Patients
A sample of 300 patients was consecutively recruited in 11 regional hospitals in the southwest of the Netherlands from January 2012 to July 2014 for a cohort study on DMARD adherence. Patients who were willing to participate were followed up for 1 year. Patients were included if they were at least 18 years old, were started on one or more DMARDs for RA for the first time, and were able to sufficiently read and understand the Dutch language. For the present analysis, we only selected those patients included before January 2014 and who were diagnosed with RA according to the European League Against Rheumatism (EULAR)/American College of Rheumatology (ACR) 2010 criteria for RA.
Participants in the study were, on a fixed time interval, seen by a research nurse or specialized rheumatology nurse after their regular rheumatologist consultation. Because the time interval in which the patient is seen by the rheumatologist differed per hospital, the time intervals vary. In the first year after diagnosis, RA patients are mostly seen every 3 months, but this time interval varied depending on the rheumatologist follow-up appointment.
The Erasmus MC medical research ethics committee gave their approval to perform the study. Each hospital’s board of directors gave their approval for participation in the study. All participants gave written informed consent for their participation and for retrieving relevant clinical data from their patient file.
Primary outcome
Every 3 months, the DAS28 was measured by a trained rheumatology nurse. The score comprises four domains: swollen joint count (SJC), tender joint count (TJC), erythrocyte sedimentation rate (ESR) and a patient general health assessment using a visual analogue scale (VAS). For patients who dropped out of the study, but did not withdraw their consent, the DAS28 score was retrieved from the patient files.
Clinical covariates
Clinical variables assessed at baseline included symptom duration before diagnosis, ACPA, RF, ESR (or C-reactive protein (CRP)) and joint involvement. ACPA and RF were combined for a RF/ACPA positivity score. Symptom duration was dichotomized in more or less than 6 weeks, according to the EULAR/ACR 2010 criteria for RA. The number of DMARDs used was counted and analyzed as a continuous measure. The use of either subcutaneous methotrexate (MTX) or biologicals was noted from the patient file and entered as a binary variable.
Psychosocial covariates
Symptoms of anxiety and depression were measured at baseline with the Hospital Anxiety and Depression Scale (HADS) [14]. The questionnaire has two subscales: one for anxiety and one for depression. The scores range between 0 and 21, higher scores indicating more symptoms of anxiety or depression.
Adherence measurement
Non-adherence was measured per DMARD using a ‘medication event monitoring system’ (MEMS) device, which consists of a medication vial and a MEMS cap. The MEMS uses a microprocessor in the medication container cap to record day and time of each vial opening. The data stored in the MEMS cap is transferred into a web-based data platform, which compiles hour-by-hour drug dosing histories, and in which medication regimen changes are noted. Indirect adherence measurement with MEMS is regarded as a gold standard, since it objectively measures a necessary behavioural step for adherence in ‘real time’ over a continuum. Disadvantages of using MEMS are the high price, the fact that it does not prove ingestion of medication and that it might be seen as an intervention, although this intervention effect is regarded as negligible [15]. Nursing and medical staff were blind to the adherence data throughout the study.
Extra openings of the MEMS cap were ignored, because these mostly do not represent medication intake, but openings by pharmacists. These would otherwise lead to an overestimation of adherence.
When patients stopped using one or more DMARD on the rheumatologist’s advice, for example in case of laboratory abnormalities, this was noted as a non-monitored period, which means that this period was not assigned as a non-adherence event.
For each individual patient and per DMARD, we calculated per day if there was medication underuse. Underuse was defined as a negative difference between the amount of observed openings minus the amount of expected openings. For MTX, we calculated the underuse not per day, but per week, since this medicine only needs to be ingested weekly. For the 12-week period before each DAS28 measurement, we calculated the proportion of days of DMARD underuse. If a patient used multiple DMARDs in the 12-week period, the mean underuse proportion was calculated. Adherence was also dichotomized using a non-adherence proportion above 0.2 (80 % or less adherence) and using a non-adherence proportion above 0.1 (90 % or less adherence).
When a patient used subcutaneous MTX, the patient was asked to put their folic acid in the MEMS container. The openings of the medication cap to take folic acid would then represent the use of subcutaneous MTX. Adherence to biologicals could not be measured. Patients that used biologicals also used other synthetic DMARDs to which adherence could be measured.
Statistical analysis
Characteristics of the study population and non-adherence per DMARD were described with means, standard deviations, medians, interquartile ranges and percentages as appropriate. Four regression models were run with DAS28 as dependent continuous outcome; at T1, over the period T1 to T2 (two time points), the period T1 to T3 (three time points), and the period T1 to T4 (four time points) respectively.
First, univariate linear regression was performed for the T1 model to identify eligible predictors for the DAS28 score at T1. Predictors entered in the univariate regression were: standardized age, sex, baseline DAS28, RF/ACPA positivity, baseline anxiety, baseline depression, number of weeks using DMARDs, education level (low, medium or high), non-adherence, number of DMARDs used, use of subcutaneous MTX and biological use. Non-adherence and covariates with a p value lower than 0.2 were entered in the multivariate model.
For the influence of non-adherence on DAS28 over T1 to T2 (two time points), T1 to T3 (three time points) and T1 to T4 (four time points), multilevel regression models were performed with patients in the upper level and their repeated measures in the lower level. Variables that were taken into account in the models to predict DAS28 over time were: standardized age, sex, baseline DAS28, RF/ACPA positivity, baseline anxiety, baseline depression, number of weeks using DMARDs, education level (low, medium or high), non-adherence, number of DMARDs used, use of subcutaneous MTX and biological use. All possible predictors were entered in a univariate multilevel regression, taking into account the evolution of disease activity over time. Non-adherence and covariates with a p value lower than 0.2 were entered in the multilevel model. Because of potential collinearity between anxiety and depression, only one of these covariates will be included in the multivariate models.
If our study was a hypothesis-testing study, a Bonferroni correction should have been applied because of the number of possible covariates in the analysis. However, because of the explorative character of our study this requirement would be too strict, since then we would need a p value below 0.004 to reach statistical significance, and then no covariates would be left over.
A p value below 0.05 was considered statistically significant.