This is a secondary analysis of the MAGIC (ClinicalTrials.gov identifier NCT02416479) and MAGIC Türkiye (ClinicalTrials.gov identifier NCT06106854) studies’ data. The designs were single-blinded (participants), two‐arm randomized controlled trials using repeated measures [14, 15]. Immunosuppressive medication intakes during the three-month screening phase were recorded using electronic monitoring. Immunosuppressive medication adherence during screening was quantified by computing the agreement of recorded intakes with the prescribed number and timing of intakes. After the screening phase, participants with an immunosuppressive medication adherence score of ≥85% exited the studies and those with a score of less than 85% were given the opportunity to enter the intervention phase of the studies where they were randomized to either the SystemCHANGE™ intervention (treatment) or the attention control intervention (control). Electronic monitoring continued in both groups. The 85% cut point was established with adult kidney transplant recipients using a cluster analysis approach where data from those who were adherent with morning and evening immunosuppressive medication doses had an electronic medication monitoring adherence score ranging from 85% to 100% [16]. Additionally, immunosuppressive medication nonadherence below 85% has been shown to impact clinical outcomes for adult kidney transplant recipients [17]. All participants received standard care. The attention control intervention delivered over 6 months included monthly receipt of an educational brochure addressing healthy living with a kidney transplant. After the six-month intervention phase a six-month maintenance phase followed during which there was no intervention. Electronic monitoring of both groups continued. The objective of the maintenance phase was to observe if the effects of the intervention remained once the intervention stopped.

The SystemCHANGE™ intervention is based on socio-ecological theory [18] and process improvement methods [19]. In an iterative process using a plan-do-check-act improvement process, patients create a ‘personalized system solution’ by embedding their medication-taking in existing daily routines, evaluating the solution’s effect on consistent medication-taking guided by objective medication adherence data and an interventionist, then revising the solution as necessary and continuing the cycle.

For the MAGIC US study, participants were recruited from five transplant centers in the midwestern and southern US. For the MAGIC Türkiye study, participants were recruited from a transplant center in Izmir, Türkiye. The inclusion criteria were: individuals 18 years of age or older who had received a kidney‐only transplant; who self‐administered at least one prescribed immunosuppressive medication taken twice daily with a functioning kidney transplant; who were not hospitalized; had no diagnosis that would immediately shorten the lifespan; had access to a telephone; had the ability to speak, hear, and understand English or Turkish; were able to open an electronic monitoring cap on a medication bottle; had agreement from the transplant physician and nephrologist to participate; and were not cognitively impaired. In both studies, all participants who met the study inclusion criteria were invited to be in the study and screened for 3 months for adherence. For the US study, 281 individuals were excluded due to high adherence while 130 participants with medication non-adherence were invited to be in the intervention; 41 of these declined and 89 were randomized. Five participants dropped out from the study during the intervention period. For the Türkiye study, 16 were excluded due to high adherence while 60 participants with medication non-adherence were invited to be in the intervention; 18 of these declined/were unable to continue in the study and 42 were randomized with no attrition during the intervention phase.

The following variables were collected at baseline: gender, age, relationship status, education level (as a 3-level categorical variable: primary school, secondary school, and college or higher), employment status, etiology of kidney disease, and type of transplant (from living or deceased donor).

Immunosuppressive medication adherence was measured using the Medication Event Monitoring System ([MEMS®], AARDEX Group, Belgium) which includes a bottle cap with a microprocessor that records the occurrence and time of each bottle opening [20].

Habit index was computed from the objective medication adherence data using a recently developed habit index [13], quantifying week-by-week consistency in the pattern of medication intakes. This habit index was previously shown to correlate with a widespread, self-reported habit metric [13]. Computation of the objective habit index is located in the Supplementary Material. The habit index was computed for each participant during the entire 3-month screening, during each month of 6-month intervention and during each month of the 6-month maintenance period. The habit index ranges between 0 and 1, with 0 corresponding to taking medication at systematically different times (more than 2 h apart) from 1 week to the other, and 1 corresponding to a participant having a perfectly consistent medication intake pattern from week to week, e.g., taking the medication at 10:47 am every Monday, at 2:59 pm every Tuesday, not taking a dose on Wednesdays, etc.

For the MAGIC US study, the Institutional Review Board determined the present secondary analysis of the MAGIC US study data was exempt (Project number 2095804). For the MAGIC Türkiye study, the Institutional Review Board (Project number 2023/27-02) approved the study. Habit strength and immunosuppressive medication adherence data were extracted by a data analyst from the medication intake data.

Demographic data were characterized using descriptive statistics. Chi-squared tests were used to compare count data between countries, and a Student’s unpaired t-test was used to compare age and baseline habit between the two populations.

A mixed linear model with random intercepts was used to model habit strength, measured at several timepoints, as a function of the following covariates: habit during screening, treatment group, phase in the study (intervention or maintenance), country, gender, age, and type of transplant (from living or deceased donor). Interaction terms between group, phase, and country were also modelled. These variables were selected a priori for inclusion in the model. Addition of more demographic variables in the model was investigated using a bidirectional forward stepwise approach and likelihood ratio tests, as detailed in the Supplementary Material. This procedure did not result in including additional demographic variables in the model.

A mediation analysis was performed to investigate whether habit was a mediator of the effect of the intervention on medication adherence. Medication adherence, more precisely its implementation component [21], was quantified as the proportion of days with exactly two immunosuppressive medication intakes. The mediation analysis used several regressions to estimate the direct effect of habit on medication adherence and its indirect effect through the intervention. The previously selected demographics were included in these regression models. Computations were performed using Python version 3.7.3 [22]. Further details on statistical methodology are provided as Supplementary Material.

This study’s follow up of 6 months for the intervention period and 6 months for the maintenance period are much longer than previous habit strength studies [13]. The SystemCHANGE™ intervention taps into reliable personal systems for linkage of medication taking routines to habit so that medication taking happens at the same time every day without relying on intention or memory. Consequently, habit strength and thus medication adherence is sustained over long periods of time [14].

The original MAGIC studies aimed at testing whether the SystemCHANGE™ intervention improved medication adherence. Studies on behavioral interventions, by design, suffer a self-selection bias, as participants who may benefit the most from an intervention are also more likely to decline participation in the study [28].

The size of the US sample was determined through a power analysis aimed at detecting an effect of the SystemCHANGE™ intervention on medication adherence. This analysis is reported in the original article [14]. On the other hand, the Turkish sample was small, which might have prevented detecting an effect of the intervention during the maintenance period. This could limit the generalizability of the findings to the population. Additionally, important structural differences between the populations such as age, education level, employment status, and type of transplant donor may introduce confounding and limit the interpretation of country-level differences. The finding that the Turkish sample was younger, less educated, retired/less disabled, and received more living donor kidney transplants could have influenced the results and therefore, country-level differences should be interpreted with caution.

This analysis did not examine other variables that may influence immunosuppressive medication taking habit strength such as time of day during which the habit occurred and stability of the context in which the habit took place [10]. Additionally, depression and anxiety, two factors that may influence habit were also not studied [29].

The reference value of the habit index was computed over the entire 3-month screening period, to replicate the methodology that was used for the computation of the medication adherence score in the original MAGIC studies. However, habit (and medication adherence) may have changed during the 3-month screening period, because of the introduction of electronic monitoring of adherence causing a Hawthorne effect. The effect of the introduction of electronic monitoring on adherence is unclear [16, 30, 31] and, to the best of the authors’ knowledge, the effect of the introduction of electronic monitoring on habit has not been studied.

Finally, because this work is a secondary analysis that was not prespecified in the design of the original MAGIC studies, it carries the associated limitations. Deciding upon this analysis a posteriori, once the trial results were already available, introduces a potential risk of analytical bias, meaning these findings should be considered exploratory.

These findings have important clinical implications. Briefly assessing habits of kidney transplant recipients around the prescribed times of immunosuppressive medication taking could provide actionable strategies for linking medication taking to existing routines therefore establishing good medication taking habits. Additionally, conversations between the provider and the patient should happen early in the post-transplant period so positive medication taking habits can be established as soon as possible after transplant.

Further trials are planned to examine whether these results are replicable in those with other chronic conditions such as chronic kidney disease and breast cancer using the SystemCHANGE™ intervention, the habit strength index, and objectively measured medication adherence using the MEMS®. Additionally, multi-level influences on habit strength should be explored, including the link between demographic variables and habit strength. Finally, the 6-month SystemCHANGE intervention involved monthly “checks” with participants collaboratively reviewing their medication adherence data and assessing whether the SystemCHANGE solution was continuing to be effective. The intervention was immediately effective within the first month with a slight drop in habit strength between months 5 and 6. Consequently, testing a 3 and 6 months SystemCHANGE “booster” seems like the next logical step in this line of research.