Flow of participants through the study. MEMS indicates Medication Event Monitoring System (Aardex, Zug, Switzerland), a type of medication-monitoring bottle cap.
Effects of adherence on international normalized ratio (INR) evaluated by the Medication Event Monitoring System (MEMS) medication bottle caps (Aardex, Zug, Switzerland) using dichotomous values for adherence. For missed openings, nonadherent is defined as more than 20% missed MEMS cap openings and adherent is 20% or fewer missed MEMS cap openings. For extra openings, nonadherent is more than 10% extra MEMS cap openings, and adherent is 10% or fewer extra MEMS cap openings. For any incorrect, nonadherent is more than 20% incorrect MEMS cap openings (either missed or extra), and adherent is 20% or fewer incorrect MEMS cap openings. B indicates INR below range; A, INR above range; and O, INR out of range. *Multivariable P<.001. †Multivariable P = .04 for comparison of adherent vs nonadherent.
Kimmel SE, Chen Z, Price M, Parker CS, Metlay JP, Christie JD, Brensinger CM, Newcomb CW, Samaha FF, Gross R. The Influence of Patient Adherence on Anticoagulation Control With WarfarinResults From the International Normalized Ratio Adherence and Genetics (IN-RANGE) Study. Arch Intern Med. 2007;167(3):229-235. doi:10.1001/archinte.167.3.229
Warfarin sodium is a highly efficacious drug, but proper levels of anticoagulation are difficult to maintain. Conflicting data exist on the influence of patient adherence on anticoagulation control.
We performed a prospective cohort study at 3 anticoagulation clinics to determine the effect of adherence on anticoagulation control. Patients treated with warfarin with a target international normalized ratio of 2.0 to 3.0 were monitored with electronic Medication Event Monitoring System (MEMS) medication bottle caps. Detailed information was collected on other factors that might alter warfarin response.
Among 136 participants observed for a mean of 32 weeks, 92% had at least 1 missed or extra bottle opening; 36% missed more than 20% of their bottle openings; and 4% had more than 10% extra bottle openings. In multivariable analyses, there was a significant association between underadherence and underanticoagulation. For each 10% increase in missed pill bottle openings, there was a 14% increase in the odds of underanticoagulation (P<.001); participants with more than 20% missed bottle openings (1-2 missed days each week) had more than a 2-fold increase in the odds of underanticoagulation (adjusted odds ratio, 2.10; 95% confidence interval, 1.48-2.96). Participants who had extra pill bottle openings on more than 10% of days had a statistically significant increase in overanticoagulation (adjusted odds ratio, 1.73; 95% confidence interval, 1.09-2.74).
Patients have substantial difficulties maintaining adequate adherence with warfarin regimens, and this poor adherence has a significant effect on anticoagulation control.
Poor adherence with medication therapy has been estimated to be responsible for over 5% of all hospital admissions, with an associated direct cost exceeding $8 billion.1 The effects of inadequate adherence would be expected to be particularly problematic with commonly used drugs that have a narrow therapeutic index, such as warfarin sodium.2 Although warfarin is highly efficacious when anticoagulation levels are maintained within the target range, anticoagulation levels above and below the target range are associated with substantial increases in bleeding and thromboembolic risk, respectively.3 Unfortunately, warfarin therapy is plagued by poor levels of anticoagulation control,4- 7 and anticoagulant therapy has been estimated to be a leading cause of preventable adverse drug events among ambulatory elderly patients.8
Despite concerns about the potential effect of inadequate warfarin regimen adherence,9- 13 the effects of inadequate adherence on anticoagulation control have not been rigorously and quantitatively evaluated. Because warfarin has a long effective half-life (about 40 hours14), it is possible that some nonadherence could have a negligible effect on anticoagulation levels.15 In fact, some small studies have suggested that typical levels of adherence in practice might play, at most, a minor role in determining the stability of anticoagulation control in clinical practice.15,16
The International Normalized Ratio (INR) Adherence and Genetics (IN-RANGE) Study has 2 major aims: quantifying the contribution (1) of warfarin adherence and (2) of genetic variability to anticoagulation control. The present article, the first report of the IN-RANGE study, presents the results of the adherence aim.
The IN-RANGE study is a prospective cohort study enrolling patients at 3 specialized anticoagulation clinics: the Hospital of the University of Pennsylvania, Philadelphia; the Philadelphia Veterans Affairs Medical Center; and the Penn State Milton S. Hershey Medical Center, Hershey, Pa. Each clinic has followed its own protocol for dosing. Two clinics used fingerstick devices, and 1 used venous samples from phlebotomy. All patients who came to 1 of the 3 clinics to initiate warfarin therapy were 21 years or older and had a target INR of 2.0 to 3.0 were considered eligible for the study. We excluded patients with abnormal INRs prior to initiating warfarin treatment and those with antiphospholipid antibody (in whom the INR measurement might not be valid17).
Information on factors that can interact with warfarin, warfarin dose, and INR was obtained prospectively by trained study interviewers. Participants also completed a prospective 1-week diary of their intake of vitamin K–rich foods at baseline,18 using a modification of a previously validated instrument.19
The Medication Event Monitoring System (MEMS) medication bottle cap (Aardex, Zug, Switzerland) records the exact time and date of pill bottle opening and was fitted either directly onto participants' warfarin bottles if they used pill bottles for their medication or on an empty bottle if they used alternate containers (eg, 7-day pill reminder boxes). These latter participants continued to use their alternate containers for their warfarin medication while enrolled in the study but were instructed to open and close the bottle with the MEMS cap each time they took their warfarin so that it would function as a diary of their doses.
Follow-up questionnaires were completed at subsequent anticoagulation clinic visits. These questionnaires included questions about changes in medications, changes in vitamin K intake or alcohol consumption, weight change, and instructions by practitioners to temporarily stop taking their warfarin. Anticoagulation clinicians were blinded to the adherence data.
For each patient, the percentage of days that the incorrect warfarin dose was taken (Pincorrect) was defined as the number of days the patient either did not open the bottle or opened it more than once divided by the number of days in the monitored period. Days in which a clinician instructed the participant to not take a dose were considered correct if the patient did not open the bottle. The percentage of overadherent days (Pover) was defined as the number of days the patient opened the bottle more than once on days they were instructed to take a pill plus the number of days that they opened the bottle when they were instructed not to take any pills, divided by the number of days in the monitored period. The percentage of underadherent days (Punder) was defined as the number of days with no bottle openings divided by the number of days when the patient was instructed to take a dose. Each adherence variable was calculated for the time interval between the current INR measurement and the immediately preceding INR measurement.
The primary outcome of the study was the participants' INR at each visit. The primary measures of INR were underanticoagulation (INR below the lower limit of the INR range) and overanticoagulation (INR above the upper limit of the INR range). In secondary analyses, we examined the association between adherence and any out-of-range INR and more extreme values of underanticoagulation (INR ≤1.5). There were too few highly elevated INRs (n = 24 >4.0) to examine this outcome.
For our primary analysis, we categorized adherence a priori into groups that divided the population into roughly equally sized, clinically meaningful groups and used indicator variables in the regression analyses. We also analyzed the adherence categories as ordinal variables to test for trends across the categories and analyzed adherence as a continuous variable in the regression analyses. In addition, we set 20% as the cutoff to dichotomize Punder (≤20% and >20%), since this value was the mean percentage of incorrect doses in our pilot data. In contrast, we set a cutoff of 10% for Pover (≤10% and >10%), since only 3% of visits were characterized by Pover of 20% or higher.
We used multivariable generalized estimating equation regression with a first-order autoregressive correlation structure to account for the lack of independence in the data. In the adjusted analysis, 3 types of potential confounders were defined. Included in the first type were variables that were forced a priori in all models and included age, sex, race, insurance status, study site, number of days since first visit, average weekly dose of warfarin, history of warfarin use, and use of medications that interact with warfarin.20 The second type included baseline variables that were potential confounders: education; employment; income; body mass index; indication for warfarin treatment; alcohol consumption; number of physician, outpatient clinic, or emergency department visits in the past 12 months; median daily vitamin K intake at baseline; variability in vitamin K intake at baseline; and concomitant medical illnesses. From among these potential confounding variables, those that were associated with INR with a bivariate P value of less than .20 for any of the INR outcomes were included in the multivariable models. The third type of potential confounding variables included time-varying variables: change in vitamin K intake and alcohol consumption; general health condition; weight gain or loss; and starting or stopping treatment with acetaminophen, aspirin, or medications that interact with warfarin since the prior interview. Information on these variables was available at anticoagulation clinic visits during which participants completed study interviews. Of these variables, only general health produced a change in any of the adherence-INR outcome odds ratios (ORs) of more than 10%.
In additional analyses, we examined whether the method of monitoring adherence modified the association between adherence and outcome by using interaction terms in our regression models for adherence by use of MEMS cap as a diary for those dispensing their medication from alternative containers vs MEMS cap used on the actively used warfarin bottle. Because adherence may have different effects on INR during the initiation phase of warfarin therapy vs the maintenance phase (after a steady-state maintenance dose has been determined),21 we also assessed the effect of phase on the adherence-INR relationship, using the appropriate interaction terms. Similarly, we included an interaction term for the variable of duration between visits to determine if differing durations between INR measurements altered the adherence-INR associations. Finally, we tested for interactions among clinical variables and our primary adherence-outcome associations using the appropriate interaction terms in multivariable models.
All analyses were performed with SAS software, version 9.1 (SAS Institute, Cary, NC). The institutional review boards at all participating hospitals approved the study, and all participants provided informed, written consent.
The flow of participants through the study is shown in Figure 1. Those who refused to use the MEMS caps were significantly more likely than those who used them to have INR measurements during follow-up that were below the target range (35% vs 26%) (P = .006) but not above target range (14% vs 15%) (P = .93). Table 1 lists details of the study cohort.
The mean follow-up per participant was 32 weeks. The 136 participants contributed a total of 1490 INR measurements (mean number of INRs per subject, 11.0: 33% in the initiation phase and 67% in the maintenance phase), and follow-up information on time-varying covariates was provided for 979 visits. The mean follow-up was 17 weeks during the initiation phase (testing every 15 days on average) and 38 weeks during the maintenance phase (testing every 20 days on average). Of the total, 40.4% of the INRs were out of range, with 25.8% being below (mean INR, 1.70; interquartile range [IQR] 1.55-1.88) and 14.6% above (mean INR, 3.49; IQR, 3.14.0-3.70) the target range.
The median Pincorrect was 15.1% (IQR, 6.4%-33.3%); the median Punder was 12.4% (IQR, 3.5%-31.1%); and the median Pover was 2.1% (IQR, 0.4%-4.8%). The breakdown of adherence by categories is summarized in Table 2. Forty percent of participants were less than 80% adherent (ie, Pincorrect, >20%). Thirty six percent had more than 20% missed pill bottle openings and 4% had more than 10% extra pill bottle openings. The median Pincorrect values at the Hospital of the University of Pennsylvania, Philadelphia Veterans Affairs Medical Center, and Penn State Milton S. Hershey Medical Center were 14%, 14%, and 7%, respectively.
The results of the primary analysis are summarized in Table 3. In bivariate analyses, underadherence was significantly associated with underanticoagulation but not overanticoagulation (Table 3). Multivariable analyses included all variables shown in Table 1. After adjusting for these variables, we still found a significant association between underadherence and underanticoagulation (P<.001) (Table 3). The associations between underadherence and underanticoagulation were statistically significant in all categories in which participants missed more than 20% of pill bottle openings (Table 3). For each 10% relative increase in missed pill bottle openings, there was a 14% increase in the odds of underanticoagulation (P<.001) and a 10% increase in the odds of out-of-range INR (P<.001). There was more than a 2-fold increase in the odds of underanticoagulation among visits with more than 20% (vs ≤20%) missed bottle openings (adjusted OR, 2.10; 95% confidence interval, 1.48-2.96) (Figure 2).
The association between underadherence and low INR was even stronger when we used an INR cutoff for underanticoagulation of 1.5. The ORs for 0.1% to 10.0%, 10.1% to 20.0%, 20.1% to 30.0%, 30.1% to 50.0%, and greater than 50.0% underadherence were: 1.06, 2.44, 4.28, 9.58, and 10.74, respectively (P<.001).
Overadherence was associated with overanticoagulation but not under anticoagulation (Table 3). Participants who had extra pill bottle openings on more than 10% of days had a statistically significant increase in elevated INRs (Table 3 and Figure 2).
Overall nonadherence (Pincorrect) was associated with underanticoagulation and out-of-range INRs but not overanticoagulation (Figure 2).
There was no significant interaction in analyses stratified by whether participants used the MEMS cap on their actively used warfarin pill bottles (n = 51) or as a pill diary (n = 85). The association between adherence and INR remained significant in both groups, although the ORs were somewhat higher among those using the MEMS caps on their pill bottles, albeit with nonsignificant interaction tests. For example, the adjusted OR for more than 20% missed warfarin doses by underanticoagulation was 2.08 (95% confidence interval, 1.16-3.73) for those using the MEMS caps on their active bottles vs 1.97 (95% confidence interval, 1.43-2.71) for those using MEMS caps as a diary (P = .87 for the interaction).
In analyses examining the effect of phase of therapy (initiation vs maintenance), there were no significant interactions for either Punder or Pover on any of the INR outcomes (P>.05 for all). For example, the effect of increasing underadherence on underanticoagulation was similar in the initiation and maintenance phases: for each respective category of increasing underadherence, ORs were 1.54, 1.65, 2.08, 2.81, and 2.59 for the initiation phase vs 1.74, 0.95, 1.73, 2.80, and 3.02 for the maintenance phase. For overadherence (>10%) by overanticoagulation, the ORs were 1.60 vs 1.57 for each phase. There also was no effect of the length of interval between visits on the INR-adherence associations (P>.10 for all). There were no significant interactions between clinical variables and adherence. Finally, we repeated the analyses excluding the first 5 visits to the anticoagulation clinics to reduce the influence of the initial dose titration, and the results were essentially unchanged.
This study demonstrates that poor adherence is potentially a major source of poor anticoagulation control, even among patients being treated in specialized anticoagulation clinics where the importance of adherence is constantly emphasized. Missing 1 to 2 doses a week (20%-30% missed doses) was associated with up to a 2-fold increased odds of subtherapeutic anticoagulation. This effect was independent of the other measured factors that can alter INR and was associated with poor anticoagulation control in both the initiation and maintenance phases of therapy. Even small reductions in the INR below therapeutic levels is associated with substantial risk for thromboembolism3 along with increased need for dose changes, additional visits, and the potential for dosing errors. Taking extra pills was also associated with overanticoagulation, but overadherence was less common than underadherence. A substantial proportion of patients, 40%, had clinically significant levels of poor adherence (>20% missed doses or >10% extra doses).
Patient self-report and clinician assessment suggest that poor adherence may contribute to anticoagulation control,22,23 but both methods are limited by the inaccuracy of the method of adherence assessment.24,25 Although a study using low-dose phenobarbitone to assess adherence noted that extreme levels of nonadherence can clearly affect anticoagulation control,26 no large studies have previously quantified the relationship between more typical levels of adherence and anticoagulation control among patients in routine practice. The authors of 2 small studies using MEMS caps concluded that adherence may not have a substantial effect on anticoagulation stability.15,16 One of these studies (≤30 participants in any 1 group) examined patients taking the longer-acting phenprocoumon and suggested a possible relationship between adherence and INR stability but did not quantify the association.16 The other study, a randomized trial of warfarin vs the shorter half-life drug acenocoumarol, did not demonstrate an association between adherence and anticoagulation control among 40 patients taking warfarin but did demonstrate an association among patients taking acencoumarol.15 However, overall anticoagulation control over the entire course of therapy was compared with overall adherence, not INR-specific antecedent adherence, and data were not presented separately for overadherence and underadherence vs overanticoagulation and underanticoagulation. Our study, whichis larger and correlates INR with antecedent adherence, demonstrates that warfarin adherence does affect anticoagulation control and that even moderate levels of nonadherence are clinically important.
There are several limitations that must be considered in interpreting our study results. First, MEMS cap monitoring does not directly measure adherence. We cannot tell for sure if patients who opened their bottles took the right dose of their medication (eg, took the medication at all or took the correct dose for those prescribed different doses on different days) or if patients who did not open their pill bottles did not, in fact, take their medications (eg, by taking a pill from a pillbox). Despite these possibilities, patients are still nonadherent while being monitored, and electronic pill cap monitoring remains perhaps the best method for measuring adherence in the clinical setting. Most importantly, these misclassifications of adherence in our study would, if anything, underestimate the effect of poor adherence on anticoagulation control. Participants using the pill bottle as a diary may be more likely to have misclassification of their adherence data. However, this misclassification was unlikely to be extreme, as evidenced by the similarly stronger relationship between adherence and anticoagulation control among those using the MEMS cap on their bottles vs as a diary. The Hawthorne effect may lead to improved adherence among study participants; however, this would only serve to limit the degree of subjects' nonadherence, not affect the relation between adherence and outcome.
Second, because of the limited size of our study population, we could not examine the effects of adherence on bleeding or thromboembolism risk. Nonetheless, the degree of anticoagulation control, which we have shown to be affected by poor adherence, is the strongest and most consistent predictor of bleeding and thromboembolism.5
Third, as with all studies of warfarin, there are many other potential reasons for poor anticoagulation control. However, we were able to demonstrate that poor adherence was associated with poor anticoagulation control independent of many factors, including demographic factors, interacting medications, and diet.
Fourth, the generalizability of our findings to other patient populations is unknown, nor can we discern the effects of adherence among those who chose not to use the MEMS caps. However, given the emphasis placed on adherence within specialized anticoagulation clinics and the possibility that patients who did not want to use the MEMS cap were those with the worst adherence (as evidenced by the greater frequency of underanticoagulation among those who refused to use MEMS caps), it is possible that adherence in general practice may be worse than we have estimated.
Finally, given the frequency of poor anticoagulation, our ORs are an overestimate of the relative risk.
Adherence to long-term medication regimens, including cardiovascular therapies, is a major challenge facing patients and health care providers.27 This is particularly true for narrow therapeutic index drugs such as warfarin. Despite patients receiving education about the importance of adherence with warfarin, they still have difficulties maintaining adequate levels of adherence, and this adherence affects their degree of anticoagulation control. Even though the levels of anticoagulation control achieved in our cohort are associated with high levels of efficacy from warfarin, improper anticoagulation is still a major cause of toxic effects, lack of efficacy, and increased cost from warfarin. Future efforts must be made to improve adherence to warfarin regimens and/or to find safer alternatives that are less susceptible to poor adherence.
Correspondence: Stephen E. Kimmel, MD, MSCE, University of Pennsylvania School of Medicine, 717 Blockley Hall, 423 Guardian Dr, Philadelphia, PA 19104-6021 (email@example.com).
Accepted for Publication: September 26, 2006.
Author Contributions:Study concept and design: Kimmel and Metlay. Acquisition of data: Kimmel and Price. Analysis and interpretation of data: Kimmel, Chen, Price, Parker, Christie, Brensinger, Newcomb, Samaha, and Gross. Drafting of the manuscript: Kimmel, Christie, and Samaha. Critical revision of the manuscript for important intellectual content: Kimmel, Chen, Price, Parker, Metlay, Christie, Brensinger, Newcomb, and Gross. Statistical analysis: Chen, Brensinger, and Newcomb. Obtained funding: Kimmel. Administrative, technical, and material support: Price, Christie, Samaha, and Gross. Study supervision: Kimmel.
Financial Disclosure: Dr Kimmel has received research funding from GlaxoSmithKline and has served as a consultant to Bayer and GlaxoSmithKline, unrelated to warfarin.
Funding/Support: This study was funded by grants R01HL066176-04 and P01HS011530 from the National Institutes of Health.
Role of the Sponsors: The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; or preparation, review, or approval of the manuscript.
Acknowledgment: We thank Sandy Barile for editorial assistance; Joseph A. Gascho, MD, for serving as the site investigator at the Penn State Milton S. Hershey Medical Center; and Sarah L. Booth, PhD, for assistance with the vitamin K data. We are also indebted to Mitchell Laskin, RPh; Mabel Chin, PharmD; and Francis Herrmann, BS, RPh, for their dedication to our field work.