The brief medication questionnaire: A tool for screening patient adherence and barriers to adherence

Patient Education and Counseling 37 (1999) 113–124

The brief medication questionnaire: A tool for screening patient adherence and barriers to adherence Bonnie L. Svarstad a

1 ,a ,

*, Betty A. Chewning 2 ,a , Betsy L. Sleath 2 ,b , Cecilia Claesson c

School of Pharmacy, University of Wisconsin-Madison, 425 No. Charter St., Madison, WI 53706, USA b Assistant Professor, College of Pharmacy, University of North Carolina at Chapel Hill, NC, USA c Swedish National Board of Health and Welfare, Stockholm, Sweden Received 3 March 1998; received in revised form 30 April 1998; accepted 16 July 1998

Abstract Self-report tools for monitoring adherence can be useful in identifying patients who need assistance with their medications, assessing patient concerns, and evaluating new programs. The aim of this study is to test the validity of the Brief Medication Questionnaire (BMQ), a new self-report tool for screening adherence and barriers to adherence. The tool includes a 5-item Regimen Screen that asks patients how they took each medication in the past week, a 2-item Belief Screen that asks about drug effects and bothersome features, and a 2-item Recall Screen about potential difficulties remembering. Validity was assessed in 20 patients using the Medication Events Monitoring System (MEMS). Results varied by type of non-adherence, with the Regimen and Belief Screens having 80–100% sensitivity for ‘‘repeat’’ non-adherence and the Recall Screen having 90% sensitivity for ‘‘sporadic’’ non-adherence. The BMQ appears more sensitive than existing tools and may be useful in identifying and diagnosing adherence problems.  1999 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Patient compliance; Drug utilization; Measurement; Questionnaire

1. Introduction Patient non-compliance or lack of adherence with drug regimens continues to be a major problem in virtually all medical specialties, patient populations, *Corresponding author. Tel.: 1 1 608 2652128; fax: 1 1 608 2623397. E-mail address: [email protected] (B.L. Svarstad) 1 William S. Apple Professor 2 Assistant Professor

and health settings [1]. Studies show that approximately 25% of all prescribed doses are omitted by patients [2] and that this non-adherence is a significant factor in cardiovascular morbidity and mortality, rejection of transplanted kidneys, leukemia relapse, vision loss in glaucoma, and other indicators of treatment failure [3–7]. Poor adherence also has been implicated in unnecessary and costly procedures and hospitalization in asthma and other conditions [8]. Researchers have identified many determinants of non-adherence [2,9], specific ways in which communication between professionals and

0738-3991 / 99 / $ – see front matter  1999 Elsevier Science Ireland Ltd. All rights reserved. PII: S0738-3991( 98 )00107-4

114

B.L. Svarstad et al. / Patient Education and Counseling 37 (1999) 113 – 124

patients contributes to non-adherence [9–11], and effective interventions [9,12]. Unfortunately, no single intervention is totally effective for all patients and it is not yet possible to predict which individual or subgroup actually needs a given intervention. Patients also are reluctant to admit non-adherence unless clinicians make specific efforts to monitor the degree of adherence on a regular basis [10,11]. Practitioners therefore need accurate, practical, and clinically relevant tools for ‘‘screening’’ or detecting adherence problems [13–16]. Accurate information about adherence also is useful in ‘‘targeting’’ interventions more effectively and efficiently [17], studying professional–patient relationships [18], interpreting drug effects [4,19], and measuring the outcomes of patient education and disease management programs [20]. Researchers have tested the accuracy of several methods of detecting non-adherence. Unfortunately, no single method is adequate [16]. One of the earliest methods of measuring non-adherence involved physician estimates. Findings revealed that this method was no more accurate than chance alone, leading researchers to abandon it [13,21]. Later studies showed that pill counts provide useful information if done in patients’ homes and the purpose of this assessment is not emphasized beforehand [14]. However, home visits are not always feasible and patients often combine medication from several bottles into a single container, making it difficult to interpret pill counts [22]. Researchers also have criticized clinic-based pill counts, because many patients do not bring containers back to the clinic [23] and those that are returned seriously overestimate adherence when compared to more objective methods [19,24]. Pharmacy refill records and drug claims provide relatively objective, unobtrusive, and inexpensive estimates of adherence in large populations over extended periods of time [25–28]. However, these methods only provide a gross measure of adherence and cannot be used for short-term regimens [20]. Researchers also have used laboratory tests, blood pressure readings, and other physiological measures for detecting non-adherence [13,29,30]; however, these methods are not always available or feasible. Another concern is that these techniques only reflect drug-taking in the day or two before the test [31].

This is a serious drawback, because patients often increase drug intake a few days before coming to the clinic, giving an erroneous impression of adherence [19]. The most innovative and sophisticated method of measuring adherence is the Medication Events Monitoring System ([MEMS], Aprex Corporation, Fremont, CA). The MEMS involves dispensing each patient’s medication in a bottle that contains a microprocessor in the cap. The microprocessor records the date and time of each bottle opening, with each opening counted as a presumptive dose. There is no assurance that patients actually consume their medication, but they would have to open and close the bottle at prescribed intervals on a daily basis to create a false pattern of adherence. Studies have demonstrated that MEMS is more accurate than other available methods and is therefore considered the ‘‘gold standard’’ of adherence measurement [16,19]. Despite these advantages, this tool has not yet been applied widely due to its cost and other practical issues that limit its use in large studies and routine clinical practice [20]. Self-report measures (face-to-face or telephone interviews, questionnaires, diaries) provide a practical and flexible method of assessing adherence and a unique opportunity to identify patient concerns. While self-reported adherence has been linked to clinical outcomes [22], there are serious concerns about the accuracy of these measures due to poor ‘‘sensitivity’’ or ability to detect true non-adherence [18,32]. In fact seven of eight published studies examining the validity of self-report adherence measures show a sensitivity level below 60% (Table 1). This means that existing tools incorrectly classify at least 40% of patients with true non-adherence. Stewart [18] attempted to improve this situation by developing a more specific and comprehensive set of questions. While her approach yielded excellent results in patients with new prescriptions (85.5% sensitivity), it failed to detect most cases of nonadherence in patients with refills (40% sensitivity). There are several explanations for the low sensitivity of existing self-report measures. First, it is possible that no single tool can detect all types of non-adherence and that multiple tools are needed. For example, Park and Lipman [33] found that their instrument was more sensitive in detecting major

B.L. Svarstad et al. / Patient Education and Counseling 37 (1999) 113 – 124

115

Table 1 Studies examining the validity of self-report measures of adherence a Study

Stewart [18]

Morisky et al. [22] Craig [30] Inui et al. [15] Gilbert et al. [13] Haynes et al. [14] Gordis et al [29] Park and Lipman [33]

Criterion measure b

Pill count; , 100% New prescriptions Refill prescriptions Blood pressure Lab test Pill count; , 75% Pill count; , 81% Pill count; , 80% Lab tests; , 75% f Pill count; , 100%

Validity of self-report measure c Sensitivity %

PPV %

85.5 40.0 43.6 e 40.0 55.3 19.0 50.0 53.8 25.0

70.8 50.0 52.0 100.0 88.3 50.0 90.6 100.0 88.2

Specificity % d d

81.0 100.0 87.9 92.0 95.8 100.0 100.0

Accuracy % 75.5 66.7 69.0 85.0 67.6 70.4 75.4 68.9 59.8

a

Adapted from Stewart [13] Table shows criterion for determination of true non-adherence. Studies were included only if objective adherence data were obtained for at least 80% of eligible subjects. c See methods section for definitions of accuracy, sensitivity, positive predictive value (PPV), and specificity. d Not reported for patients with new versus renewed prescriptions; specificity was 69.8% for total sample e Table shows measure’s ability to detect non-adherence as measured by blood pressure (24 / 55) f Patients were considered nonadherent if , 75% of urine specimens were positive for the prescribed drug over a six month study period. b

(repeat) dosage errors than minor (sporadic) dosage errors. Unfortunately, most studies do not distinguish repeat and sporadic non-adherence and simply lump them together when calculating sensitivity. Second, survey researchers have developed a number of time-proven techniques for minimizing the different types of reporting errors that are known to occur when asking people to report the frequency of behaviors that are potentially embarrassing, threatening, or difficult to report accurately for other reasons [34]. Unfortunately, these survey techniques have not been applied rigorously to the design of self-report tools for measuring adherence. For example, survey methodologists generally recommend asking about behavior during a specific time period and using a shorter recall period [34]. However, existing tools for measuring adherence often use long recall periods or an unspecified time period [15,29]. Another limitation of existing tools is that they include questions that might be worded more carefully to reduce memory errors, the level of threat or embarrassment experienced by patients who want to make a favorable impression, and other sources of response error. Examples in the published literature include overly broad questions that ask about multiple drugs in the same item [14], ambiguously worded

questions that ask how ‘‘regularly’’ a drug was used [15], narrowly worded questions which mention only one or two types of nonadherent behavior such as forgetting doses or stopping a medication [22], leading questions which prompt or remind patients about what their doctor expects them to do [13,15], and questions which suggest that non-adherence is ‘‘careless’’ behavior [22]. Whether different types of questions and alternative wording yield more accurate self reported adherence is an issue that clearly deserves further study. The purpose of this study is to test the validity of a new self-report instrument for measuring and monitoring adherence and barriers to adherence from the patient’s perspective. The instrument, referred to as the Brief Medication Questionnaire (BMQ), builds on existing theory and research about patient adherence and survey methodology. Our aims were to develop an instrument that is brief and easy to use, has good sensitivity or ability to detect different types of non-adherence, and has the potential for self-administration by patients with multiple drugs. In this paper, we test the BMQ’s ability to detect repeat and sporadic non-adherence using a 5-item Regimen Screen that asks patients how they took each of their medications in the past week, a 2-item

116

B.L. Svarstad et al. / Patient Education and Counseling 37 (1999) 113 – 124

Belief Screen about drug efficacy and bothersome features, and a 2-item Recall Screen about remembering difficulties.

2. Methods

2.1. Patient recruitment and data collection The Medication Event Monitoring System ([MEMS], Aprex Corporation, Fremont, CA) was used to evaluate the sensitivity of a 2-page Brief Medication Questionnaire (BMQ) that we developed for measuring adherence in patients who were prescribed enalapril and captopril (angiotensin-converting enzyme [ACE] inhibitors). Patients were recruited in three pharmacies operated by a Midwestern health maintenance organization (HMO). Patients were eligible if they resided in a non-institutional setting, did not use special containers, and had three or more scheduled medications (including enalapril or captopril). Eligible patients were approached by an HMO pharmacist to determine interest in the study. Written consent was then sought by a research pharmacist who was not affiliated with the HMO. Patients were informed that they may or may not receive a special container that recorded each opening and that everyone would be interviewed about medication use four weeks after enrollment. Of 48 eligible patients, only four refused. Consenting patients were randomly assigned to two groups: one group received their enalapril or captopril in a MEMS container (n 5 22) and the other group received a standard vial (n 5 22). We employed the MEMS, because it provides detailed information about adherence during a particular week or month and is considered the ‘‘gold standard’’ of adherence measurement [16,19]. It was not possible to blind the patient or interviewer as to MEMS and non-MEMS assignment. However, we were concerned that MEMS patients might try harder to adhere because the MEMS container appears different than a standard vial. We therefore collected pill count data from MEMS and non-MEMS patients, allowing us to test the potential effects of receiving a MEMS container. Of 44 consenting patients, one was unable to complete the study. The remaining patients were

interviewed in their homes 1 month after enrollment. The interview was done by a research pharmacist using the 2-page BMQ. After completing the BMQ, the researcher asked to see the patient’s medication bottles and performed a ‘‘pill count’’. If the patient had a MEMS cap, it was collected and replaced with a standard cap. The research pharmacist later reviewed medical and pharmacy records to verify regimens. The researcher was blinded to MEMS results until all BMQ data were entered into a computerized file.

2.2. MEMS and pill count measures of adherence MEMS data were used to construct three objective measures of ACE inhibitor adherence: (i) type of non-adherence in the past week (repeat, sporadic, none), (ii) rate of dose omission in the past week (proportion of prescribed doses not taken for 7 days prior to interview), and (iii) rate of dose omission in the past month (proportion of prescribed doses not taken for 30 days prior to interview). Type of nonadherence fell into three categories: ‘‘repeat nonadherence’’ (took $ 20% over or under prescribed number of doses); ‘‘sporadic non-adherence’’ (took 1–19% under or over prescribed number of doses); and ‘‘no non-adherence’’ (took 100% of prescribed doses). We used a 20% cutoff for repeat non-adherence to facilitate comparison with past studies [13– 15,29]. To assess the effect of MEMS assignment, we compared rates of dose omission in the past month using pill count data for patients in the MEMS and non-MEMS groups. Rate of dose omission was defined as the proportion of prescribed doses not taken for the past 30 days.

2.3. BMQ measures of adherence and barriers to adherence In this paper, we analyze three generic ‘‘screens’’ or sets of questions from Part A of the BMQ. The first screen includes five items that measure adherence behavior and is called the ‘‘Regimen Screen’’ for potential non-adherence. It begins with a neutral, open-ended question that asks the patient to list all medications taken in the past week. For each medication listed, four questions are asked: ‘‘How many

B.L. Svarstad et al. / Patient Education and Counseling 37 (1999) 113 – 124

days did you take it?’’ ‘‘How many times per day did you take it?’’ ‘‘How many pills did you take each time?’’ and ‘‘How many times did you miss taking a pill?’’ We focused on the past week, because a shorter recall period may reduce reporting error, as suggested earlier. Patients receive a score of ‘‘1’’ if their initial or spontaneous report indicates potential non-adherence with the current regimen for the target medication and a score of ‘‘0’’ if this report indicates no non-adherence. Indicators of potential non-adherence including: failing to mention the target medication (without prompting or interviewer assistance), stating that one cannot answer or remember, reporting any interruption or discontinuation due to a late refill or other reason, reporting any missed doses, or reporting any extra doses. Patient responses to the five items also are used to calculate the rate of dose omission (proportion of prescribed doses omitted according to initial report). (See Appendix A for a copy of the questions and procedures used to score them). After patients finish reporting how each mentioned drug was taken, the interviewer asks one or more neutral probes to identify additional medications that the patient did not mention for some reason (e.g. ‘‘Do you have any other medications that you may have stopped for some reason?’’). If the patient identifies additional medications, the interviewer should list them and repeat the above mentioned questions. However, it is important to mark medications listed in the patient’s initial (spontaneous) report versus additional medications identified after any interviewer probes, reminders, or prompts about a specific drug that was omitted from the patient’s listing of drugs used in the past week. Patients obviously can forget to mention certain drugs taken in the past week, but previous work [10] suggests that patients who ‘‘forget’’ to mention a prescribed drug often have stopped or reduced their use of that drug according to more objective measures. We therefore rely on the patient’s initial (spontaneous) report when scoring the Regimen Screen. The Belief Screen measures two beliefs that have been linked to non-adherence in past studies [9,11]. These particular items address patient concerns or doubts about the efficacy of a given medication and concerns about unwanted side effects, short-term or long-term risks, or other bothersome features of a

117

given medication. For each medication, the patient is asked: ‘‘How well does (did) this medication work for you?’’ (very well, ok, not well). After reporting how well each medication works, the patient is asked: ‘‘Do any of your medications bother you in any way?’’. Patients receive a score of ‘‘1’’ if they respond ‘‘not well’’ or ‘‘don’t know’’ when asked how well the target medication works for them and a score of ‘‘1’’ if the medication was identified as bothersome. Item scores are summed to obtain a total belief score, with positive scores indicating one or more belief barriers (range 5 0–2). The third screen is called the ‘‘Recall Screen’’ and includes two items that measure potential problems remembering all doses. These barriers are identified by reviewing the dosage regimen and by asking the patient ‘‘How hard is it for you to remember to take all the pills?’’ (very, somewhat, not at all). Patients receive a score of ‘‘0’’ they have a single dose regimen (once daily) and report that it is ‘‘not at all’’ hard to remember all the pills, a score of ‘‘1’’ if they have a multiple dose regimen (two or more times per day), or report that it is ‘‘very’’ or ‘‘somewhat’’ hard to remember all the pills, and a score of ‘‘2’’ if both indicators are present. The BMQ also asks about difficulties opening the container, reading labels, obtaining refills, and other practical issues. However, we have excluded them from the present analysis due to limited variability in this study population.

3. Data analysis To assess the effects of MEMS assignment, we compare rates of dose omission for patients in the MEMS and non-MEMS groups using pill count data and the t-test. We then test the BMQ’s sensitivity or ability to detect true non-adherence, as measured by MEMS. Standard epidemiologic methods [18,35] are used to define and calculate ‘‘sensitivity’’, ‘‘specificity’’, ‘‘positive predictive value’’, and overall ‘‘accuracy’’ of the Regimen, Belief, and Recall Screens (Table 2). The Fisher’s exact test is used to analyze the relationship between categorical measures of adherence, and Pearson correlation techniques are used to analyze relationships between continuous measures of adherence.

B.L. Svarstad et al. / Patient Education and Counseling 37 (1999) 113 – 124

118

Table 2 Calculation of BMQ sensitivity, specificity, positive predictive value, and overall accuracy BMQ (self-report)

Objective criterion (MEMS) Dosage error present

Dosage error absent

Positive screen Negative screen

a c

b d

Sensitivity 5 a /(a 1 c) 3 100. Specificity 5 d /(b 1 d) 3 100. Positive predictive value (PPV) 5 a /(a 1 b) 3 100. Overall accuracy 5 (a 1 d) /(a 1 b 1 c 1 d) 3 100. Note: BMQ refers to Brief Medication Questionnaire; MEMS refers to Medication Events Monitoring System; a positive screen indicates patient reported some non-adherence or barrier to adherence and a negative screen indicates patient did not report any non-adherence or barrier to adherence.

4. Results

4.1. Patient characteristics Of 43 patients who completed the study, 21 had a standard vial and 22 had a MEMS container (Table 3). Sixty percent of study participants were male and

95% were white. Patient age ranged from 30 to 74 years (mean 5 52.6); education ranged from 8 to 20 years (mean 5 13.8); and number of scheduled medications ranged from 3 to 9 (mean 5 4.5). Patients had been taking their ACE inhibitor for an average of 25.4 months (range, 1 to 96). Twenty one patients were prescribed captopril and 22 were prescribed enalapril. Dosage regimens for ACE inhibitors ranged from once daily (n 5 23) to twice daily (n 5 20). According to the BMQ, 19% (n 5 8) of the patients had a concern or doubt when asked ‘‘How well is (this) medication working for you?’’, 19% (n 5 8) identified the ACE inhibitor when asked ‘‘Do any of your medications bother you in any way?’’, and 34.9% (n 5 15) reported at least one of these belief barriers. One of every four patients (n 5 11) also reported that it is at least somewhat hard to ‘‘remember to take all the pills’’. We compared the characteristics of MEMS and non-MEMS patients and found no significant differences by age, education, race, number of medications, type of ACE inhibitor, duration of treatment, or dosage regimen (P . 0.05). The MEMS group did have a higher proportion of male patients (P 5 0.05);

Table 3 Patient characteristics and adherence by MEMS assignment Total

Non-MEMS

MEMS

Number of cases

43

21

22

Patient background Age, mean years Sex, % male Education, mean years Race, % white Scheduled drugs, mean no. Drug type, % captopril Drug length, mean months Drug regimen, % multiple doses

52.6 60.5 13.8 95.3 4.5 48.8 25.4 46.5

51.1 42.9 14.0 95.2 4.6 47.6 29.6 47.6

54.0 77.3 a 13.6 95.5 4.5 50.0 21.4 45.4

BMQ and adherence measures Doubts how well drug works, % Believes drug is bothersome, % Reports $ 1 belief barriers, % Reports remembering problem, % % doses missed / week (BMQ) % doses missed / week (MEMS) % doses missed / month (pill count) % doses missed / month (MEMS)

18.6 18.6 34.9 25.6 4.1 – 10.7 –

4.8 23.8 28.6 23.8 2.6 – 10.8 –

31.8 13.6 40.9 27.3 5.5 19.2 10.5 13.0

a Chi-square test for non-MEMS versus MEMS group was significant at 0.05 level. No significant differences were found for other background, BMQ, or adherence measures.

B.L. Svarstad et al. / Patient Education and Counseling 37 (1999) 113 – 124

however, patient sex was unrelated to any of the subjective or objective adherence measures (P . 0.05). To determine the potential effects of MEMS assignment on adherence, we compared rates of dose omission using pill count data for the month following enrollment. We found no significant difference between the proportion of doses omitted in the two groups (0.105 and 0.108 in MEMS and non-MEMS group, respectively; F 5 0.0047; P . 0.05). Further analysis of MEMS patients showed a significant correlation between rate of dose omission as measured by MEMS and pill count (r 5 0.655; P 5 0.001); however, pill counts underestimated the rate of omission in comparison to MEMS (0.105 versus 0.131). We conclude that MEMS did not have a measurable effect on adherence and that it is a suitable method for testing the BMQ. We therefore restrict the following analysis to patients for whom MEMS data were available.

4.2. Prevalence of non-adherence according to MEMS Twenty patients provided usable MEMS data for the week prior to the home interview. Of these patients, 15 did not adhere fully to the prescribed regimen during the week prior to the interview. One-fourth of the patients had repeat non-adherence (took $ 20% more or less than prescribed); one-half had sporadic non-adherence (took 1–19% more or less than prescribed); and one-fourth had no nonadherence in the past week. Of those with nonadherence, all except one took less than prescribed. Overall, patients omitted 19.2% of the prescribed doses in the past week and 13.1% of the prescribed doses in the past month, suggesting that patients did not increase their drug-taking in the week prior to the BMQ interview.

4.3. BMQ’ s ability to predict repeat and sporadic non-adherence Fig. 1 shows the BMQ’s ability to predict any non-adherence in patients with repeat and sporadic non-adherence according to MEMS. We present results of the Regimen, Belief, and Recall Screens in Panels A–C, respectively. A ‘‘positive screen’’ indicates that the patient reported some non-adherence

119

(or barrier to adherence) in response to a given screen, whereas, a ‘‘negative screen’’ indicates that no non-adherence or barrier was reported. While none of the screens is adequate by itself, as a group they yielded better results than might be expected based on past studies. For example, a positive Regimen Screen was obtained in four of five patients with repeat non-adherence and none of the 15 remaining patients with sporadic or no nonadherence (Panel A). The Regimen Screen therefore had very good sensitivity (4 / 5, 80%), specificity (15 / 15, 100%), positive predictive value (4 / 4, 100%), and overall accuracy (19 / 20, 95%) when examining the type of non-adherence of greatest concern to researchers and clinicians. On the negative side, it failed to detect non-adherence in patients with sporadic non-adherence (0 / 10, 0% sensitivity). The 2-item Belief Screen also performed well. A positive Belief Screen was obtained for all patients with repeat non-adherence and 3 of 15 patients with sporadic or no non-adherence (Panel B). This resulted in a 100% sensitivity level (5 / 5), 80% specificity (12 / 15), 62% positive predictive value (5 / 8), and an overall accuracy of 85% (17 / 20) with respect to repeat non-adherence. Like the Regimen Screen, the Belief Screen failed to identify sporadic non-adherence: only 1 of 10 patients with this type of non-adherence was identified, yielding a sensitivity level of 10%. A different picture emerged when we tested the Recall Screen (Panel C). In contrast to the other screens, the Recall Screen had poor sensitivity (40%) when examining repeat non-adherence and good sensitivity for sporadic non-adherence. Of 10 patients with sporadic non-adherence, 9 had a recall barrier. This resulted in good sensitivity (90%), specificity (80%), and overall accuracy (85%) when examining this type of non-adherence. A summary of these findings and statistical results is provided in Table 4.

4.4. BMQ’ s ability to predict rates of dose omission in past week and month When we compared rates of dose omission, we found a strong correlation between the reported rate of omission using the Regimen Screen and the true rate of dose omission in the past week (r 5 0.67;

120

B.L. Svarstad et al. / Patient Education and Counseling 37 (1999) 113 – 124

Fig. 1. Comparison of BMQ and MEMS measures of adherence.

Table 4 Validity of BMQ by type of screen and type of non-adherence according to MEMS, n 5 20 Type of screen and type of non-adherence a

Regimen Screen Repeat non-adherence Sporadic non-adherence Belief Screen Repeat non-adherence Sporadic non-adherence Recall Screen Repeat non-adherence Sporadic non-adherence

Validity of BMQ screen b Sensitivity (%)

PPV (%)

Specificity (%)

Accuracy (%)

80.0 c 0.0

100.0 –

100.0 37.5

95.0 70.0

100.0 c 10.0

62.5 12.5

80.0 42.9

85.0 25.0

40.0 90.0 c

18.2 81.8

40.0 80.0

40.0 85.0

a Type of non-adherence in past week according to MEMS: repeat 5 took at least 20% over or under the prescribed amount; sporadic 5 took 1–19% over or under the prescribed amount. b See Table 2 for definition of sensitivity, positive predictive value (PPV), specificity, and accuracy. c Fisher’s exact test, P , 0.01 (1-tailed).

B.L. Svarstad et al. / Patient Education and Counseling 37 (1999) 113 – 124

P 5 0.001) and past month (r 5 0.89; P 5 0.001) according to MEMS. Findings also showed significant correlations between the total Belief Screen score and true rate of dose omission in the past week (r 5 0.71; P 5 0.001) and past month (r 5 0.61; , 0.01). In contrast, there was a slightly negative but non-significant relationship between the patient’s score on the Recall Screen and actual rate of dose omission in the past week (20.13; ns) and past month (20.08, ns). These findings are consistent with earlier results and reinforce the importance of using several tools and assessing different types of non-adherence.

5. Discussion Our results are consistent with previous studies which show that patients generally under report their non-adherence. However, it is clear that the sensitivity and overall accuracy of self-report measures can be improved by employing established principles of survey methodology. By applying these principles more rigorously, we obtained a Regimen Screen with a sensitivity level of 80%, a positive predictive value and specificity level of 100%, and an overall accuracy of 95%. These results compare favorably to published self-report adherence tools, which show an average sensitivity level of approximately 46% and an average accuracy level of 71% (Table 1). We also found strong correlations between the reported rate of omission in the past week and true rates of omission in the past week and past month (r 5 0.67 and r 5 0.89, respectively). These findings also compare favorably with past studies which show correlations between subjective and objective adherence measures ranging from 0.43 to 0.74 [14,22,30]. Our study is among the first to demonstrate that sensitivity levels vary by type of non-adherence and type of screening tool. We found that the Regimen and Belief Screens had good sensitivity when examining repeat non-adherence and poor sensitivity with regard to sporadic non-adherence, whereas, the Recall Screen had good sensitivity with regard to sporadic non-adherence and poor sensitivity with regard to repeat non-adherence. We suspect that sporadic non-adherence is under reported in the Regimen Screen, largely because this behavior is

121

more difficult to report due to its unintentional, infrequent, or erratic nature. Patients seem aware of their remembering difficulties, but unable to pinpoint when or how often they occur. In contrast, repeat non-adherence probably reflects deliberate changes in the regimen by patients who have unresolved concerns or doubts about the drug and how it is affecting them. While these patients may be reluctant to disclose the full extent of their non-adherence, many admit at least some non-adherence and concern about a particular drug regimen when asked carefully worded questions designed to reduce the different types of reporting errors that can occur in this context. Our results are encouraging, because they are based on a test population with multiple drugs and refill prescriptions, factors known to reduce the sensitivity of self-report adherence measures [18]. Larger studies in other settings clearly are needed to assess BMQ performance in patients with other characteristics. Further research also is needed to test the BMQ’s ability to predict future behavior and to determine how BMQ results are affected by variations in interviewer training and background, how and where the instrument is administered, number of administrations, modifications in question wording and order, interviewer reactions or attempts to change patient behavior and beliefs, and alternative methods of scoring. Any of these could influence interview or questionnaire results.

6. Practice implications Clinical studies are needed to evaluate the BMQ’s utility in various medical and pharmacy practice settings. However, we believe it can add important dimensions to adherence monitoring and enhance communication between patients and their care givers. First, it provides a clinically relevant and flexible method of screening non-adherence in patients with diverse drugs and drug regimens. When and how often it is administered depends on the patient population and how the information will be used. Second, it has excellent potential for self-administration by patients, thus providing an inexpensive and easy-to-use tool for screening adherence on a regular basis. Additional time, training, and reallocation of

122

B.L. Svarstad et al. / Patient Education and Counseling 37 (1999) 113 – 124

tasks often are necessary in any effort to improve communication between professionals and patients; however, busy practitioners should be able to test the utility of BMQ items with minimal effort. Third, the Belief and Recall Screens can help ‘‘diagnose’’ and potentially resolve different types of adherence barriers. Initially, the questions can be used as ‘‘screening’’ tools to identify, classify, and document patient concerns requiring further assessment and follow-up. Once specific concerns have been evaluated, appropriate interventions can be developed in consultation with patients and other care givers. Patients with recall barriers can be given simpler regimens, assistance in tailoring regimens to fit daily routines, special containers or memory aids, or other support designed to increase remembering. Patients with unwanted side effects or bothersome features can be referred for medical evaluation and counseling, evaluated as candidates for less bothersome medications, given appropriate reassurance, or offered other assistance in coping with bothersome aspects of therapy. Patients with concerns about efficacy can be given further counseling and feedback, greater involvement in evaluating treatment

goals and outcomes, assistance in using home monitoring devices, and / or supports known to enhance motivation and satisfaction with therapy. After interventions have been implemented, BMQ items can be used to evaluate outcomes. BMQ data also can be entered into various databases and used in continuous quality improvement programs. It is important to explore these different clinical applications if we are to gain a better understanding of the potential advantages and disadvantages of the BMQ and other self report tools for measuring and monitoring adherence problems and concerns.

Acknowledgements This research was supported in part by a grant from the National Corporation of Swedish Pharmacies. We thank all patients, pharmacists, and clinic staff who participated in the study and our colleagues and students who provided helpful suggestions. We especially thank Ingegard Agernas, Larry Boh, and Cynthia Raehl.

B.L. Svarstad et al. / Patient Education and Counseling 37 (1999) 113 – 124

123

124

B.L. Svarstad et al. / Patient Education and Counseling 37 (1999) 113 – 124

References [1] Cramer, JA, Spilker B, editors. Patient compliance in medical practice and clinical trials. New York: Raven Press, 1991. [2] Cramer JA. Compliance with contraceptives and other treatments. Obstetr Gynecol 1996;88:S4–S12. [3] Rudd P. Clinicians and patients with hypertension: Unsettled issues about compliance. Am Heart J 1995;130:572–9. [4] Urquhart J. Patient compliance as an explanatory variable in four selected cardiovascular studies. In: Cramer JA, Spilker B, editors. Patient compliance in medical practice and clinical trials. New York: Raven Press, 1991:301–22. [5] Didlake RH, Dreyfus K, Kermman RH, et al. Patient noncompliance: a major cause of late graft failure in cyclosporine-treated renal transplants. Transplant Proc 1988;20:63– 9. [6] Snodgrass W, Smith S, Trueworthy R, et al. Pediatric clinical pharmacology of 6-mercaptopurine:lack of compliance as a factor in leukemia relapse. Proc Am Soc Clin Oncol 1984;3:204. [7] Priddy JT, Kass MA, Gordon MO, et al. Factors related to compliance with topical pilocarpine treatment. Invest Ophthalmol Visual Sci 1987;28:37. [8] Rand CS, Wise RA. Measuring adherence in asthma medication regimens. Am J Resp Crit Care Med 1994;149:S69–76. [9] Meichenbaum D, Turk E. Facilitating treatment adherence: a practitioner’s guide. New York: Plenum Press, 1987. [10] Svarstad BL. Physician–patient communication and patient conformity with Medical Advice. In: Mechanic, D editor. The growth of bureaucratic medicine. New York: Wiley, 1976:220–238. [11] Svarstad B. Patient–practitioner relationships and compliance with prescribed medical regimens. In: Aiken L, Mechanic D, editors. Applications of social science to clinical medicine and health policy. New Brunswick, NJ: Rutgers University Press, 1986:438–459. [12] Roter DL, Hall JA, Merisca R, Ruehle B, Cretin D, Svarstad B. Effectiveness of interventions to improve patient compliance: A meta-analysis. Med Care 1998 (in press). [13] Gilbert JR, Evans CE, Haynes RB, Tugwell P. Predicting compliance with a regimen of digoxin therapy in family practice. Can Med Assoc J 1980;123:119–22. [14] Haynes RB, Taylor DW, Sackett DL, et al. Can simple clinical measurements detect patient compliance?. Hypertension 1980;2:757–64. [15] Inui TS, Carter WB, Pecoraro RE. Screening for non-compliance among patients with hypertension: Is self-report the best available measure?. Med Care 1981;19:1061–4. [16] Cramer JA, Mattson RH, Prevey ML, Scheyer RD, Ouellette VL. How often is medication taken as prescribed?. J Am Med Assoc 1989;26:3273–728. [17] Kontz MM. Compliance redefined and implications for home care. Holistic Nurs Pract 1989;3:54–64.

[18] Stewart MS. The validity of an interview to assess a patient’s drug taking. Am J Prev Med 1987;3:95–100. [19] Rudd P, Ahmed S, Zachary V, et al. Improved compliance measures: Applications in an ambulatory hypertensive drug trial. Clin Pharmacol Ther 1990;48:767–85. [20] Rittenhouse BA. A novel compliance assessment technique: The randomized response interview. Int J Technol Assess Health Care 1996;12:498–510. [21] Roth HP, Caron HS. Accuracy of doctors’ estimates and patients’ statements on adherence to a drug regimen. Clin Pharmacol Ther 1978;23:361–70. [22] Morisky DE, Green LW, Levine DM. Concurrent and predictive validity of a self-reported measure of medication adherence. Medical Care 1986;24:67–73. [23] Myers ED, Calvert EJ. Information, compliance and sideeffects: A study of patients on antidepressant medication. Br J Clin Pharmac 1984;17:21–5. [24] Pullar T, Kumar S, Tindall H, Feely M. Time to stop counting the tablets?. Clin Pharmacol Ther 1989;46:163–8. [25] Enlund H, Tuomilehto J, Turakka H. Patient report validated against prescription record for measuring use of and compliance with antihypertensive drugs. Acta Med Scand 1981;209:271–5. [26] Steiner JF, Koepsell TD, Fihn SD, Inui TS. A general method of compliance assessment using centralized pharmacy records. Med Care 1988;26:814–23. [27] Steiner JF, Prochaska AV. The assessment of refill compliance using pharmacy records: Methods, validity, and validation. J Clin Epidem 1997;50:105–16. [28] Sclar DA, Skaer TL, Robison LM, et al. Antihypertensive pharmacotherapy: Economic outcomes in a health maintenance organization. Curr Ther Res 1994;55:1056–66. [29] Gordis L, Markowitz M, Lilienfeld AM. The inaccuracy in using interviews to estimate patient reliability in taking medications at home. Med Care 1969;7:49–54. [30] Craig HM. Accuracy of indirect measures of medication compliance in hypertension. Res Nurs Health 1985;8:61–6. [31] Urquhart J. Role of patient compliance in clinical pharmacokinetics. A review of recent research Clin Pharmacokinet 1994;27:202–15. [32] Jimenez VA, Ballestero AG, Martinez VP, et al. Descriptive study of patient compliance in pharmacologic antihypertensive treatment and validation of the Morisky and Green test [Spanish]. Atencion Primaria 1991;10:767–70. [33] Park LC, Lipman RS. A comparison of patient dosage deviation reports with pill counts. Psychopharmacologia 1964;6:209–302. [34] Sudman S, Bradburn NM. Asking questions: a practical guide to questionnaire design. San Francisco: Jossey-Bass, 1982. [35] Ibrahim MA. Epidemiology and health policy. Chapel Hill: Aspen Systems Corporation, 1985.