Exploratory Analysis of the Impact of an mHealth Medication Adherence Intervention on Tacrolimus Trough Concentration Variability: Post Hoc Results of a Randomized Controlled Trial

Abstract

Background:

Medication nonadherence is a leading cause of late allograft loss in kidney transplantation (KT). Tacrolimus trough coefficient of variation (CV), measured using the coefficient of variation, is strongly correlated with acute rejection, graft function, and graft loss.

Objective:

The objective of this study was to determine if this mobile health (mHealth) intervention aimed at improving medication adherence in a nonadherent KT population would affect high intrapatient tacrolimus variability.

Methods:

A 6-month, prospective, parallel-arm, randomized controlled clinical trial was conducted to determine the effects of an mHealth intervention on tacrolimus CV. Intervention arm participants utilized an electronic medication tray and an mHealth app to monitor home-based adherence. Tailored motivational reinforcement messages were delivered to promote competence for adherence. Tacrolimus CV was measured using a 12-month rolling average, assessed at monthly intervals (6-month intervention period and 6 months after completion of the study); 80 were included, 40 in each arm.

Results:

At baseline, tacrolimus CV was similar between arms (37% ± 15% intervention, 37% ± 13% control, P = 0.894). Patients randomized to the intervention had a significant reduction in mean 12-month tacrolimus CVs (P = 0.046) and a significant improvement in the proportion achieving low tacrolimus CV (tacrolimus CV < 40%; P = 0.001), as compared with the control arm.

Conclusion and Relevance:

High tacrolimus CV is a risk factor for acute rejection and graft loss; these results offer the potential promise of improved medication adherence and clinical outcomes through the use of innovative technology.

Introduction

Kidney transplantation (KT) is the gold standard treatment for eligible patients with end-stage renal disease, demonstrating improved life expectancy, superior quality of life, and better psychosocial functioning, as compared with hemodialysis.^1–4 Although short-term outcomes after KT are excellent, with a 97% 1-year graft survival, long-term outcomes remain suboptimal, with a graft half-life of only 9.9 years.⁵ Recipient nonadherence to prescribed immunosuppression is common after KT and has been identified as a primary risk factor for acute rejection, graft loss, and death.^6–10

In modern KT, tacrolimus is the backbone of maintenance immunosuppression with nearly 95% of all KT recipients initiated on the medication.^11,12 Although there is ample evidence that tacrolimus is highly effective at reducing the risk of acute rejection and graft loss,^13–15 the drug has limitations, most notably its side effect burden and pharmacokinetic profile.¹⁶ Tacrolimus is associated with deleterious effects, including neurotoxicity, nephrotoxicity, and metabolic sequelae. Tacrolimus has high interpatient and intrapatient variabilities with regard to absorption, distribution, and clearance. This high interpatient variability and narrow therapeutic range have led clinicians to routinely perform therapeutic drug monitoring using 12-hour trough concentrations.

Recent evidence demonstrates that intrapatient variability in tacrolimus 12-hour trough concentrations is a significant predictor of long-term outcomes in KT recipients. High intrapatient tacrolimus variability is associated with increased risk of acute rejection, graft dysfunction, graft fibrosis, and graft loss.^17–23 The initial evidence that high intrapatient tacrolimus variability was associated with poorer outcomes was published by Borra et al.¹⁷ In that study of 297 KT recipients, tacrolimus trough levels were measured between months 6 and 12 following transplantation. The within-patient variability in clearance was calculated and related to a composite end point that included graft loss, biopsy-proven chronic allograft nephropathy, or a doubling of serum creatinine over the course of follow-up. Sapir-Pichhadze et al¹⁹ published the first study of the long-term effects of tacrolimus variability in KT recipients. This retrospective cohort study of 356 patients examined the association between the SD of tacrolimus levels starting at 1 year after transplant and the composite end point of late rejection, graft glomerulopathy, or graft loss (including death). They found that there was a 27% increase in the adjusted hazard ratio of the composite end point for every 1-unit increase in the SD of tacrolimus levels. In 2016, Shuker et al²⁰ confirmed high intrapatient tacrolimus variability as an independent risk factor for adverse KT outcomes. In this study of 808 kidney transplant recipients (KTRs), tacrolimus intrapatient variability was calculated from trough levels and related to a composite end point of graft failure, late biopsy-proven rejection, transplant glomerulopathy, or doubling of serum creatinine. Patients with high intrapatient tacrolimus variability were found to have a 1.4 times higher risk of reaching the composite end point.²⁰ Several other studies have found high intrapatient tacrolimus variability to be associated with poor graft function,¹⁸ accelerated progression of histological lesions,²² and the development of de novo donor-specific antibodies.²⁴ Although these publications provide strong evidence that tacrolimus variability is a risk factor for poorer outcomes, they offer no insight into whether tacrolimus variability is causal or mutable, or whether improvements in tacrolimus variability will be reflected by better outcomes.

Rahamimov et al²⁵ recently demonstrated that not only were those with high tacrolimus variability at significantly increased risk of graft loss, but that a return to low tacrolimus variability ameliorated that risk. This finding suggests that interventions aimed at reducing tacrolimus variability may have a positive effect on long-term graft survival.

To our knowledge, it has not previously been demonstrated that high intrapatient tacrolimus variability is mutable through targeted intervention. We recently conducted an efficacy randomized controlled trial (RCT) in nonadherent KTRs using mobile health technology in an attempt to positively affect adherence to their prescribed medical regimen. The primary aim of this study was to determine if this mHealth intervention aimed at improving medication adherence in a nonadherent KTR population would affect high intrapatient tacrolimus variability.

Materials and Methods

Study Design

This was an institutional review board–approved, single-center, post hoc exploratory analysis of a 6-month, 2-arm RCT involving KT recipients with poor medication adherence. The primary objective of the RCT was to assess the efficacy of an mHealth intervention, as compared with usual care, in improving medication adherence and blood pressure (BP) control. This post hoc ancillary analysis was focused on assessing if the intervention improved intrapatient tacrolimus trough variability. After a 1-month screening period, patients were randomized to the mHealth-delivered intervention or attention control arm and followed for 6 months, with a planned postintervention 6-month follow-up as well (12 months in all). A total of 80 patients were randomized (40 in each arm). The Institutional Review board of the Medical University of South Carolina approved the study (Pro00053471).

Participants

KT recipients were included in the study if they were adults (≥18 years old) and verified as having both poorly controlled BP and poor medication adherence based on the 1-month screenings of timed dosage intake using electronic medication trays, which housed all medications from the patient’s regimen. Exclusion criteria were as follows: severe cognitive impairment; <6 months or >5 years since transplant; not prescribed BP medications; inability to self-administer medications, measure own BP, or use a mobile phone; inability to speak, hear, or understand English; history of psychiatric illness or substance abuse; or planned pregnancy.

Sample Size Justification

The sample size is based on a power analysis for the primary outcome of a larger study. The primary outcomes of the larger study included the proportion of patients with >0.90 medication adherence scores (based on the date/time stamped scores from the electronic trays) and proportion of patients meeting and sustaining adherence to the Kidney Disease: Improving Global Outcomes (KDIGO) guidelines for clinic-based systolic BP control (<130 mm Hg). For an intent-to-treat analysis with 40 patients per group, a 2-sided χ² test (α = 0.05) will have >90% power to detect a difference of 35% in proportions of those who are medication adherent (or meeting and sustaining adherence to the KDIGO BP guidelines) in the intervention group compared with those in the attention control group at the 3-month time point. A difference of 35% in medication adherence between the groups would be considered clinically relevant and warrant change in clinical practice. The current study is an ancillary analysis of data acquired from a larger study; therefore, the sample size is based on calculations for the larger study.

Intervention

A detailed description of the intervention has previously been published.²⁶ In brief, following a 1-month screening to confirm uncontrolled hypertension and poor medication adherence, participants were randomized to either an mHealth intervention or attention control arm for a 6-month active trial with a 6-month follow-up period. The mHealth (Smartphone Medication Adherence Saves Kidneys [SMASK]) participants utilized an electronic medication tray with reminder capabilities enabled, a Bluetooth-enabled BP monitor, and the SMASK smartphone app to monitor home-based adherence to their medical regimen. The electronic medication tray has 28 compartments for up to 4 doses for 7 days, time stamps compartment use, and is capable of providing auditory (eg, chimes) and visual (ie, blue blinking light) reminders of each patient’s medication schedule. At the prescribed dosing day and time, a blue light blinks on the compartment for 30 minutes. If the compartment remains unopened after 30 minutes of the visual alarm, an intermittent loud chime activates for another 30 minutes. Finally, after another 30 minutes of nonadherence, an automated reminder phone call or SMS text message is delivered to the participant’s phone. Additionally, tailored motivational and positive reinforcement feedback messages were automatically delivered based on the previous day’s calculated medication adherence. The use of tailored motivational reinforcement feedback messages was guided by the underlying tenants of competency and autonomous motivation from the self-determination theory. The library of more than 600 messages was developed by enhancing a generic group of messages used in earlier mHealth medication adherence studies.^27,28 The messages were tailored to the participant based on responses to a values, belief, and goals questionnaire administered at baseline used to identify underlying motivating themes (eg, family, faith, friends, work, community events) to promote consistent engagement in a participant’s medical regimen. The responses to the questionnaire were used in a tree-structured algorithm to generate tailored motivation messages used to promote self-efficacy for medical regimen adherence and autonomous regulation for sustained behavior change.

Additionally, the intervention group was instructed and reminded via SMS messages to use the A&D Bluetooth BP monitor every 3 days, once in the morning and once at night. The BP monitor interfaced with the SMASK app that was installed on their smartphone, and home-based BPs were transferred securely and encrypted to our on-campus HIPAA (Health Insurance Portability and Accountability Act) compliant servers for compilation and analysis by the participants’ health care providers on a biweekly basis.

The attention control participants also used an electronic medication tray but with its reminder capabilities disabled and received text messages with healthy lifestyle tips that included links to PDFs and brief video clips for attention control. The messages included tips related to physical activity, dietary intake, nonexposure to firsthand or second-hand smoke, and benefits of limiting alcohol intake. The messages were delivered every 3 days and required less than 5 minutes to review content.

In short, medication adherence scores were calculated using timestamps of openings of Vaica pill tray compartments for both SMASK and attention control groups. Medication doses taken within a 3-hour window (ie, 1.5 hours on either side of scheduled dosage time) of a patient-selected scheduled time received a full score (100%) for that dose. Timestamps from openings occurring within a 3- to 6-hour window received partial credit (50%) for that dose, and those openings outside 6-hour windows did not count for any adherence credit.²⁹

Outcomes

The primary outcome of this ancillary analysis was the proportion of patients obtaining normal tacrolimus trough variability, compared across study arms. This was defined using an intrapatient 12-month rolling average of the coefficient of variation ([CV]; calculated as [Mean/SD] × 100%). A cutpoint of 40% was used to define normal versus high variability. CV was chosen because most previous studies validating high intrapatient tacrolimus trough variability use this measure, and it is advantageous over other measures because it normalizes variability regardless of mean.³⁰ This allows comparisons between patients with different goal ranges for tacrolimus; 40% was chosen as the cutoff to define normal versus high intrapatient variability because we have previously validated this in our KTR population.

Statistical Analysis

Tacrolimus variability as measured through the aforementioned CV was assessed at monthly intervals for each patient (12-month rolling average), starting at the time of randomization and continuing for 12 months (6 months during the intervention period and 6 months after completion of the study). Rolling averages were calculated and estimated on a monthly interval. Each month, 1 additional month of tacrolimus levels is added to the CV calculation, and the levels from the previous month are removed. Statistical analysis was conducted using a generalized linear mixed model to account for correlation of repeated measures and the nondependence of measures over the course of time. Descriptive statistics are displayed as means ± SDs (continuous variables), medians ± interquartile ranges (ordinal variables), and percentages (categorical variables). Independent univariate comparisons were conducted using the Student t-test, Mann Whitney U test, and Fisher exact test based on variable type and distribution. Repeated measures (tacrolimus CV) were assessed using generalized estimating equations to compare the change over time between the 2 treatment arms. A 2-sided P value of <0.05 was considered statistically significant. All data were analyzed using SAS 9.4 (SAS Institute, Cary, NC).

Results

Participant Characteristics

A total of 587 patients were screened for potential enrollment. The majority of exclusions were controlled systolic BP (<130 mm Hg) at prior clinic visits, >5 years from the date of transplant, and not prescribed medications for hypertension; 384 patients met criteria and were approached for consent, of whom 152 agreed to participate. After the 1-month screen-in period, of the 152 enrolled, 82 were randomized and 71 completed the 6-month study. The majority of screening failures were a result of either poor cellular connectivity at home, well-controlled systolic BP, or good medication adherence. The mean age of the study population was roughly 52 years and 69% were male, with 79% being African American. Most were at least 1 year posttransplant, with a mean time from transplant to randomization of 2.1 ± 2.2 years and with the exception of 1 intervention participant, and it was the first KT for all participants. A total of 905 tacrolimus CV values were incorporated into the analysis, averaging 11.03 per participant over a year of surveillance. The majority of participants were married (51%), had some college education (59%), and a mean annual income of <$50 000 (71%). Most were not working, because of retirement, disability, or unemployment (83%). These participant sociodemographic characteristics were similar between the attention control and intervention arms (see Table 1).

Table 1.

Descriptive Characteristics of Participants.

	Intervention arm (n = 40)	Control arm (n = 40)
Mean age (years ± SD)	52.1 ± 11.3	51.5 ± 12.5
Male	72.5%	65.0%
Female	27.5%	35.0%
Race
African American	80%	77.5%
White	20%	22.5%
Years since transplant	2.0 ± 2.0	2.1 ± 2.4
Number of medications	7.4 ± 1.1	6.1 ± 1.4
Received pancreas transplant, percentage (n)	2.5 (1)	10 (4)
Marital status
Single	20.0%	15.4%
Married/Living with significant other	47.5%	53.8%
Separated/Divorced	20.0%	20.5%
Widowed	12.5%	10.3%
Education level
High school or less	42.5%	40.0%
Partial/College graduate	57.5%	60.0%
Annual income
$0-$25 000	35.0%	52.5%
$25-$50 000	25.0%	30.0%
>$50 000	25.0%	7.5%
Not reported	15.0%	10.0%
Employment status
Full-/Part-time	20.0%	15.0%
Retired/Disabled	70.0%	77.5%
Unemployed	10.0%	7.5%

Medication Adherence

Consistent with EMERGE guidelines,³¹ adherence outcomes are reported as they relate to the implementation phase of medication adherence. The SMASK and attention control’s medication adherence averages during the 1-month screening prior to the onset of the active trial were 53.1% and 45.3%, respectively (P > 0.40). Average medical regimen adherence, as indicated by timestamped medication intake and BP monitoring for the SMASK group, was significantly higher than that of the attention control group (all P values <0.001) at each evaluation point across the 6-month trial. The SMASK group’s average adherence rates were 89.8%, 90.1%, and 88.6% for the 1-, 3-, and 6-month time points, respectively. The attention control group’s average rates for medication adherence were 45.3%, 38.5%, and 45.7% for the 1-, 3-, and 6-month time points, respectively. The electronic Vaica medication trays were returned on completion of the active trial at the 6-month visit; therefore, there are no medication adherence data during the 6-month follow-up period.

Tacrolimus Variability Analysis

At baseline, tacrolimus variability, as measured by the 12-month rolling CV, was similar between the SMASK mHealth intervention arm (37% ± 15%) and the attention control arm (37% ± 13%, P = 0.894). Patients randomized into the SMASK intervention arm had a significant reduction in the mean 12-month rolling average of the CV (see top of Figure 1; P = 0.046). The estimated change in the CV between the control and intervention arms was 0.48% per month (95% CI = 0.03% to 0.94%), translating to a difference of 5.8% (95% CI 0.36% to 11.3%) over the course of the 12-month analysis. Furthermore, in repeated-measures analyses, patients randomized into the SMASK arm had a significant improvement in the proportion of patients achieving an intrapatient tacrolimus CV <40% (see the bottom of Figure 2; P = 0.001) as compared with the attention control arm. In the attention control arm, the proportion of patients achieving a tacrolimus CV of <40% remained consistent at 65% to 70%, whereas in the SMASK intervention arm, the proportion of participants achieving a tacrolimus CV <40% started at 63% and increased to >80% by 12 months after randomization.

Figure 1.

Consolidated Standards of Reporting Trials (CONSORT) diagram for Recruitment, 6-month active trial and 6-month follow-up of the SMASK Trial.

Abbreviations: BP, blood pressure; KTR, kidney transplant recipient; MA, medication adherence; RCT, randomized controlled trial; SBP, systolic BP; SMASK, Smartphone Medication Adherence Saves Kidneys.

Figure 2.

Tacrolimus variability compared between study arms. The top panel is the rolling 12-month average tacrolimus CV. The bottom panel is the proportion of participants achieving a rolling 12-month average tacrolimus CV of <40%.

Abbreviations: Tac, tacrolimus. CV, coefficient of variation

Discussion

The primary findings of this ancillary study provide new preliminary evidence that through a deliberate, technologyaided, multidisciplinary intervention aimed at improving medication adherence, high tacrolimus variability can also be significantly reduced. These findings are novel because we are not aware of any previous RCTs demonstrating significant reductions in tacrolimus variability. Given the robust and growing body of literature demonstrating that high tacrolimus variability is associated with increased risk of graft fibrosis, acute rejection, graft loss, and patient death, these results are also clinically meaningful. Although more studies are warranted to confirm these preliminary findings, the results do offer hope that high tacrolimus variability can, in fact, be modified through mHealth-enabled medical regimen self-management programs guided by a user-centered, theory-guided, iterative design process. Future large-scale studies are needed to fully assess if reducing tacrolimus variability will translate into improved long-term clinical outcomes among KTRs.

The recently described guidelines for reporting on medication adherence interventions, the EMERGE taxonomy,³¹ identifies 3 phases of medication adherence: (1) initiation, (2) implementation, and (3) persistence. Although there are certainly challenges in KT in all 3 phases, this investigation is primarily focused on the implementation phase. Medication adherence is critical for optimal KT outcomes, but testing of interventions directed at improving medication nonadherence (MNA) is sparse. A review of MNA studies performed in solid organ transplant recipients found a meager 12 studies, 7 of which involved KTRs.³² Intervention approaches included patient or primary care provider education and patient-focused motivational, behavioral, or psychological/affective state change. Less than half of the studies observed a significant improvement in adherence to even a single medication. One 6-month electronic monitoring/cognitive behavioral skills enhancement RCT was successful in improving MNA among nonadherent KTRs.³³ The RCT included telephone-delivered skills training to improve medication adherence. MNA was tracked via electronic medication pill bottle caps (MEMS caps) that store date and time stamps of all bottle openings. Compared to an attention education control group (n = 5), the intervention group (n = 8) demonstrated greater improvement in adherence scores (0.72 to 0.88 vs 0.75 to 0.77; P = 0.04). Although these are promising results compared to other MNA intervention trials in KTRs, sample size was small, MNA was not monitored in real time, feedback to the individuals was delivered monthly, and impact of the intervention on any clinical measure was not evaluated. Furthermore, although the MEMS caps were used to track MNA, they did nothing to actively encourage adherence with visual, auditory, or electronic reminder cues. There remains a significant need for evidence-based strategies that show sustained improvement in medication adherence and BP control among KTRs. Although we will perform a formal analysis on contributing factors to our improved and sustained medication adherence and BP control, we can prematurely make informed inferences. Participants who experienced success with the mobile health intervention usually had a spouse or were living with a significant other, endorsed above average self-efficacy, indicated above average health literacy by the month 1 evaluation and after viewing a demonstrative video of SMASK at baseline, and answered positively to a question inquiring whether they believed SMASK would be useful.

In 2019, Rahamimov et al²⁵ published a single-center retrospective study of 878 adult KTRs who received tacrolimus for at least 1 year. The aim of the study was to determine if, in a population with an earlier history of high tacrolimus variability, an improvement in tacrolimus variability was associated with improved graft survival. Exposure to tacrolimus variability was calculated using a time-weighted coefficient of variability (TWCV), with the time after transplant divided into 6-month intervals. With a median follow-up of 1263 days, the authors found that a cumulative TWCV >25% was associated with poorer graft survival (hazard ratio = 3.66). Interestingly, of the 480 patients who had, at one point, a cumulative TWCV >25%, 110 subsequently attained a cumulative TWCV <25% and were found to be at no increased likelihood of graft loss. Although this retrospective study cannot answer questions about causality or mutability, these results do suggest that the reduction of tacrolimus variability might lead to improved outcomes. The authors conclude that MNA is likely an important contributor to tacrolimus variability and that prospective trials are needed to determine whether interventions aimed at reducing tacrolimus variability result in improved graft survival.

The results of our study suggest that with an intervention aimed at improving medication adherence intrapatient variability can be reduced significantly. Larger scale prospective studies with longer follow-ups are needed to determine what the impact of improved tacrolimus variability might be on acute rejection rates, long-term graft function, graft histology, and graft survival. Although a large RCT with longer posttrial follow-up testing of this intervention could answer questions related to causality, mutability, and improvements in outcomes, a pragmatic trial is more feasible. We suggest that attention be turned away from demonstrating associations between tacrolimus variability and different outcome measures and toward efforts aimed at improving tacrolimus variability, with the goal of documenting improved outcomes. Ultimately, these efforts should lead to a change in the standard of care.

The presented findings must be evaluated within the context of several limitations of the study. First, all participants were recruited from a single transplant center, which may call into question the generalizability of the findings. However, this center is the sole transplant service provider for the State of South Carolina and has a catchment population of more than 4.6 million persons that encompasses a wide range of ethnic, educational, and socioeconomic backgrounds. Second, those who chose to participate in the mHealth-based RCT might be predisposed to a more positive attitude toward mHealth and thereby introduce a positive bias. That 71% of those approached agreed to participate suggests that a significant bias toward mHealth is unlikely. Third, the follow-up time and sample size precluded us from having sufficient power to assess hard clinical outcomes in a comprehensive manner, including acute rejection, graft survival, and death. Despite randomization, the 2 groups were unbalanced in income and simultaneous kidney/pancreas (SPK) transplantation, with the control group having more participants with lower income and more receiving SPK. Anecdotally, financial constraints were not noted to be a problem in either group in regard to medication adherence. This could be a result of the fact that our cohort’s average time since transplant was 2 years, indicating that Medicare Part B would still cover the majority of the prescribed immunosuppressants. Study coordinators have no recollection of any participants experiencing barriers to obtaining their full regimens. Furthermore, the immunosuppression regimen for patients with kidney transplant alone and SPK patients are similar; therefore, we do not expect that any bias exists between the groups. Finally, the inclusion of only transplant recipients with poorly controlled BP for this analysis is a limitation in terms of generalizability because undoubtedly those patients with controlled BP would benefit from minimizing tacrolimus variability as well. Because this was an ancillary analysis of a larger project with a primary outcome variable of change in systolic BP among uncontrolled hypertensive KTRs, the inclusion criteria transcended into the current analysis. Clearly, future large-scale studies are warranted in this regard.

Conclusion and Relevance

This study demonstrates potentially improved tacrolimus trough concentration variability through the use of an mHealth-enabled medication adherence system. Because tacrolimus variability is associated with graft outcomes, these results provide the promise that improved medication adherence through the use of innovative technologies may lead to improved clinical outcomes in KTRs.