Does Electronic Monitoring Influence Adherence to Medication? Randomized Controlled Trial of Measurement Reactivity

Patients were recruited from 13 primary care clinics in Oxfordshire, Buckinghamshire, Suffolk, Essex and Huntingdonshire (UK). Patients were eligible for inclusion if aged 18 years or over with type 2 diabetes of at least 3 months’ duration, able to give informed consent, currently taking any oral glucose-lowering agent and with a HbA_1c ≥ 7.5 % (58 mmol/mol). (An HbA_1c level of 7.5 % or above is widely used in clinical practice to indicate suboptimal blood glucose control which may in turn reflect inadequate adherence to glucose-lowering medication.) Those approached were deemed by their general practitioner to be appropriate for tight glycemic control and independent in medication taking.

We used a parallel group trial design with 1:1 randomization to electronic container or standard packaging and follow up at 8 weeks (Fig. 1). Randomization was conducted before the baseline visit to the clinic (visit 1). Prior to visit 2 at 8 weeks, a second randomization was conducted to evaluate the effect of an adherence intervention delivered during that visit. The adherence intervention trial is reported elsewhere [17]; the present study focuses on the 8-week period up to and including visit 2.

Randomization of patients was carried out by the trial statistician independently of the trial co-ordination and clinic teams. A partial minimization procedure was used to dynamically adjust randomization probabilities to balance the baseline stratification variables (clinic, duration of diabetes, HbA_1c result from the clinic record and self-reported adherence). After the participant had been allocated to group, both the participant and the clinic nurse who saw the participant became unblinded to the participant’s group allocation. However, the laboratory staff who measured HbA_1c were blind to allocation. The London multi-centre research ethics committee reviewed and approved the protocol (06/MRE02/3).

The primary behavioral outcome was self-reported medication adherence measured by postal questionnaire sent 1 week before visit 2 (8 weeks) using the Medication Adherence Report Scale (MARS) developed by Horne and colleagues [18]. We used a five-item version of the MARS that asked about “using your diabetes medicines in the last month”, with item responses scored on a 5-point scale from “always true” [1] to “never true” [5]. Scores are summed to give a score ranging from 5 to 25 with a higher score indicating higher self-reported adherence (Table 1). In a pilot study, we found that this version of the MARS had slightly better psychometric properties than alternative versions (e.g., a version that did not specify a time frame) and compared with the similar questionnaire developed by Morisky and colleagues [19]. In the present study, MARS at follow up correlated significantly (r = 0.27; p = 0.02; n = 81) with electronically measured adherence (using only the data from the last 30 days of the 8-week follow-up period, to match the time frame of the self-report measure). Cronbach’s alpha was 0.82 at baseline and 0.67 at follow-up. The primary clinical outcome was HbA_1c at 8 weeks.

A secondary behavioral outcome measure, available only for participants in the group allocated to using electronic containers, was whether or not the correct number of doses of the main glucose lowering medication was taken as prescribed on each day over the 8-week follow-up period as measured by electronic monitoring.

The clinic nurse identified eligible patients registered with the clinic. Eligible patients were sent a letter from the clinic giving details of the trial, and a questionnaire asking about basic demographics, medication regimen, medication adherence (including the MARS) and beliefs about taking diabetes medicines. Responders were telephoned by the clinic nurse to arrange a recruitment visit at the clinic (visit 1). In advance of the visit, patients eligible and willing to take part were allocated to be dispensed medication in the electronic medication-monitoring device (TrackCap, Aardex, Zurich, Switzerland) or in standard packaging. At the visit, informed consent was obtained, clinical data were collected, blood was taken, and questionnaires completed. For those patients allocated to the electronic container, the nurse briefly explained its purpose and use and gave them an information sheet explaining how to use it. The patient was told that the container counts the number of times that the lid is removed, that they should only open and close the container at the time a tablet is to be taken, and that they should only remove the amount of medication required at the time. The nurse then referred them to the clinic dispenser or pharmacist to dispense their usual prescription for metformin or alternative oral glucose lowering agent in the device. Other medication was dispensed as usual, in standard packaging. For those allocated to the other trial arm, the clinic dispenser or pharmacist dispensed their medication as usual, in standard packaging. Patients in both trial arms who used a pillbox to organize their medications were able to continue to use this method during the trial, with the proviso that those in the electronic container arm had to use the electronic containers for their main oral glucose lowering agent. A second visit was arranged in 8 weeks.

In advance of the 8-week visit (visit 2), patients were sent a questionnaire (which included the MARS) from the coordinating center and were centrally randomized to the adherence intervention or a standard care visit. The patients were not told their intervention group allocation before they attended the visit. At the beginning of the visit, patients in both groups had blood samples taken for measurement of HbA_1c.

The adherence intervention trial was planned to follow up 200 patients [20]. For the measurement effect trial, this sample size gave 80 % power to detect a difference of one point in the MARS self-reported adherence measure assuming a standard deviation of 2.5 (a small to medium effect size), and 80 % power to detect a 0.5 % difference in HbA_1c assuming a standard deviation of 1.25 % (two-sided tests, 5 % significance level).

Analysis of the trial data was by intention to treat. Outcomes were analyzed using analysis of covariance adjusting for their corresponding baseline value to improve precision. The missing indicator method [21] was used so that patients with a missing baseline value could be included. Since the MARS score was skewed, the results were checked using a Mann–Whitney U test.

An exploratory subgroup analysis was conducted to examine whether the effect of using the electronic container was different among participants who used a pillbox to organize their medications compared with the rest of the sample. This was done by extending the analysis of covariance to incorporate a test for interaction between trial arm and pillbox use (yes/no). An additional analysis (suggested by one of the reviewers) was conducted to examine whether the effect of using the electronic container was different among participants who were less adherent at baseline (as measured by self-report) compared with those who were more adherent; this involved testing the interaction between trial arm and baseline adherence (treated as continuous).

Generalized estimating equations (GEE), which take account of the dependence of the observations within individuals, were used to analyze change over time in the proportion of participants in the electronic container group who were adherent according to the electronic monitoring data on each day over the 8-week follow-up period. The analysis specified an autoregressive (order 1) correlation matrix and robust standard errors and tested linear and quadratic effects of time. Analyses were conducted in Stata 11 and PASW 16.