093. A p-value and confidence interval (CI) was performed for logistic regression with a two-tailed p-value of 0.05 considered significant.