Design
This exploratory post-hoc study is embedded within the RECODE study. The RECODE study was a pragmatic 1:1 cluster randomized controlled trial with 12-month and 24-month follow-up that assessed the long term (cost-)effectiveness of IDM incorporated in primary care in COPD patients.(13, 15) The primary outcome was the difference in health-related QOL measured by the Clinical COPD Questionnaire (CCQ), after 12 months of follow-up (regardless of clinically significant improvement). To assess whether effects were maintained the total study duration was 24 months. Forty clusters of primary care teams were randomized either to the IDM intervention or to usual care.
The IDM intervention included a two-day training in essential components of COPD IDM for general practitioners, practice nurses and specialised physiotherapists. They received information on proper diagnosis, optimizing medication adherence, motivational interviewing, smoking cessation counselling, applying self-management plans including early recognition and treatment of exacerbations, and physical (re)activation and nutritional support, and they were trained in using an IT application that aimed to facilitate communication within the team and with patients. Subsequently, each primary care team designed its own practice plan to implement IDM in daily practice. Usual care was based on Dutch general practice COPD guidelines, in line with GOLD guidelines(2, 25), and typically consisted of regular scheduled visits with a practice nurse with a main focus on spirometry or ad-hoc visits to a general practitioner in case of an exacerbation. IDM with different intervention components delivered by a multidisciplinary team was not regularly implemented in primary care. Practice nurses in the usual care group received a course on technical performance of spirometry only, to divert attention from topics related to our intervention. The study was conducted in The Netherlands between 2010 and 2013. Full details of the design, IDM intervention and usual care are provided elsewhere.(13, 15)
Participants
Patients were included if they were diagnosed according to national/international guidelines with COPD by their treating physician. Patients were invited for formal lung function assessment if spirometry data were unavailable. Exclusion criteria were terminal illness, dementia or cognitive impairment, abuse of hard drugs or alcohol and inability to fill in Dutch questionnaires. In total 1086 COPD patients were included. Participant characteristics are provided elsewhere.(13, 14)
Procedure
Measures
Predictor variables (baseline). Several variables were assessed in the RECODE trial, of which those relevant to the current analyses are reported below.
Socio-demographic factors. Participants provided their gender; age; living condition (alone vs. together); and socioeconomic status based on their educational level (recoded into ‘low education’ referring to no education or lower vocational education, or ‘no low education); and employment status.
Lung function and symptoms. We extracted Forced Expiratory Volume in 1 second (FEV1), post-bronchodilator, % predicted according to age and height from medical files. FEV1 was measured by practice nurses or respiratory nurses if unknown. Dyspnea was assessed with the Medical Research Council [MRC] Dyspnea Scale), with 2 as cut-off value for dyspnea.(26) Data on the total number of moderate/severe exacerbations in the previous year were extracted from medical records, with an exacerbation being defined as a worsening of daily symptoms requiring oral prednisone and/or antibiotic courses and/or hospitalizations.
Comorbidity. Comorbidity variables included presence of major cardiovascular disease, hypertension, diabetes and depression (yes/no). In addition, the Charlson co-morbidity index was used to assess overall comorbidity severity, with higher scores indicating more severe conditions and higher mortality risk according to comorbidities.(27)
Lifestyle, illness behaviour and knowledge We assessed smoking status (self-report); physical activity in Metabolic Equivalent Time [MET] minutes, measured by the International Physical Activity Questionnaire [IPAQ](28); and self-management, measured as taking initiative, investment behavior and level of self-efficacy with the Self-Management Scale-30 [SMAS-30].(29)
Outcome variables (12-month and 24-month follow-up).
Clinical COPD Questionnaire (CCQ). Clinical improvement was assessed with the CCQ, a 10-item COPD-specific QOL questionnaire that includes a symptoms, functional and mental domain(30). Each item is scored on a scale from 0 (best possible score) to 6 (worst possible score). The minimal clinically important difference (MCID) is a decline of 0.4 points.(30, 31) The CCQ was found to be responsive to change in previous studies.(30, 32) In this study two categorical variables were created in order to classify each participant as improved patient’ (i.e. decline in CCQ ≥ 0.4 between baseline and follow-up) or ‘unimproved patient’ (i.e. decline in CCQ < 0.4), at 12-month and 24-month follow-up, respectively. In addition, the numeric CCQ difference scores were used in sensitivity analyses (see Statistical analyses).
Statistical analyses
Preliminary analyses were performed using descriptive statistics to examine how many patients showed clinically relevant improvement. We then performed four sets of logistic regression analyses to examine which baseline characteristics were associated with clinical improvement in the intervention and control group at 12-month and 24-month follow-up (RQ1). Specifically, for each time point and group we first performed univariable logistic regression analyses with the dichotomous CCQ improvement variable as outcome. Significant predictors (p < 0.05) were then added to the multivariable model to examine which characteristics were independently related to clinical improvement. Sensitivity analyses were performed by repeating this procedure with the numeric CCQ difference scores as outcome variables, using univariable and multivariable linear regression analyses. We also performed generalized linear mixed model (GLMM) analyses to account for the effect of clustering, with cluster team added as random effect (see Supplementary material). However, these models often failed to converge, possibly due to small cluster effects. The cluster covariance values were indeed very small and results of GLMM were very similar to logistic regression results.
For RQ2 we performed a set of hierarchical logistic regression analyses to examine whether predictors of clinical improvement differed significantly between the intervention and control group, i.e. moderation analysis. Specifically, we entered the main effects of predictors that were significant in the univariable analyses (in either group), as well as the treatment condition variable, in Step 1. The interaction between the treatment condition variable and a specific predictor (focusing on predictors that showed different effects in the multivariable analyses in the intervention and control group) was entered in Step 2. Participants with full data for the variables included in a specific model were included in the respective analysis. We ensured that assumptions of all analyses were met. Statistical analyses were performed using IBM Statistics SPSS version 23.