Pillbox Use and INR Stability in a Prospective Cohort of New Warfarin Users

BACKGROUND: Warfarin, a frequently prescribed oral anticoagulant, is well known for its narrow therapeutic index. Adherence to warfarin may help to achieve a stable international normalized ratio (INR), but little data are available regarding the impact of using a pillbox as a potential adherence aid device. OBJECTIVE: To evaluate the association between pillbox use and time in therapeutic range (TTR) < 60% and INR instability pattern. METHODS: This study was based on a prospective cohort of 1,069 new warfarin users who initiated warfarin between May 2010 and July 2013 within 17 hospitals in Quebec, Canada. Demographic, lifestyle, and clinical data were collected for 3 months to a year after warfarin initiation, and genetic factors were assessed. Patients usingh self-prepared and pharmacist-prepared pillboxes were compared with nonusers for the 3- to 12-month follow-up period. The primary outcome was a TTR < 60%, which represents a low percentage of time in the INR therapeutic range or an unstable patient. The secondary outcome was the INR instability pattern (unstable below range; unstable over range; unstable with erratic pattern; and stable) to better describe patient INR profiles. A multivariate generalized linear mixed model was used for the primary outcome, along with a multivariate multinomial linear mixed model for the secondary outcome. RESULTS: The cohort included patients with a mean age of 70.4 ± 11.7 years; 61.8% of patients were men; 76.3% had atrial fibrillation as warfarin’s primary indication; and 35.6% had a previous history of myocardial infarction or angina. Self-prepared and pharmacist-prepared pillbox use was not associated with TTR < 60% or a specific INR instability pattern. A sensitivity analysis showed that self-prepared pillbox users had a higher TTR than nonusers (3.55% ± 1.69%; P = 0.036). This effect was greater among patients aged < 70 years (5.48% ± 2.50%; P = 0.029) than among older patients (1.92% ± 2.31%; P =0.406). CONCLUSIONS: Pillbox use was not associated with TTR < 60% or a specific INR instability pattern. The impact of self-prepared pillbox use was greater among younger patients, but results were not clinically significant. Future studies adjusting for concomitant drug use are needed to clarify these results.

D espite the addition of new oral anticoagulants on the market, warfarin remains a frequently prescribed drug to prevent thromboembolic events. 1 However, warfarin's narrow therapeutic index is still an ongoing challenge for clinicians and patients. Multiple blood tests and dosage adjustments are required for patients to achieve an international normalized ratio (INR) within the targeted therapeutic range, defined according to a patient's indication. A subtle change in a patient's clinical profile and behaviors, such as adherence, can have a large impact on warfarin's effectiveness and safety.
As reported in the literature, patients who are taking warfarin and whose INR target range is between 2 and 3 have an increased risk of bleeding, if they are above the targeted range instead of within or below (odds ratio [OR] = 3.2; 95% confidence interval [CI] = 1. 2-8.3). Inversely, patients with an INR below the targeted therapeutic range have an increased risk of having a thromboembolic event (OR = 5.1; 95% CI = 2.9-8.8). 2 In order to assess the quality of warfarin anticoagulation control, a linearly interpolated time in therapeutic range (TTR) is commonly used. 3 A low TTR (< 60%) is associated with higher rates of major bleeding and thromboembolic events, which indicates the importance of good INR monitoring strategies. 4,5 Multiple INR monitoring programs do exist, as well as specialized anticoagulation clinics, but patients attending such clinics only have had a mean TTR of 65.4%. 6 Multiple studies have shown that adherence to warfarin is a key factor in the achievement of a therapeutic and stable INR. [7][8][9][10][11][12] Missing a warfarin dose at least 20% of the time significantly increases the odds of having a subtherapeutic INR of up to • Adherence to warfarin is a key factor involved in the achievement of a therapeutic and stable international normalized ratio (INR), but little is known about the potential benefits of pillbox use. • Pillbox use has been proven to help patients achieve a higher level of adherence in other therapeutic areas.

What is already known about this subject
• This study reported that pillbox use in a real clinical setting was not associated with INR stability. • Patients aged < 70 years who used a self-prepared pillbox achieved a greater INR stability than older users, but this finding was not clinically significant.

What this study adds
drug use, and natural products use), with the exception of physical activity and smoking, were updated via 4 follow-up telephone interviews that took place 3, 6, 9, and 12 months after initiation of warfarin by research nurses and trained research assistants. Because of the noninterventional nature of this study, patients were asked whether they used a pillbox for warfarin at each of these interviews. If pillboxes were used, patients were asked to clarify whether the pillboxes were self-prepared or prepared by a pharmacist. Those lifestyle and pillbox use data defined patients' habits for the following 3 months to avoid temporality bias. Patients were asked whether they missed any of their warfarin doses or experienced any known deviations in dosage during the previous 3 months to evaluate self-reported adherence. Before the interview, patient-reported warfarin dose regimen for the last 7 days were collected, along with the prescribed warfarin dose regimen corresponding to this period. There was no significant difference between reported and prescribed warfarin dose (Pearson coefficient = 0.969), as described in a study using data from the same cohort. 17 INRs were obtained from each patient's monitoring site. At the 12-month follow-up, 76.2% of patients were monitored at a hospital-based anticoagulation clinic; 13.9% were monitored by their physicians; 7.2% were monitored by a pharmacist; and 1.9% were monitored with other means, such as selfmonitoring.

CYP2C9 and VKORC1 Genotyping
Patients provided a blood sample at the beginning of the follow-up for DNA extraction. Genotyping was performed at the Beaulieu-Saucier Pharmacogenomics Centre of the MHI. The Sequenom iPLEX ADME CYP2C9/VKORC1 Panel v1.0 (Sequenom, San Diego, CA) was used to genotype samples from QWCS patients. The mass array was analyzed using Sequenom MassArray Typer software (version 4.0.25.73). Genotypes were subjected to data quality check procedures that ensured single nucleotide polymorphism (SNP) and sample completion rates of 98% or more, absence of plate bias effects, absence of sample duplicates, and related individuals based on identical by descent estimates and Hardy-Weinberg equilibrium (P < 0.05). Two SNPs from the gene CYP2C9 (rs1799853 or CYP2C9*2; rs1057910 or CYP2C9*3) and 1 SNP from the gene VKORC1 (rs9934438) were used in the this study.

Outcomes Measures
The TTR was calculated using linear interpolation between available INR measures for each of the follow-up periods (3-6, 6-9, and 9-12 months). 18 Interview dates were used as delimiters between follow-ups to consider the time-dependant nature of pillbox use. The TTR was also calculated for the entire study period (3-12 months). A TTR was not calculated for the first 3 months of warfarin therapy to discriminate the potential impact of a patient's stabilization period and inexperience with warfarin. A TTR was excluded from the analysis for the corresponding follow-up period if (a) there were more than 70 days (10 weeks) between 2 INR measures and (b) if the patient temporarily 2.8-fold (P < 0.001). Inversely, overadherence at least 10% of the time significantly increases the odds of having a supratherapeutic INR of up to 1.7-fold (P = 0.02). 13 Patients can typically be overadherent to warfarin by taking an additional dose 1 day because they forgot that they had already taken their daily dose earlier the same day.
In order to improve adherence, and ultimately clinical outcomes, patients taking warfarin are often encouraged to use a pillbox-a container with compartments that correspond to a day or a period of the day into which the medication can be placed in advance. Pillbox use has already been linked to health improvements among HIV-infected patients taking antiretroviral therapy and patients with uncontrolled hypertension who are taking at least 3 antihypertensive drugs. 14,15 Therefore, there is an implicit belief that using a pillbox for warfarin improves adherence and ultimately clinical outcomes, but very limited data with a sufficient sample size are actually available to support this belief. 16 Moreover, studying the link between adherence and pillbox use in a cohort of warfarin users is subject to bias, since pharmacy claims data are unreliable because of the multiple warfarin dose adjustments, and the use of electronic pillboxes (MEMs) would prevent the use of a comparison group. In order to assess the potential benefits from using a pillbox, the purpose of this study is to evaluate the association between pillbox use and 2 clinical outcomes: TTR < 60% and the INR instability pattern.

■■ Methods Patient Selection
Data were used from the Quebec Warfarin Cohort Study (QWCS), an ongoing prospective and multicentric cohort study that assesses the genetic, clinical, and environmental risk factors associated with the effectiveness and safety of warfarin. The ethics committees of the Montreal Heart Institute (MHI) and the 17 participating hospitals approved the study. All of the participating hospitals had cardiology departments where research nurses could screen patients in order to select potential eligible participants.
Study participants started using warfarin between May 1, 2010, and July 31, 2013. The primary indication for warfarin treatment was atrial fibrillation, mechanical valve replacement, or mitral stenosis. Participants had to be aged at least 18 years in order to participate. Patients excluded from the QWCS were those with a history of clinically important bleedings or a recent gastrointestinal bleeding or hemorrhagic stroke (< 3 months before warfarin initiation) and those with known coagulation factors deficiency, chronic thrombocytopenia, hematologic malignancy, cirrhosis, chronic hepatitis, jaundice, or cognitive impairment (unreliability).

Data Collection and Measurements
Sociodemographic data, comorbidities, previous history of cardiovascular disease, and lifestyle data were collected at baseline by a research nurse approximately 2 weeks after warfarin initiation. Lifestyle data (e.g., alcohol intake, diet, over-the-counter stopped taking warfarin for more than 7 days. The distribution of the TTR calculated for each segmented follow-up period (3-6, 6-9, and 9-12 months) suggested a lack of normality, which was confirmed by a Shapiro-Wilk (W) test. The TTR calculated for the overall follow-up period (3-12 months) was distributed normally (W = 0.96). We opted to categorize TTRs to obtain an indicator for INR instability (TTR < 60%), which represents the primary outcome. This indicator was previously associated with a higher rate of thrombosis and bleeding events. 4,5 The secondary outcome was the INR instability pattern, which categorizes patients according to the pattern of their outof-range INRs. 19 Patients with TTR < 60% and ≥ 75% of the outof-range time below the therapeutic range were categorized as unstable below range. Patients with TTR < 60% and ≥ 75% of the out-of-range time above the therapeutic range were categorized as unstable above range. Patients were categorized as unstable with an erratic pattern when they did not fit the first 2 definitions. Patients with TTR ≥ 60% were the reference group.

Covariates
Covariates included age at initiation of warfarin, sex, level of education, and body mass index. The following lifestyle variables were also considered: alcohol intake, green vegetable intake, smoking status, and level of physical activity. Physical activity was evaluated according to the Stanford Brief Activity Survey, a validated questionnaire for cardiac patients. 20 History of hypertension, diabetes, dyslipidemia, myocardial infarction (MI) or angina, and stroke were also covariates, along with the genotype based on the presence of ≥ 1 SNP on either the CYP2C9 and VKORC1 gene.

Statistical Analysis
Users with self-prepared and pharmacist-prepared pillboxes were compared with nonusers for the primary and secondary outcomes between 3 and 12 months of warfarin therapy. The primary outcome (TTR < 60%) was tested with a multivariate generalized linear mixed model. As a sensitivity analysis, this outcome was tested using a different threshold (TTR < 45%). 21

C. Losses to Follow-up and Exclusions
The secondary outcome (INR instability pattern) was tested with a multinomial mixed linear model. Primary and secondary models were implemented with the GLIMMIX procedure in SAS to allow for the time-dependant nature of pillbox use and potential confounders. They could account for up to 3 outcome measures per patient (3-6, 6-9, and 9-12 months of warfarin therapy).
The multivariate models with the lowest Bayesian information criteria were selected. 22 The final models were adjusted for age, sex, warfarin's indication, history of hypertension, dyslipidemia, diabetes, MI or angina, stroke, green vegetable intake, and genotype. Statistical analyses were conducted with SAS version 9.3 (SAS Institute, Cary, NC).

Sensitivity Analyses
The overall TTR (3-12 months) was tested as a continuous variable, since it was distributed normally. A multivariate generalized linear regression model was used that was implemented with a GLM procedure in SAS. Since this model cannot account for time-dependent variables, only patients who did not change their habits in terms of pillbox use or nonuse were included (n = 817, 76.4%). Using this model, the potential age-associated differential association between pillbox use and TTR was also explored by stratifying the population based on the cohort mean age at the initiation of warfarin (< 70 years; ≥ 70 years). Age, mean (SD) 70.4 (11.5) 68. 5

INR Instability Patterns
The majority of unstable patients were below range between 3 and 6 months of warfarin therapy, but this proportion decreased until the end of the follow-up period (58.1%-41.0%). On the other hand, there was a growing proportion of patients who were unstable above therapeutic range or with an erratic pattern (24.4%-35.8% and 17.8%-23.2%, respectively).

Sensitivity Analyses
Using the overall TTR as a continuous variable, self-prepared pillbox users had a significantly higher adjusted TTR than nonusers (3.55 ± 1.69; P = 0.036; Table 4). Using the same model, but subgrouping for age, the impact of self-prepared pillbox use seemed greater among patients aged < 70 years (5.48 ± 2.50; P = 0.029) than older patients (1.92 ± 2.31; P = 0.406). Finally, women had a significantly lower TTR than men (3.70 ± 1.67; P = 0.027), and patients with an INR target between 2.5 and 3.5 had a lower TTR than patients with an INR target between 2.0 and 3.0 (-7.35 ± 2.62; P = 0.005).

■■ Discussion
The impact of pillbox use on INR instability among new users of warfarin was assessed by testing the association of pillbox use with a TTR < 60% indicator and the INR instability pattern. Results were not significant with regards to the primary and secondary outcomes. In the sensitivity analysis, patients using a self-prepared pillbox had a significantly, but not clinically, higher TTR than nonusers. Patients aged < 70 years seemed to derive more benefit from self-prepared pillbox use than older patients.
According to this study's results, patients aged ≥ 70 years had a more stable INR profile during the follow-up period than younger patients. Previous studies have reported an association between younger age and warfarin nonadherence and INR instability. 11,23,24 This association could support this ■■ Results

Demographic and Clinical Data
As of July 31, 2013, 1,069 participants were enrolled in the QWCS to complete a 1-year follow-up. Of those participants, 836, 771, and 721 were included in the analyses for the followup periods at 3-6, 6-9, and 9-12 months, respectively ( Figure 1). The major reason for sample size reduction was related to warfarin cessation. At each of the follow-up periods, approximately 43% of patients were using a pillbox that included warfarin, of which about 75% self-prepared their pillboxes. An increase of the mean TTR was observed across the follow-up periods, as well as a high variability among all subgroups of patients ( Figure 2).
As shown in Table 1, the demographic and clinical characteristics are quite similar between nonusers and self-prepared pillbox users (except for age). On the other hand, pharmacistprepared pillbox users were older, disproportionally female, less educated, differed in their alcohol and green vegetable intake, were less active, more diabetic, had a greater history of MI or angina, and were less prone to have a mechanic valve replacement as a main indication for warfarin than nonusers.
When changing the threshold to define INR instability to TTR < 45% and using the same confounders, results were similar. Patient-prepared pillbox users (OR = 0.81; 95% CI = 0.64-1.03; P = 0.087) and pharmacist-prepared pillbox users (OR = 1.04; 95% CI = 0.72-1.50; P = 0.851) were still comparable with nonusers (data not shown).   , N = 1,069) research objectives, since a randomized controlled trial may raise ethical concerns. Pillbox use is recommended by guidelines on warfarin management and is already part of the dayto-day life of 53.1% of patients taking warfarin. [31][32][33] A particular strength of this study is the ability to discriminate between self-prepared and pharmacist-prepared pillbox users. In addition, the use of an incident-user cohort has allowed consideration of the time-dependent nature of pillbox use and other covariables in a mixed regression model. Also, the use of the INR instability pattern as a secondary outcome allows for greater precision in the evaluation of the impact of interventions among warfarin users.

Limitations
Some of the limitations of this study should be mentioned. In particular, it was not possible to directly measure the impact of pillbox use on adherence. Pharmacy claims data can be unreliable because of the multiple warfarin dose adjustments, and the use of electronic pillboxes (MEMs) would create a bias in this particular study. However, no significant difference was found between reported and prescribed warfarin dose (Pearson coefficient = 0.969), suggesting a good compliance from the QWCS participants to prescriber recommendations. 15 Indeed, it was possible to test for self-reported adherence at each interview, with some reservation, since this variable could be biased. When asked, pillbox users tend to recall missed doses better than nonusers, since the latter do not have any visual clue of their omissions and are therefore more quick to perceive themselves as perfectly adherent. This recall bias could result in falsely considering pillbox use as a risk factor for study's finding of a greater impact from self-prepared pillbox use among young patients taking warfarin. MI or angina history was associated with an increased INR instability, which could be an indicator of decreased adherence because of its link with depression and anxiety or co-medication use. [25][26][27] Previous work conducted with the QWCS found that MI history was associated with a lower concordance between self-reported and prescribed warfarin doses. 17 Hypertension was associated with an increased odds of having TTR < 60% but not with a lower TTR. History of MI or angina and hypertension could be indicators of an increased number of concomitant drugs and thus an increased risk of drug interactions with warfarin. Previous studies reported that various comorbidities were linked to a lower TTR and lower adherence, which reinforced this hypothesis. [28][29][30] Finally, female sex and an INR target range of 2.5-3.5 were risk factors for a low TTR, which is concordant with results from previous studies. 24,29 This study is the first prospective cohort study that reports the impact of pillbox use on INR instability among warfarin users. It is also the first study to evaluate the impact of selfprepared pillbox use on this outcome in a real clinical setting. Results are generalizable to other populations of new warfarin users who have access to an anticoagulation monitoring program. A previous study did show an improved TTR from the use of clinician-prepared pillboxes, but the sample size was small (n = 13). Moreover, the data were compared with preenrollment data, which suggests that the results might actually reflect a change in patients' behaviors in this 3-month study (i.e., Hawthorne effect). 14 (N = 1,069) a nonadherence. Using an age-adjusted logistic regression, it was demonstrated that there was indeed a tendency to a higher risk of reporting nonadherence among self-prepared pillbox users when compared with nonusers (OR = 1.22; 95% CI = 0.90-1.65). Similar but significant results were also seen among pharmacist-prepared pillbox users (OR = 2.36; 95% CI = 1.55-3.59).

Multivariate Model of INR Instability Pattern Between 3 and 12 Months of Warfarin Therapy
The absence of concomitant drug information could explain the inability to detect an association with primary and secondary outcomes. Pillbox users could use a greater number of medications than nonusers and be more subject to drug interactions and an increase in INR instability. Therefore, it is possible that true association was overestimated in these analyses. In addition, data on pillbox use was only collected every 3 months during the first year of warfarin therapy, so the exact date of any change in pillbox use or nonuse was not captured. To prevent any temporality bias, the pillbox use data reported at the interview before the follow-up period analysed was considered, not the interview following. Moreover, 76.4% of patients did not change their pillbox use habits during the follow-up, which minimized the issue.
Another limitation of this study was the growing number of losses to follow-up and exclusions from analysis by the time of the 12-month follow-up interview. The main cause was the discontinuation or temporary cessation of warfarin, which was an inevitable event in this observational study. Finally, a healthy user bias may have been present in this study. Pillbox users could have had generally healthier behaviors than nonusers.

■■ Conclusions
This study suggests that pillbox use was not associated with TTR <60% or INR instability, but these results may be biased by unmeasured confounders such as concomitant drug use. Selfprepared pillbox users had a slightly higher TTR than nonusers, which was more apparent in younger patients. Although further research is still needed in this area, the promotion of pillbox use among warfarin users may help achieve therapeutic goals more quickly and safely, which would reduce the need for frequent monitoring visits.

DISCLOSURES
This study was funded by Canadian Institutes of Health Research (CIHR) and the Centre for Excellence in Personalised Medicine. Both funding sources were not involved in the design, conduct, and reporting of this study. The data used for this study came from the Quebec Warfarin Cohort Study (QWCS), which was supported by the CIHR and the Centre for Excellence in Personalised Medicine. Dumas received a doctoral training award from the CIHR. Perreault and Dubé received a salary award from the Fonds Québécois de Recherche en Santé.
Study concept and design were contributed by Talajic, Tardif, Dubé, and Perreault. Dumas, Rouleau-Mailloux, Bouchama, and Lahcene collected the data, which was interpreted by Dumas, Dubé, and Perreault. The manuscript was written and revised by Dumas, Dubé, and Perreault.