Development and Evaluation of the Adherence to Refills and Medications Scale (ARMS) among Low-Literacy Patients with Chronic Disease
ABSTRACT
Objectives: Patient literacy affects many aspects of medication use and may influence the measurement of adherence. The aim of the study is to design and evaluate a medication adherence scale suitable for use across levels of patient literacy.
Methods: The Adherence to Refills and Medications scale (ARMS) was developed, pilot tested, and administered to 435 patients with coronary heart disease in an inner-city primary care clinic. Psychometric evaluation performed overall and by literacy level, included an assessment of internal consistency, test–retest reliability, and factor analysis. Criterion-related validity was evaluated by comparing scores with Morisky's self-reported measure of adherence, medication refill adherence, and blood pressure measurements. Lexile analysis was performed to assess the reading difficulty of the instrument.
Results: The final 12-item scale had high internal consistency overall (Cronbach's α = 0.814) and among patients with inadequate (α = 0.792) or marginal/adequate literacy skills (α = 0.828). Factor analysis yielded two subscales, which pertained to taking medications as prescribed and refilling medications on schedule. The ARMS correlated significantly with the Morisky adherence scale (Spearman's rho = −0.651, P < 0.01), and it correlated more strongly with measures of refill adherence than did the Morisky scale. Patients with low ARMS scores (which indicated better adherence) were significantly more likely to have controlled diastolic blood pressure (P < 0.05), and tended to have better systolic blood pressure control. Lexile analysis demonstrated that the instrument had a favorable reading difficulty level below the eight grade.
Conclusion: The ARMS is a valid and reliable medication adherence scale when used in a chronic disease population, with good performance characteristics even among low-literacy patients.
Introduction
Only 50% to 75% of patients are adherent to medications prescribed for the management of chronic illnesses [1,2]. The forms of nonadherence can be grouped into not filling or refilling medications correctly, and not taking medications correctly [3]. Examples of the former include not filling a new prescription, experiencing gaps between refills, and discontinuing medications. Examples of not taking medications correctly include taking a different dose than prescribed, and taking medication less often than prescribed. Medication nonadherence, in its various forms, is associated with higher hospitalization rates [4], health-care costs [4], and mortality [5].
Patient self-report remains a common method of assessing medication adherence [6]. It is relatively easy and inexpensive, especially when compared with medication event monitoring systems, the measurement of drug levels in the body, or the calculation of refill compliance from claims data. Self-reported assessments of adherence demonstrate different levels of concordance with other measures of medication adherence, a relationship influenced in part by questionnaire design and administration technique [7]. One aspect of questionnaire development which is gaining more attention is suitability across different levels of patient literacy [8].
Limited literacy is a very common problem in the United States, affecting over 90 million adult Americans [9]. Literacy is associated with understanding of drug indications and their potential side effects [10]. Patients with limited literacy skills are less able to identify their own medications and distinguish one from another [11]. They are also twice as likely to misinterpret prescription drug labels and auxiliary warning labels [12,13]. In a study of patients with human immunodeficiency virus, Kalichman and colleagues demonstrated that limited literacy was associated with worse self-reported adherence to antiretroviral medications [14]. Nevertheless, other research has not found a clear relationship [10,15,16]. For example, in a prospective cohort study by Gazmararian and colleagues, inadequate literacy skills were significantly associated with poor medication refill adherence in unadjusted analyses, but fully adjusted models showed a strong trend without a statistically significant effect [17].
As investigators continue to explore the relationship between literacy and medication use, it will be important to use validated measures that can be administered to low-literacy populations and still provide an accurate assessment of patients' behaviors. For self-reported measures of adherence, this may mean ensuring that such instruments perform well across different levels of literacy [18]. For instance, low-literacy patients may have difficulty understanding certain terms, or they may interpret questions differently from patients with higher literacy skills, potentially leading to biased responses [19]. To our knowledge, no published evaluations have assessed literacy-related issues in the measurement of self-reported adherence. This article describes the development and psychometric evaluation of a medication adherence scale intended for use among patients with chronic medical conditions. We describe its reliability, validity, and performance by level of patient literacy.
Methods
Item Development and Description of the Instrument
In early 2004, we assembled a multidisciplinary team with expertise in medication use, health education, literacy, psychology, and chronic disease management to review the medical literature for scales pertaining to self-reported medication adherence. We sought instruments with simple wording that would be appropriate for use among low-literacy patients. Generalizability across medical conditions was also desired.
We identified one instrument, a four-item measure developed by Morisky and colleagues that has been in use for approximately 20 years [20]. Another instrument, the Hill-Bone Compliance to High Blood Pressure Therapy scale, was also identified [21]. The Hill-Bone instrument was based on the Morisky scale, but it is specific to antihypertensive therapy and also includes items pertaining to lifestyle modification in the setting of hypertension. Although the items from these instruments appear relatively straightforward, neither has been formally evaluated in the context of literacy. The multidisciplinary team reviewed and modified items from these questionnaires to broaden their clinical context and simplify the wording where needed. New items were also written.
Items for the questionnaire were compiled with two subscales in mind—adherence with the filling or refilling of prescriptions, and adherence with taking medications. A total of 14 items initially comprised the instrument, the Adherence to Refills and Medications scale (ARMS). Each item was structured for response on a Likert scale with responses of “none,”“some,”“most,” or “all” of the time, which were given values from 1 to 4. Most items were written so that a lower score indicated better adherence.
Pilot Testing
Ten patients from an inner-city primary care clinic volunteered to assist with pilot testing the initial questionnaire. They completed cognitive interviews, which are helpful in evaluating the thought processes used to interpret and answer questions. The patients also evaluated the wording of the questionnaire draft and provided suggestions to improve the clarity of the items. Their comments were reflected in the final version of the questionnaire.
Setting and Population
The ARMS was administered orally as part of the enrollment interview for a randomized control trial, the Improving Medication Adherence through Graphically Enhanced interventions in Coronary Heart Disease (IMAGE-CHD) study. The investigation took place in the primary care clinics at Grady Memorial Hospital, an urban teaching hospital in Atlanta, Georgia.
From March 30, 2004 to March 7, 2005, all patients who arrived at the clinic for a scheduled appointment were screened for the study. Patients were eligible if they had a documented history of CHD, received their prescription medications from the health system pharmacy, and managed their own medications. Patients were ineligible if they had a corrected visual acuity worse than 20/60, a serious psychotic or mood disorder (schizophrenia, bipolar disorder, or schizoaffective disorder), or had delirium or dementia as determined by several short screening questions. Because the study required longitudinal follow-up, patients were also ineligible if they lacked a mailing address or phone number, were in police custody, or were unable to communicate in English. A total of 970 patients with CHD were approached, 490 met the full eligibility criteria, and 435 participated in the study (89% of eligible patients).
Study Protocol
Consenting patients completed an interviewer-assisted questionnaire on the day of the scheduled clinic visit. The questionnaire included demographic information, an assessment of patients' literacy skills, the ARMS, and the self-reported adherence measure developed by Morisky and colleagues [20]. To minimize the potential effect of literacy on questionnaire responses, the interviewer placed a printed response scale in front of the patient for each set of items, oriented the patient to the response choices, and indicated that the patient could respond verbally or by pointing to the desired choice. The interviewer read all questionnaire items aloud in a private examination room.
Literacy was measured using the Rapid Estimate of Adult Literacy in Medicine (REALM) [22]. The REALM is a 66-item word recognition and pronunciation test, which uses common terms from the health-care setting. It is the most widely used measure of literacy in health care [23]. Scores on the REALM range from 0 to 66. In this study, we dichotomized REALM scores to indicate patients with inadequate literacy (raw score 0–44, less than or equal to sixth grade reading level) or marginal/adequate literacy skills (score 45–66, greater than sixth grade reading level).
Upon completion of the enrollment interview, patients were compensated $5 and randomized to one of the four study groups—an illustrated daily medication schedule [24], postcard refill reminders, both interventions, or usual care (control). As per study protocol, approximately 3 months after enrollment, patients completed a follow-up questionnaire, which included the ARMS. Blood pressure measurements from the date of enrollment were abstracted from patients' clinic charts.
The study was approved by the Emory Institutional Review Board and the Grady Research Oversight Committee. Statistical analyses were performed with SPSS version 14.0 for Windows (Chicago, IL, USA). A P-value <0.05 was considered statistically significant.
Statistical Analysis
The internal consistency reliability of the ARMS was determined by computing Cronbach's α and by evaluating the item-total correlation coefficient for scale items [25]. These methods assess whether the items in a scale are measuring the same construct. In general, items with a correlation ≥0.3 with the total scale are considered conceptually similar, and an α≥0.7 is considered to indicate adequate internal consistency [26].
To assess test–retest reliability, the baseline and 3-month follow-up responses of patients in the control group of the randomized trial were compared using Spearman's correlation coefficient. Correlations ranging from 0.7 to 0.9 are considered very good.
Principal components exploratory factor analysis was performed to evaluate the internal structure of the ARMS. This technique indicates the domains that may be represented by sets of items within the full questionnaire. First, a correlation matrix of all scale items was examined to note the mean inter-item correlation and to gain an initial understanding of whether the scale would factor adequately in subsequent analyses. To determine the initial number of factors, eigenvalues >1 were used, and scree plots were also examined. Sets of items generated by the Varimax-rotated component matrix were evaluated to determine whether they fit into conceptually sound domains. Items that demonstrated a loading of ≥0.4 were considered to adequately measure a factor.
The final scale was established primarily using results from the internal consistency evaluation to eliminate items that did not fit with the rest of the scale. If the analyses indicated that Cronbach's α would increase as a result of removing a particular item, then before removal, the content of that item was reviewed to determine whether its inclusion in the scale was justified conceptually. Factor analysis was performed on the full and reduced scales, and the loading of items onto subscales was reviewed to verify the appropriateness of scale reduction. Internal consistency analysis was then performed on the reduced scale and among the items of each newly established factor.
To assess the performance of the scale among low-literacy patients, internal consistency and factor analyses were repeated separately in patients with inadequate literacy skills and in those with marginal or adequate literacy skills. We hypothesized that the scale properties would remain relatively stable across literacy levels.
We also performed a Lexile analysis of scale items to assess language complexity, aiming for a reading difficulty of eighth grade or lower as recommended by the Institute of Medicine [27]. Lexile scores are based on sentence length and familiarity of word choice [19]. Use of uncommon words leads to higher scores, which indicate a higher level of difficulty [28]. Lexile scores can range from 0 to 2000L, and scores less than 1000L correspond to an eighth grade level or below. For comparison, Lexile scores were also computed for the Morisky and Hill-Bone instruments.
After the final scale was established, we analyzed criterion-related validity through several steps. First, we assessed the correlation of the ARMS with the measure of self-reported adherence developed by Morisky and colleagues [20], using Spearman's rho. The Morisky scale is comprised of four yes-no items, and each is scored as 1 or 2 to create a composite measure that ranges from four to eight points. Higher scores indicate better adherence. Second, we correlated the ARMS with medication refill adherence during the previous 6 months using Spearman's rho. The cumulative medication gap (CMG) was used as a measure of refill adherence [17]. CMG values may range from 0 (i.e., no gaps between refills, perfect adherence) to 1.0 (i.e., large gaps between refills, poor adherence). It is a common method for assessing refill adherence [29]. Third, we examined predictive validity by correlating ARMS scores with medication refill adherence during the subsequent 12 months among patients in the control group of the randomized trial, again using CMG as the measure of refill adherence. Fourth, among patients with hypertension, we dichotomized ARMS scores at the median and compared the percentage of patients with controlled blood pressure among respondents with low versus high ARMS scores, using chi-square tests. We hypothesized that patients with low ARMS scores, indicative of better medication adherence, would be more likely to have controlled blood pressure. According to national guidelines, controlled blood pressure was defined as a level <140/90 in patients without diabetes, or <130/80 among patients with diabetes [30]. Similar analyses were performed to compare dichotomized ARMS subscale scores with blood pressure control.
Results
Patient Characteristics
The parent trial included 435 patients, all of whom completed the ARMS during the baseline enrollment interview. The study population was 55.6% female, had a mean age of 63.7 years, and was 91% African American (Table 1). Patients had completed a mean of 10.9 years of school. Approximately 45% of the population had inadequate literacy skills as measured by the REALM. This CHD population had high levels of medical comorbidities, including hypertension (98.6%), hypercholesterolemia (87.4%), and diabetes (45.1%). They were prescribed a median of six medications. Among patients with hypertension, 45.7% were considered to have adequately controlled blood pressure at baseline.
| Characteristic | |
|---|---|
| Female sex | 242 (55.6) |
| Age (years), mean ± SD | 63.7 ± 10.3 |
| <65 | 230 (52.9) |
| Race | |
| African American | 396 (91.0) |
| Caucasian | 31 (7.1) |
| Other | 8 (1.8) |
| Marital status | |
| Married or living with someone | 67 (15.4) |
| Divorced or separated | 172 (39.5) |
| Widowed | 123 (28.3) |
| Single or never married | 73 (16.8) |
| Employment | |
| Unemployed | 68 (15.6) |
| Employed, full time or part time | 33 (7.6) |
| Other (e.g., retired, disabled) | 334 (76.8) |
| Years of education | |
| <12th grade | 204 (46.9) |
| ≥12th grade | 231 (53.1) |
| Literacy | |
| Inadequate (REALM score 0–44) | 196 (45.1) |
| Marginal or adequate (REALM score 45–66) | 239 (54.9) |
| Cognitive function | |
| Impaired (MMSE < 24) | 157 (36.1) |
| Not impaired (MMSE ≥ 24) | 278 (63.9) |
| Hypertension | 429 (98.6) |
| Diabetes | 196 (45.1) |
| Hypercholesterolemia | 380 (87.4) |
- Values are presented as n (%) unless otherwise noted.
- MMSE, Mini Mental State Examination; REALM, Rapid Estimate of Adult Literacy in Medicine.
Reliability
Interitem correlations ranged from 0.021 to 0.498 (mean = 0.250). Item-total correlations ranged from 0.153 to 0.589 (Table 2). Question 9 demonstrated the lowest item-total correlation, and it also had the lowest levels of correlation with other individual items (range 0.021 to 0.143). Question 1 also had relatively low correlation with other items (range 0.103 to 0.320) and the overall scale (0.364), and on further review, it was judged to be conceptually different.
| Items | Mean ± SD | Original 14-item scale | Reduced-12 item scale | |||
|---|---|---|---|---|---|---|
| Item-total correlation coefficient | Cronbach's α if item is deleted | Item-total correlation coefficient | Cronbach's α if item is deleted | Lexile score | ||
| 1. How often do you miss scheduled appointments? | 1.51 ± 0.58 | 0.364 | 0.810 | — | — | 680L |
| 2. How often do you forget to take your medicine? | 1.41 ± 0.54 | 0.589 | 0.794 | 0.579 | 0.791 | 630L |
| 3. How often do you decide not to take your medicine? | 1.26 ± 0.50 | 0.468 | 0.803 | 0.451 | 0.802 | 680L |
| 4. How often do you forget to get prescriptions filled? | 1.19 ± 0.45 | 0.466 | 0.804 | 0.466 | 0.801 | 730L |
| 5. How often do you run out of medicine? | 1.58 ± 0.67 | 0.481 | 0.802 | 0.475 | 0.799 | 520L |
| 6. How often do you skip a dose of your medicine before you go to the doctor? | 1.40 ± 0.59 | 0.492 | 0.801 | 0.485 | 0.798 | 970L |
| 7. How often do you miss taking you medicine when you feel better? | 1.23 ± 0.53 | 0.574 | 0.796 | 0.571 | 0.792 | 840L |
| 8. How often do you miss taking your medicine when you feel sick? | 1.23 ± 0.52 | 0.405 | 0.807 | 0.412 | 0.804 | 850L |
| 9. How often do you take someone else's medicine? | 1.20 ± 0.15 | 0.153 | 0.819 | — | — | 640L |
| 10. How often do you miss taking your medicine when you are careless? | 1.32 ± 0.51 | 0.515 | 0.800 | 0.500 | 0.798 | 860L |
| 11. How often do you change the dose of your medicines to suit your needs (like when you take more or less pill than you're supposed to)? | 1.17 ± 0.44 | 0.356 | 0.810 | 0.353 | 0.809 | 960L |
| 12. How often do you forget to take your medicine when you are supposed to take it more than once a day? | 1.31 ± 0.53 | 0.548 | 0.797 | 0.548 | 0.794 | 1110L |
| 13. How often do you put off refilling your medicines because they cost too much money? | 1.32 ± 0.63 | 0.497 | 0.800 | 0.509 | 0.796 | 1100L |
| 14. How often do you plan ahead and refill your medicines before they run out?* | 1.89 ± 0.99 | 0.411 | 0.820 | 0.417 | 0.820 | 1000L |
- * This item was reverse coded.
Cronbach's α for the full 14-item scale was 0.816, which demonstrated good internal consistency. Upon dropping items 1 and 9, Cronbach's α remained high (α = 0.814). The mean inter-item correlation for the reduced scale was 0.287. Item-total correlations in the reduced scale ranged from 0.353 to 0.579 (Table 2).
Of the 96 patients in the control group, 93 (97%) completed the follow-up interview. The assessment of test–retest reliability among these patients showed a good correlation between baseline and follow-up responses (Spearman's rho = 0.693, P < 0.001).
Factor Analysis
Initially, factor analysis of all 14 items based on eigenvalues revealed a three-factor solution which accounted for 47.9% of the variance. Forcing a two-factor solution for all 14 items did not produce clear separation of the items as intended in the instrument design.
Nevertheless, when factor analysis was performed on the reduced 12-item scale (without questions 1 and 9), and a 2-factor solution was forced, the items clustered as expected (Table 3). This supported the scale reduction to 12 items. Factor 1 had an eigenvalue of 4.209 and explained 35.1% of the variance. It contained eight items that assessed adherence to taking medications correctly. Factor 2 had an eigenvalue of 1.199 and accounted for 10.0% of the variance. It was comprised of four items that assessed adherence to refilling medications on schedule.
| Factor 1 rotated component loading | Factor 2 rotated component loading | |
|---|---|---|
| Eigenvalue | 4.209 | 1.199 |
| % variance explained | 35.077% | 9.988% |
| Items | ||
| 2. How often do you forget to take your medicine? | 0.652 | |
| 3. How often do you decide not to take your medicine? | 0.659 | |
| 4. How often do you forget to get prescriptions filled? | 0.707 | |
| 5. How often do you run out of medicine? | 0.751 | |
| 6. How often do you skip a dose of your medicine before you go to the doctor? | 0.625 | |
| 7. How often do you miss taking you medicine when you feel better? | 0.717 | |
| 8. How often do you miss taking your medicine when you feel sick? | 0.435 | |
| 10. How often do you miss taking your medicine when you are careless? | 0.699 | |
| 11. How often do you change the dose of your medicines to suit your needs (like when you take more or less pill than you're supposed to)? | 0.424 | |
| 12. How often do you forget to take your medicine when you are supposed to take it more than once a day? | 0.617 | |
| 13. How often do you put off refilling your medicines because they cost too much money? | 0.673 | |
| 14. How often do you plan ahead and refill your medicines before they run out? | 0.602 | |
- * Questions 1 and 9 were deleted.
- ARMS, Adherence to Refills and Medications scale.
For the first subscale, Cronbach's α was 0.794, and the item-total correlations ranged from 0.344 to 0.598. For the second subscale, the item-total correlation ranged from 0.408 to 0.514, and Cronbach's α was 0.641.
Distribution of Scores
When composite scores were created by treating each item on the ARMS as a four-point question, scores on the 12-item instrument ranged from 12 to 34 [mean = 16.32, standard deviation (SD) = 4.06]. On the eight-item taking medications subscale, scores ranged from 8 to 29 (mean = 10.33, SD = 2.66). On the four-item refilling medications subscale, scores from 4 to 14 were reported (mean = 5.99, SD = 1.98). Individual item means and SDs are provided in Table 2. Lower scores indicate better adherence.
Reading Difficulty and Performance by Literacy Level
The Lexile analyses performed on individual items showed scores that ranged from 520L to 1110L (Table 2). The overall Lexile score of the reduced 12-item scale was 920L, which corresponded to a reading level below eighth grade. The Lexile scores of the Hill-Bone adherence scale were similar (item range of 390L to 1020L, overall score of 850L). Lexile scores for the Morisky scale were lower (item range of 420L to 840L, overall score of 650L), as it consists of short items with few situational contexts.
The internal consistency of the reduced ARMS scale was high among patients with inadequate literacy skills (Cronbach's α = 0.792), as well as those with marginal or adequate literacy skills (Cronbach's α = 0.828). All item-total correlations remained ≥0.3, which also indicated good internal consistency across literacy levels.
When using only the responses of patients with inadequate literacy skills, principal components factor analysis identified the same two factors as found in the overall factor analysis. Nevertheless, when the factor analysis was repeated among respondents with marginal or adequate literacy skills, several questions loaded on more than one factor, and the items did not separate into the intended domains.
Validity Assessment
Scores on the ARMS and its subscales correlated significantly with other measures of medication adherence—the four-item scale by Morisky and colleagues, and medication refill adherence during the previous 6 months (Table 4). Scores on the ARMS and medication taking subscale also predicted refill adherence during the subsequent 12 months. Although the ARMS refill subscale correlated with the Morisky scale and with refill adherence during the previous 6 months, it did not predict subsequent refill adherence. Compared with the Morisky scale, the ARMS correlated more strongly with measures of refill adherence.
| ARMS | ARMS medication taking subscale | ARMS prescription refill subscale | Morisky scale | |
|---|---|---|---|---|
| Morisky scale | −0.651* | −0.686* | −0.431* | 1.000 |
| 6-month refill adherence (retrospective) | 0.323* | 0.303* | 0.227* | −0.223* |
| 12-month refill adherence (prospective) | 0.291* | 0.368* | 0.161 | −0.199† |
- * P < 0.01.
- † P = 0.056.
- Correlation coefficients are Spearman's rho. n = 410 for 6-month refill results. n = 93 for 12-month refill results.
- ARMS, Adherence to Refills and Medications scale.
Of the 429 patients with hypertension, 49.4% had controlled systolic, 77.4% had controlled diastolic, and 46.9% had controlled overall blood pressure. Patients with low ARMS scores (below the median of 16) were significantly more likely to have controlled diastolic blood pressure (81.3% vs. 73.2%, P < 0.05). Patients with ARMS scores <16 also tended to have both controlled systolic (52.2% vs. 46.3%) and overall (50.4% vs. 42.9%) blood pressure, but these differences were not statistically significant. When we examined blood pressure data across ARMS scores to look for gradients in the relationship, we found that diastolic blood pressure control declined with increasing ARMS scores, particularly above scores of 21. Such gradients were less apparent with systolic and overall blood pressure control.
We also examined separately the relationship between scores on the ARMS subscales and blood pressure control. Results were similar to those of the overall scale, except patients with low ARMS refill scores (below the median of six) were significantly more likely to have controlled diastolic blood pressure (81.9% vs. 72.9%, P < 0.05), but the relationship between low ARMS medication taking scores (below the median of 10) and controlled diastolic blood pressure was not statistically significant (80.5% vs. 74.4%, P = 0.13).
Discussion
The ARMS was developed to evaluate self-reported adherence to taking and refilling medications among patients with chronic disease. Psychometric analyses revealed high internal consistency, test–retest reliability, and criterion-related validity. To our knowledge, the ARMS is the first measure of adherence to demonstrate stability across levels of patient literacy.
Clearly, more studies are needed to develop and validate instruments for use among low-literacy patients [8,18,19,31], given the high prevalence of limited literacy skills in the US population [9]. Instruments that address aspects of disease self-management, including medication adherence, are probably most important and will help decipher the complex relationship between literacy and health [32,33]. Other techniques to administer instruments to low-literacy patients, such as reading the items, simplifying response scales, and presenting visual cues, are also needed [19,31].
The ARMS has a number of strengths as an adherence measure for patients with chronic diseases. It was developed and tested among patients with coronary heart disease and other chronic conditions, including hypertension, dyslipidemia, and diabetes mellitus. It is easy to use. It was also designed to include two distinct subscales, and this was supported in the overall factor analysis. The 8-item medication taking subscale assesses a patient's ability to correctly self-administer the prescribed regimen. The 4-item prescription refill subscale assesses a patient's ability to refill medications on schedule. Conceptually, these represent different types of problems in medication use [3,6]. It is desirable for a measure of self-reported adherence to capture these difficulties separately so that interventions can be tailored appropriately [34]. A better understanding of the reasons for nonadherent behavior, accompanied by targeted interventions, will be critical to reducing the excess morbidity and mortality associated with nonadherence in the management of chronic diseases [4,5,34].
Several aspects of the present analysis support the validity of the ARMS. It correlates highly with the Morisky scale, which is perhaps the most commonly used self-reported measure of adherence [20]. Moreover, the ARMS has a stronger correlation with measures of refill adherence than does the Morisky scale, indicating the ARMS may be superior in some respect. Ironically, however, the ARMS medication taking subscale was more highly correlated with refill adherence than was the ARMS refill subscale. The reason for this is unclear and requires further study. It is possible that the refill subscale would benefit from more items (it has four while the medication taking subscale has eight), and we have begun testing an expanded version that addresses different circumstances around refill adherence. A significant association with diastolic blood pressure control also provides evidence of validity for the overall scale. The inability to demonstrate a significant association with systolic and overall blood pressure control may indicate the importance of factors other than medication adherence (e.g., diet and lifestyle) in the control of blood pressure.
A major advantage of the ARMS is its suitability for use among minority populations and patients with limited literacy skills, groups that appear to have lower levels of adherence [17]. Lexile analysis demonstrated that the instrument has a reading difficulty below the eighth grade level, as do the Hill-Bone and Morisky adherence measures. Our experience in administering the ARMS also indicates that patients of all literacy levels are able to complete the assessment when it is presented verbally. Importantly, reliability analyses demonstrated high internal consistency of responses across literacy levels.
One limitation of this study is its performance at a single inner-city hospital that serves a predominately African American population. This may limit generalizability to other settings. Second, patients in the study were taking six medications on average, and the scale may perform differently in populations with less medication use. A third limitation was the 3-month gap between the initial and subsequent administrations of the ARMS. We expect that test–retest reliability would have been higher if the tests had been administered across a more ideal time interval, such as 2 weeks. Nevertheless, the large time gap was required because of the overall study design. Fourth, data were collected at a scheduled clinic appointment. To the extent that appointment keeping reflects adherence to other health behaviors, patients who completed the ARMS may have been relatively adherent. Fifth, the distribution of scores on the ARMS was skewed, with most patients indicating adherent behavior. This is common with self-report scales and may result from social desirability bias, in addition to the scale being administered at a clinic appointment.
In conclusion, the ARMS is a valid and reliable self-reported measure of medication adherence, which performs well across literacy levels. Future research will need to assess its performance in other settings, as well as its ability to measure changes that might result from interventions to enhance adherence [34].
Source of financial support: Supported by a grant from the American Heart Association.
