cc

[

V A L U E I N H E A L T H 1 4 ( 2 0 1 1 ) 1 0 9 2 – 1 1 0 0

Avai lable onl ine at www.sc iencedirect .com

journal homepage: www.elsevier .com/ locate / jva l

Toward Identifying the Causes and Combinations of Causes Increasingthe Risks of Nonadherence to Medical Regimens: Combined Results ofTwo German Self-Report SurveysThomas Wilke, PhD1,*, Sabrina Müller, BS1, Donald E. Morisky, ScD, MSPH, ScM2

1Institut für Pharmakoökonomie und Arzneimittellogistik, Hochschule Wismar, Wismar, Germany; 2Department of Community Health Sciences, UCLA School

of Public Health, Los Angeles, CA, USA

A B S T R A C T

decicdChNtKa

C

Objectives: This study aimed to identify the causes of the nonadher-ence (NA) of German patients to their prescribed medication. In theourse of the investigation, the NA risk profiles resulting from theombination of the various causes were identified. Methods: Two

cross-sectional surveys with a total of 1517 patients (comprising 1177 pa-tients contacted by telephone and forming survey 1 and a different set of340 patients interviewed in-depth and face-to-face forming survey 2) wereconducted. Self-reported NA was measured by the generic Morisky Med-ication Adherence Scale (MMAS). Survey 1 used a four-item MMAS andSurvey 2 an eight-item MMAS. Results: Approximately 35% to 40% of thepatients can be described as nonadherent. In survey 1, a few causes ex-plain the NA (chronic disease, younger age, and fewer medications re-quired to be taken). The more detailed survey 2 shows that the existenceof intentional NA has considerably more influence than any other causal

factors. Positive medication beliefs, a positive mood, and a good patient- O

●

●

●

oöko

al So

doi:10.1016/j.jval.2011.06.002

octor relationship reduce the NA risk. Furthermore, patients who areasily able to recognize the correct medication, as evidenced by ability toorrectly identify the packaging, have a reduced NA probability. Concern-ng additive risk, patients who are chronically ill but display no otherauses of risk have an NA probability of 10.4%. By contrast, in patientsisplaying all the identified causes of risk, the rate increases to 93.9%.onclusions: About one-third of patients can be classified as nonad-erent. Intentional/medication-based NA causal factors explain theA considerably better than do socioeconomics. The existence of more

han one cause of risk considerably increases the NA risk of a patient.eywords: nonadherence, noncompliance, nonadherence/noncompli-nce causes.

opyright © 2011, International Society for Pharmacoeconomics and

utcomes Research (ISPOR). Published by Elsevier Inc.

Introduction

Medication-based nonadherence (NA) in chronic indications canbe defined as the extent to which a person’s drug-taking behaviordoes not correspond to instructions from a health-care provider[1–3]. Because approximately 30% to 50% of patients are affectedby it [2,4,5], NA is a considerable challenge. NA significantly influ-ences clinical (surrogate) parameters [3,6] and long-term mortality7] and increases health-care costs [8,9]. By contrast, there exists

often sobering evidence of the lack of efficacy of intervention pro-grams. Only 50% of such programs/interventions are capable ofsignificantly reducing the NA of patients, and only a third of theinterventions improve clinical surrogate parameters [4,10]. Thereare multiple causes of this state of affairs. However, one of themost important reasons for the lack of effectiveness is that only avery few adherence programs are based on a careful audit of thepatient-specific causes for NA (adherence barriers). If NA causeswere known in detail, they could be used, possibly even on anindividual or a patient cluster basis, to develop specific barrier-reducing measures [10–14]. Detailed knowledge of the relevantcauses of NA is therefore an important precondition for increasingthe effectiveness of adherence programs.

* Address correspondence to: Thomas Wilke, Institut für Pharmakof Applied Sciences), PF 1210, 23952 Wismar, Germany.

E-mail: thomas.wilke@hs-wismar.de.1098-3015/$36.00 – see front matter Copyright © 2011, Internation

Published by Elsevier Inc.

A variety of factors are discussed as potential causes of NA [5].More recent classification methods of NA causes distinguish inparticular intentional and unintentional factors [10,15,16]. In themain, we follow this approach, which distinguishes five groups ofcausal factors (Fig. 1). The evidence to date concerning the impor-tance of these five factors in explaining NA can be summarized asfollows:

● Within the medication-based factors, the complexity of medi-cation regimens [17–19] and the possibility of side-effects[11,20] are the main contributors.Direct and indirect medication-related costs that patients haveto bear (e.g., copayments, waiting times, long journeys to reachthe doctor) are the most important health care–related factors[21–24].Sociodemographic factors play only a small part [25,26]. In ad-dition, particularly with regard to aging, there is not a consis-tent direction of influence on NA [10].Factors leading to unintentional NA have to date only beensporadically investigated. Factors such as depression [27,28],dementia, or increased forgetfulness or carelessness [29–31], or

nomie und Arzneimittellogistik, University of Wismar (University

ciety for Pharmacoeconomics and Outcomes Research (ISPOR).

●

noatatcctorlpw[amtptsfvscs

ii

lmfpti

1093V A L U E I N H E A L T H 1 4 ( 2 0 1 1 ) 1 0 9 2 – 1 1 0 0

a lower than average level of health literacy [32] increase boththe extent and probability of the occurrence of NA.Recently, there has been much more attention paid to factorsin the context of intentional NA. Even though many questionsremain unresolved, it is apparent that these factors are capableof explaining a considerable part of the NA [10]. In this regard,negative general attitudes and negative medication beliefs[33,34] as well as inadequate coping behavior [35–38] and a poorrelationship with health-care providers [26,39,40] are the mostimportant.

Previous research concerning NA causes has two central weak-esses: 1) Multivariate analysis is used only rarely [10,41]. On thene hand, this applies to numerous articles written on the basis ofdministrative/claims-based data [42]. Due to the data situation,hese are only as a rule able to measure the sociodemographic andfew of the indication- and medicine-based NA causes [43,44]. On

he other hand, even studies that have examined further NAauses only examine a few or only one specific factor [4,10], typi-ally, medication-related factors such as the number of medica-ions to be taken daily were analyzed in considerable detail with-ut consideration of other potential NA causes [45]. Therefore, theeciprocal effects between the factors and their additive effects areargely unknown. 2) Numerous publications concerning NA, andarticularly survey-based investigations, have used methodicallyeak NA measurement instruments that have not been validated

46,47]. Quite regularly, a single simple direct question failing topply even rudimentary techniques of psychometric analysis andeasurement development is used instead of a validated and de-

ailed self-report to measure NA [48–51]. Self-report instrumentser se have their limitations, such as respondent bias and errors inhe recall of behavior [48]. In addition, in research designs usingimple direct questions about medication that has possibly beenorgotten, there is the potential for activating what the inter-iewee sees as socially desirable answer behavior [49,50]. Moreophisticated NA self-report questionnaires use indirect, nonac-usatory questioning techniques to minimize the tendency to give

Fig. 1 – Illustration shows important nonadherence causes dgroups.

ocially desired answers. Therefore, validating NA measurement

nstruments before using them or using only previously validatednstruments is strongly recommend.

In the context of the research into NA causes and its current chal-enges detailed previously, this study reports the results of two Ger-

an surveys (a telephone survey and a subsequent in-depth face-to-ace survey). It aimed to identify the factors that can be used toredict the NA of German patients to their prescribed medicationherapy. In the course of the investigation, the NA risk profiles result-ng from the combination of the various causes of risk are identified.

Methods

This article reports the results of two cross-sectional surveys thatincluded a total of 1577 patients. In the first survey, 1177 patientswere interviewed by a structured telephone interview. This surveywas originally conducted to assess the extent of NA in outpatientthromboprophylaxis after major orthopedic surgery [34]. The in-terviewed patients were drawn from 22 German acute or rehabil-itation orthopedic clinics (informed consent). The potential sam-ple included all the patients in these clinics who were scheduledfor or had already had hip or knee arthroplasty and who were ableto answer questions in a telephone interview (consecutive recruit-ment) 40 to 45 days after surgery. The initial sample size (1771patients) was reduced because 16% (276 patients) refused to givetheir informed consent or could not be interviewed because oftheir poor health status. Consequently, interviews were per-formed with 1495 patients. The data gathered from 78.7% of thesepatients (1177 patients) were included in a further analysis be-cause the patients confirmed that they were taking regular, atleast daily, medication. Two focus groups, one with five membersand one with four members, were used to pretest the question-naires in terms of comprehensibility and practicability. Somequestions were found to be difficult to understand in a telephoneinterview and were therefore rewritten. The interviews were con-ducted by trained interviewers, each of whom had previously con-ducted three to five test interviews under supervision. The super-

ssed in the literature, and their classification into five

iscu

vised interviews were audio recorded and subsequently evaluated

iLSULmlpas

apmot

otdtita5tTawtf

●

●

●

●

Tnitatttmbdi5

●

●

●

●

●

●

●

Tbvt

gdiithr

dgtpa

1094 V A L U E I N H E A L T H 1 4 ( 2 0 1 1 ) 1 0 9 2 – 1 1 0 0

in the presence of the interviewers to give performance feedbackand increase the learning effects. The interviewers had to learnthe content of the questionnaire before conducting an interviewand were educated in the normal treatment of a patient who hadundergone major orthopedic surgery before release from the acutehospital. The interviewers did not have medical degrees becausethe level of medical knowledge necessary to conduct the inter-views was not advanced. The data were collected in real time in aMicrosoft Access databank. The real-time collection was based onan online tool developed for this purpose, which included pre-defined data entry fields to reduce the risk of data entry errors.Furthermore, an audio recording of each interview was also madeto provide help should discrepancies later arise. During the study,these recordings were also used to supervise the interviewers. Theaverage interview was conducted on day 41 after orthopedic sur-gery and lasted approximately 10 to 15 minutes. The question-naire consisted of 30 questions.

The second in-depth face-to-face survey was conducted ex-plicitly to promote the aims of this study, and its main focus was tocollect additional information on potential NA causes. This surveywas conducted with 340 pharmacy customers in 17 German phar-macies, again using pretested questionnaires. The pretest wasconducted with 20 patients in one pharmacy. The patients whotook part in the second survey were not the same ones who tookpart in the first survey. As in the first survey, only patients onregular, at least daily, medication were included in the survey. Theinterviews were conducted by trained interviewers without anymedical degrees. Again, the interviewers first conducted three tofive test interviews under supervision. All the interviews were su-pervised, this time, however, by means of the completed question-naires instead of audio recording. The collected data was takendown in writing and later transferred to a Microsoft Access data-bank. Double data entry was conducted, to minimize errors. Anindependent supervisor checked all the data entries to control forconsistency and to identify unexpected values. The people se-lected to be interviewed were average patients visiting a Germanpharmacy. This was achieved by each interviewer being given anage and sex distribution ratio identical to the average distributionof these characteristics at the particular pharmacy and told tofollow it in selecting interviewees. After giving their informed con-sent, the patients were interviewed in the participating pharmacy,either immediately or at a later date of their choice. Each interviewlasted 20 to 30 minutes and was based on a questionnaire with 26questions. All the variables (surveys 1 and 2) included in this anal-ysis are listed in Table 1.

In each of the surveys, the general medication-based adher-ence of the interviewees was measured by a self-report. This wasdone without considering particular indications or medications. Insurvey 1, the NA measurement instrument used was the four-itemMorisky Medication Adherence Scale (MMAS-4) [52], and in survey2, the eight-item MMAS (MMAS-8) [6] was used. (Use of the MMASs protected by US copyright laws. Permission for use is required.icensure agreement is available from Donald E. Morisky, ScD,cM, MSPH, Professor, Department of Community Health Services,CLA School of Public Health, 650 Charles E. Young Drive South,os Angeles, CA 90095-1772.) Both indices are validated measure-ent instruments in current use. The existence of NA was estab-

ished by the scoring logic of the MMAS used. In survey 1, everyatient who reached an index score of �4 was classified as non-dherent [52]. In survey 2, every patient who reached an indexcore of �6 was classified as nonadherent [6].

Table 1 gives an overview of sociodemographic characteristicss well as additional information obtained during the surveys andotentially capable of explaining the occurrence of NA. The infor-ation gathered in survey 1 is suitable for analyzing the influence

f sociodemographic factors on NA. By contrast, survey 2 allows

he detailed evaluation of the influence of other causal factors m

n possibly intentional/medication-based NA (Fig. 1). The sys-ematic recording of these factors was conducted using vali-ated measurement instruments. However, due to the limitedime available for the interviews, these instruments were usedn a shortened/summarized form: the medication beliefs sec-ion was reduced to four questions, a selection from the Beliefsbout Medicines Questionnaire (BMQ), an instrument using-point Likert scales [33]. In this study, only the general part ofhe German validated version of the BMQ [53] was considered.his part consists of eight questions that reflect general harmnd general overuse scales. Because the focus of our interestas on the general harm context, we chose the four questions in

he BMQ that had in the validation study attained the highestactor load for the general harm scale [33]:

Medicines do more harm than good.People who take medicines should stop their treatment for awhile every now and again.Most medicines are addictive.All medicines are poisons.

he doctor-patient relationship was examined using a question-aire from DiMatteo [54]. This questionnaire has not yet been val-

dated in Germany, so after translation into German, it was back-ranslated and the German version adjusted until a match waschieved. The emotional state of the interviewees was deduced byhe Self-Assessment Manikin (SAM) concept [55]. The SAM ques-ionnaire consists of three panels (valence, arousal, domains). Inhis survey, only the first panel was used because only the general

ood was to be measured. The state and level of the relationshipetween customers and pharmacists as well as possible practicalifficulties experienced by patients with medication or its packag-

ng were documented by using the following questions (4- or-point Likert scales):

Relationship between customers and pharmacists

I believe that my pharmacist helps me.I believe that my pharmacist understands me.I think that my pharmacist gives clear instructions.

Difficulties experienced by patients with medication/packaging

How difficult is it to distinguish the medicines on the basis oftheir packaging?How difficult is it to find how much of the medicine to take?How difficult is it to open the package?How difficult is it to read the leaflet?

hese last questions were developed specifically for this studyecause as far as the authors know, there was not a preexistingalid and practicable instrument available to assess these poten-ial NA causes.

The data were subjected to multivariate analysis (logistic re-ression) using NA as the dependent variable. To test for depen-encies due to clustering effects (22 centers in survey 1, 17 centers

n survey 2), each regression estimate was redone using general-zed estimating equation (GEE) methods [56]. The differences be-ween the logistic regression estimates and the GEE estimates,owever, were very small so we chose to report only the logisticegression estimate results.

In principle, all the documented information was included as in-ependent variables in the first model calculations. Multivariate re-ression analysis was carried out using backward stepwise elimina-ion to determine the most parsimonious factors associated with NArobabilities. In the backward stepwise iterations, regression termst a significance level of P � 0.10 were successively eliminated. All

odels, however, were adjusted for sex, living circumstances, and

1095V A L U E I N H E A L T H 1 4 ( 2 0 1 1 ) 1 0 9 2 – 1 1 0 0

Table 1 – Collected variables and descriptive statistics.

Variable Survey 1 Survey 2

N 1177 340Mean age in years Ø 68.5 (range 21–94; SD 9.002) Ø 55.8 (range 17–90; SD 17.042)Sex, no. (%)

Female 720 (61.2) 218 (64.1)Male 457 (38.8) 122 (35.9)

Chronic diseases (more than one answer possible), no. (%)None 436 (37.0) 108 (31.8)Hypertension 365 (31.0) 119 (35.0)Diabetes (both types) 152 (12.9) 51 (15.0)Rheumatism 37 (3.1) 11 (3.2)Asthma/allergy 35 (3.0) 44 (12.9)Other 294 (25.0) 138 (40.6)

Living circumstances, no. (%)Living alone 270 (22.9) 93 (27.4)Living with spouse/life partner 794 (67.5) 247 (72.6)Living with children 84 (7.1) —Other circumstances 29 (2.5) —

Highest achieved education level, no. (%)Without any formal apprenticeship 160 (13.6) 35 (10.3)Completed apprenticeship 806 (68.5) 225 (66.2)University degree 152 (12.9) 80 (23.5)Other degrees 59 (5.0) —

Current working status, no. (%)Blue-collar worker — 16 (4.7)White-collar worker — 107 (31.5)Entrepreneur — 24 (7.0)Pensioner — 146 (43.0)Unemployed — 12 (3.5)Other — 35 (10.3)

Emotional state of respondents at the time of the interview(based on Manikin scale)

Very good — 82 (24.1)Good — 158 (46.5)Moderate — 80 (23.5)Bad — 16 (4.7)Very bad — 4 (1.2)Information not collected — —

No. of different drugs taken per dayMean no. 4.1 (range 1–16, SD 2.671) —

No. of prescribed medicationsMean no. — 3.1 (range 1–15, SD 2.456)

Dosing frequency prescribed medicationsMore than once daily — 150 (47.0)Once daily — 155 (48.6)At least once a week — 10 (3.1)Less than once a week — 4 (1.3)

No. of OTC medications takenMean no. of drugs taken per day — 2.3 (range 1–8, SD 1.407)

Dosing frequency OTC medicationsMore than one time daily — 29 (11.3)Once daily — 97 (37.9)At least once a week — 41 (16.0)Less than once a week — 89 (34.8)

MMAS-4 score1 13 (1.1) —2 101 (8.6) —3 343 (29.1) —4 720 (61.2) —

(continued on next page)

1096 V A L U E I N H E A L T H 1 4 ( 2 0 1 1 ) 1 0 9 2 – 1 1 0 0

education. If any data were missing from the set for a particularinterviewee (listwise deletion), the entire set was discarded. As a con-sequence, the data from eight patients were excluded from the mul-tivariate analysis of survey 2. A significance level of 0.05 was chosenfor analysis.

Survey 2 covers many potential risks of NA. To quantify the addi-tive risk that may occur in the presence of several of these risks, in

Table 1 (continued).

Variable

MMAS-8High adherence (score � 8)Moderate adherence (score 6–7)Low adherence (score �6)

Doctor-patient relationship (patient’s perspective)Physician takes the lead without previous discussion with patientPhysician takes the lead as previously discussed with the patientPhysician and patient act as equal partnersInformation not collected

Pharmacy-patient relationship (patient’s perspective index: 3 �good relationship, 12 � bad relationship)

3–67–910–12Information not collected

Medication (beliefs index: 4 � positive beliefs,20 � negative beliefs)

4–89–1213–1617–20Information not collected

Difficulties with medication packagesIdentification of different drugs

Very difficultDifficultModerateEasyVery easyInformation not collected

Reading medication dosageVery difficultDifficultModerateEasyVery easyInformation not collected

Opening packagesVery difficultDifficultModerateEasyVery easyInformation not collected

Reading leafletsVery difficultDifficultModerateEasyVery easyInformation not collected

The table shows all variables that were included in the data set andMMAS-4, four-item Morisky Medication Adherence Scale; MMAS-8, eaverage.

whatever combination, all the significant independent variables (P �

0.05) from the multivariate regression model from survey 2 were di-chotomized. Subsequently, the estimation of the occurrence of NAwas repeated on the basis of the resulting dichotomized explanatoryparameters and for every interviewee. The results of the model wereused to predict the probability of each of the interviewees being non-adherent. On the basis of these probabilities, the average risk of theoccurrence of NA was estimated for each group/each combination of

Survey 1 Survey 2

— 97 (28.5)— 123 (36.2)— 120 (35.3)

— 34 (10.0)— 91 (26.8)— 212 (62.3)— 3 (0.9)

— 325 (95.6)— 11 (3.2)— 2 (0.6)— 2 (0.6)

— 123 (36.2)— 90 (55.9)— 24 (7.0)— 0 (0.0)— 3 (0.9)

— 8 (2.4)— 33 (9.7)— 63 (18.5)— 131 (38.5)— 101 (29.7)— 4 (1.2)

— 13 (3.8)— 67 (19.7)— 73 (21.5)— 117 (34.4)— 66 (19.4)— 4 (1.2)

— 7 (2.0)— 27 (7.9)— 83 (24.4)— 110 (32.4)— 109 (32.1)— 4 (1.2)

— 61 (17.9)— 101 (29.7)— 84 (24.7)— 61 (17.9)— 29 (8.6)— 4 (1.2)

analyzed to identify NA causes.tem Morisky Medication Adherence Scale; OTC, over-the-counter; Ø,

wereight-i

the independent risk factors.

swctc

ai6�

4

ieaop

1a

ttmcpnmaapdtttass

1097V A L U E I N H E A L T H 1 4 ( 2 0 1 1 ) 1 0 9 2 – 1 1 0 0

Results

Table 1 shows the interviewees in survey 1 to be a relatively oldergroup (average age in survey 1 was 68.5 years compared with 55.8years in survey 2, P � 0.001). This is explained by the fact thaturvey 1 was concerned with average orthopedic patients,hereas survey 2 was concerned with average German pharmacy

ustomers. Both groups of interviewees are characterized by ex-ensive dependency on medication as well as the occurrence ofhronic disease (63.0% in survey 1 and 68.2% in survey 2).

In the first survey, the mean MMAS-4 score was 3.50 (SD 0.699)nd 38.8% of the patients reached an index score of �4 (values �4ndicate NA). In the second survey, the mean MMAS-8 score was.14 (SD 1.954) and 35.3% of the patients reached an MMAS score of6 (values �6 indicate NA). Both of these analyses revealed 35% to0% of the patients to be affected by NA.

The results of the multivariate analysis of potential NA causesn survey 1 makes it clear that only a few factors are capable ofxplaining the probability of NA in patients (Table 2). These factorsre the occurrence of chronic disease, relative youth, and havingnly a few medications to take. Consequently, the explanatoryower of the model developed here is limited (Nagelkerke R2 is

0.061). Education level, sex, and living circumstances of the patientare not significant predictors of NA (P � 0.100). These parametersremained in the estimate as control variables, but are not shown inTable 2 as relevant NA causes.

The multivariate analysis conducted on the basis of the data insurvey 2 explains a much larger part of the self-reported NA than themodel based on survey 1; R2 is almost six times higher than in survey. The results of the first analysis regarding the role of chronic diseasend age are replicated (Table 2). By contrast, the number of medica-

Table 2 – Multivariate logistic regression estimates of facto

Age in yearsExistence of chronic disease

NoYes

No. of medications to be takenMood (5-point Likert scale: 1 � very good; 5 � very bad)Patient-doctor-relationship (1 � physician and patient act as equal

partners; 3 � physician takes the lead without previousdiscussion with patient)

Patient-pharmacy relationship: “Pharmacy gives effective/understandable explanations” (4-point Likert scale: 1 � fullyagree; 4 � fully disagree)

Negative medication beliefs: “People who take medicines shouldstop their treatment for a while every now and again” (5-pointLikert scale: 1 � fully disagree; 5 � fully agree)

Difficulties with medication packages: identification of differentdrugs (5-point Likert scale: 1 � no difficulties; 5 � majordifficulties)

Table shows results of logistic regression estimates.CI, confidence interval; OR, odds ratio.* A backward elimination method was performed including all varia

models.† Eight patients without complete data for all relevant independent‡ Derived from the logistic models are adjusted for sex, living circum

§ Reference group.

ions that have to be taken has no independent influence. Causeshat can be classified as part of the context of intentional NA or are

edication related, however, are both important and highly signifi-ant. These causes are negative medication beliefs (proposition: peo-le who take medicines should stop their treatment for a while everyow and again; P � 0.001), a negative mood or fundamental pessi-ism (P � 0.005) as well as a low level of trust in doctors (P � 0.001),

nd all increase the probability of a person classifying him- or herselfs nonadherent. A factor that as far as the authors know has notreviously been investigated in detail can be added to the list: theifficulties experienced by people in distinguishing between medica-ions and medication packaging and understanding the contents ofhe packaging. The results show that if people find it difficult to iden-ify their required medication by means of the packaging, they havesignificantly higher probability (odds ratio 1.708, P � 0.001) of clas-

ifying themselves as nonadherent. The following variables did notignificantly explain the measured NA of respondents (P � 0.100):

● Current working status● Number of different drugs taken per day● Belief that medicines do more damage than good● Belief that most medicines are addictive● Belief that natural remedies are safer than medicines● Difficulty reading medication dose from medication packages● Difficulty opening medication packages● Difficulty reading leaflets

As in survey 1, education level, sex, and living circumstances of thepatient are not significant predictors of NA (P � 0.100) Again, theseparameters remained in the estimate as control variables, but are notshown in Table 2 as relevant NA causes. When the explanatory vari-

plaining nonadherence.

del summary* (N � 1177):survey 1

Model summary* (N � 332†):survey 2

Nagelkerke R2

0.061Nagelkerke R2

0.352

OR (95% CI)‡ P value No. OR (95% CI)‡ P value

0.965 (0.951–0.978) �0.001 332 0.980 (0.963–0.997) 0.023

1.000§ 0.031 105 1.000§ 0.0371.333 (1.027–1.731) 227 1.952 (1.042–3.657)0.920 (0.876–0.967) 0.001

332 1.601 (1.157–2.217) 0.005332 2.027 (1.336–3.074) �0.001

332 1.530 (0.923–2.538) 0.099

332 2.195 (1.519–3.171) �0.001

332 1.708 (1.299–2.244) �0.001

hown in Table 1. Variables with P value � 0.10 remained in the final

bles were excluded from the analysis.

ces, and education.

rs ex

Mo

No.

1177

436741

1177

bles s

varia

stan

1098 V A L U E I N H E A L T H 1 4 ( 2 0 1 1 ) 1 0 9 2 – 1 1 0 0

ables derived from survey 2 are dichotomized, the factors identifiedas significant and independent contribute cumulatively to the NArisk (Fig. 2). Only 2.0% of the patients did not exhibit any of the sixidentified causes. For 75.5%, more than one causal factor was impor-tant, whereas for 22.5%, only one was important. It follows that mul-ticausality is the rule and not the exception. In the case of interview-ees who are merely chronically ill without having additional causesof risk, their NA probability is on average 10.4%. The combination ofchronic illness with any of the remaining five factors increases theprobability of NA to at least 19.3% or to a maximum of 29.9%. In oursample, in which an interviewee displayed all the identified NAcauses, the NA probability increased to 93.9% (Fig. 2).

Discussion

This study began with the intention of identifying the factors thatcan be used to predict the NA of German patients to their pre-scribed medication therapy. In the course of the investigation, theNA risk profiles resulting from the combination of the variouscauses of risk were identified. As far as the authors are aware, oursis the largest study of medication-based NA and its causes con-ducted in Germany. The results show that on the basis of self-reports, approximately a third of patients can be considered to be

Fig. 2 – Illustration (on the basis of the multivariate causal avarious risk factors (stand-alone as well as in combination).combination class (based on the combination of NA causes iadditional NA risk related to the added causal factor.

nonadherent. Sociodemographics alone do not explain the NA as

well as factors that can be thought of as part of the context ofintentional/medication-related NA.

The authors acknowledge some limitations of the study. First,neither of the surveys is representative of the average chronicallyill German patient. Nevertheless, the analyzed groups are samplescharacterized by a high average rate of medication dependenceand occurrence of chronic disease. Because neither sex nor multi-ple morbidities represent significant explanations of the occur-rence of NA, the samples analyzed here do not appear to displaybias. Only the relative youth of the participants in survey 2 is ca-pable of causing an overestimation of the extent of NA in thisdatabase. Second, due to the limited time available for each inter-view that arose on organizational grounds and the design of thequestionnaire, it was not possible to record data concerning alltheoretically possible NA causes. The variables not recorded con-cern the health beliefs of the interviewees [57,58], the approach todisease/coping behavior [35], the influence of the health literacy[32], patients’ copayments [59,60], and specific indication/medica-tion factors including the side effects of particular medicines. Inaddition, neither survey was used as a basis to analyze any factorsdescribing explicit unintentional NA causes and therefore objec-tive adherence barriers. On the other hand, survey 2, in particular,even in its existing form, is a database for the explanation of NA of

sis in survey 2) of the nonadherence (NA) connected tot gray areas indicate the base case in each riske previous class), whereas dark gray areas indicate the

nalyLighn th

a size and form that is internationally relatively rare. Third, the

1099V A L U E I N H E A L T H 1 4 ( 2 0 1 1 ) 1 0 9 2 – 1 1 0 0

reliance on self-reports is one of the weaknesses of our study.The available evidence shows that self-reporting underesti-mates the extent of NA [61– 64]. In this regard, the inclusion ofother indicators such as administrative data or Medication EventMonitoring System/pill counts would considerably increase thevalidity of the NA measurement [34,65,66]. Limited data availabil-ity, however, is the challenge here. Nevertheless, the self-report-ing measures used are good and validated instruments. Both sur-veys used scales that have been validated (discriminant validity)with regard to social desirability [6,52]. Fourth, this investigationonly concerned medication-based NA. Health outcomes andhealth-care costs may both be influenced by other elements of thehealth-care provision system. Our study describes only one, al-though important, aspect of health-care provision.

The results presented here are compatible with other findingsin the literature. The extent of self-reported NA is comparable tothat found in other studies [2,10,44,67]. The results of our causalanalysis provide confirmation of the relevance of the direction ofthe more recent NA research, which emphasizes intentional/med-ication-related NA [10,30,68,69]. Our study confirms that severalintentional factors exert independent influence on the NA proba-bility. Survey 2 shows that in cases in which one patient displaysall the factors in combination, the result is almost certainly NA.Interestingly, the following factors that are discussed as NA predic-tors in the literature do not explain any NA in our samples: the sex oreducation of patients, number of medications taken (survey 2), occu-pational status, specific chronic diseases, or relationship to pharma-cists/pharmacies. Evidently, intentional factors found in our sampleto be NA predictors exert such a dominant influence on NA that thereremains no independent influence of those other factors.

The results of our study provide important starting points forfuture research as well as for the further development of adherenceprograms. Future research into NA causes should focus on three is-sues: 1) whether the interpretation of different NA causes dependson specific care or patient-related factors (patient segmentation); 2)what the degree of interaction with other NA explanations is; and 3)whether important factors explaining NA change over the course oftime and are thereby the cause of the well-known phenomenon ofdecreasing persistence of patients after initial treatment [41,70].

The clinical and health economics effectiveness of adherenceprograms depends in large part on the extent to which importantadherence barriers can be addressed. Our results show that causes ofintentional/medication-related NA occupy a key position in this con-text. Reducing intentional NA requires thorough and in-depthknowledge about patients’ views and preferences as well as exten-sive knowledge about how interactions between patients and otherhealth-care providers influence possible NA [4]. Most of the adher-ence interventions used/tested so far, however, fail to address thecauses identified in our surveys as the most important reasons forNA [4,71]. The promotion of positive medication beliefs, a positiveview of illness and therapy, an independent measurement and opti-mization of the doctor-patient relationship, and the optimization ofways of dealing with the medication packaging appear to be ele-ments that are essential for adherence intervention effectiveness.Lessons drawn from these conclusions should be combined with ex-isting knowledge concerning factors already known to partly explainthe effectiveness of adherence interventions. The mere provision ofinformation does not reduce the NA, but providing training to pro-mote the active involvement of patients can be effective [72,73]. Fur-thermore, a combination of more than one intervention measure ismore effective than any single measure alone [4,74]. Finally, long-term programs are more effective than single interventions [72,75].Obviously, addressing intentional/medication-related NA under thefactors just named as leading to success increases the complexityand the cost of an adherence intervention. On the other hand, inter-

vention programs that do not manage to sustainably reduce NA due

to multiple causes will not be cost-effective, no matter how low theintervention costs are.

Acknowledgments

The authors thank the participating pharmacies. They also thankStephan Obert, Tobias Loder, Matthias Pfannkuche, and Karin Ros-sol-Haseroth for their support. Finally, they thank Dallas Reese forproofreading and correcting the manuscript.

Source of financial support: The surveys reported here werefinancially supported by AXEA Pharma GmbH, Lindau, Germany,and Boehringer Ingelheim Pharma GmbH & Co. KG.

R E F E R E N C E S

[1] Haynes RB, Taylor DW, Sacket DL, et al. Can simple clinicalmeasurements detect patient noncompliance. Hypertension 1980;2:757–64.

[2] Osterberg L, Blaschke T. Adherence to medication. N Engl J Med 2005;353:487–97.

[3] Cramer JA, Roy A, Burrell A, et al. Medication compliance andpersistence: terminology and definitions. Value Health 2008;11:44–7.

[4] Haynes RB, Ackloo E, Sahota N, et al. Interventions for enhancingmedication adherence. The Cochrane Database Syst Rev2008;2:CD000011.

[5] World Health Organization. Adherence to long-term therapies. WHOLibrary Cataloguing-in-Publication Data, 2003.

[6] Morisky DE, Ang A, Krousel-Wood M, et al. Predictive validity of amedication adherence measure in an outpatient setting. J ClinHypertens 2008;10:348–54.

[7] Simpson SH, Eurich DT, Majumdar SR, et al. A meta-analysis of theassociation between adherence to drug therapy and mortality. BMJ2006;333:18–9.

[8] Salas M, Hughes D, Zuluaga A, et al. Cost of medication nonadherencein patients with diabetes mellitus: a systematic review and criticalanalysis of the literature. Value Health 2009;12:915–22.

[9] Gorenoi V, Schönermark MP, Hagen A. HTA-Bericht 65. Maßnahmenzur Verbesserung der Compliance bzw. Adherence in derArzneimitteltherapie mit Hinblick auf den Therapieerfolg. Köln.Deutsches Institut für Medizinische Dokumentation und Information(DIMDI), 2007.

[10] Horne R, Weinman J, Barber N, et al. Concordance, adherence andcompliance in medicine taking. Report for the National Co-ordinatingCentre for NHS Service Delivery and Organisation R & D, 2005.

[11] Horne R, Buick D, Fisher M, et al. Doubts about necessity and concernsabout adverse effects: identifying the types of beliefs that are associatedwith non-adherence to HAART. Int J STD AIDS 2004;15:38–44.

[12] Tibaldi G, Clatworthy J, Torchio E, et al. The utility of the necessity–Concerns Framework in explaining treatment non-adherence in fourchronic illness groups in Italy. Chronic Illn 2009;5:129–33.

[13] Horne R, Parham R, Driscoll R, et al. Patients’ attitudes to medicinesand adherence to maintenance treatment in inflammatory boweldisease. Inflamm Bowel Dis 2009;15:837–44.

[14] Karamanidou C, Weinman J, Horne R. Improving haemodialysispatients’ understanding of phosphate-binding medication: a pilotstudy of a psycho-educational intervention designed to changepatients’ perceptions of the problem and treatment. Br J HealthPsychol 2008;13(Pt 2):205–14.

[15] Odegard PS, Capoccia K. Medication taking and diabetes: a systematicreview of the literature. Diabetes Educ 2007;33:1014–29.

[16] Mitchell AJ, Selmes T. Why don’t patients take their medicine? Reasonsand solutions in psychiatry. Adv Psychiatr Treat 2007;13:336–46.

[17] Paes AH, Bakker A, Soe-Agnie CJ. Impact of dosage frequency onpatient compliance. Diabetes Care 1997;20:1512–7.

[18] Claxton AJ, Cramer J, Pierce C. A systematic review of the associationsbetween dose regimens and medication compliance. Clin Ther 2001;23:1296–310.

[19] Ingersoll KS, Cohen J. The impact of medication regimen factors onadherence to chronic treatment: a review of literature. J Behav Med2008;31:213–24.

[20] Chao J, Nau DP, Aikens JE. Patient-reported perceptions of side effectsof antihyperglycemic medication and adherence to medicationregimens in persons with diabetes mellitus. Clin Ther 2007;29:177–80.

[21] Piette JD, Heisler M, Wagner TH. Problems paying out-of-pocket

medication costs among older adults with diabetes. Diabetes Care2004;27:384–91.

[

[

[

[

[

[

[

[

[

[

[

[

[

[

1100 V A L U E I N H E A L T H 1 4 ( 2 0 1 1 ) 1 0 9 2 – 1 1 0 0

[22] Strauss K, MacLean C, Troy A, et al. Driving distance as a barrier toglycemic control in diabetes. J Gen Intern Med 2006;21:378–80.

23] Colombi AM, Yu-Isenberg K, Priest J. The effects of health plancopayments on adherence to oral diabetes medication and healthresource utilization. J Occup Environ Med 2008;50:535–41.

24] Maciejewski ML, Bryson CL, Perkins M, et al. Increasing copaymentsand adherence to diabetes, hypertension, and hyperlipidemicmedications. Am J Manag Care 2010;16:e20–34.

25] Phatak HM, Thomas J. Relationships between beliefs aboutmedications and nonadherence to prescribed chronic medications.Ann Pharmacother 2006;40:1737–42.

26] Karamanidou C, Clatworthy J, Weinman J, et al. A systematic review ofthe prevalence and determinants of nonadherence to phosphatebinding medication in patients with end-stage renal disease. BMCNephrol 2008;9:2.

27] Kilbourne AM, Reynolds CF, Good CB, et al. How does depressioninfluence diabetes medication adherence in older patients? Am JGeriatr Psychiatry 2005;13:202–10.

28] Nau DP, Aikens JE, Pacholski AM. Effects of gender and depression onoral medication adherence in persons with type 2 diabetes mellitus.Gend Med 2007;4:205–13.

29] Arlt S, Lindner R, Rösler A, et al. Adherence to medication in patientswith dementia: predictors and strategies for improvement. DrugsAging 2008;25:1033–47.

30] Lehane E, McCarthy G. An examination of the intentional andunintentional aspects of medication non-adherence in patientsdiagnosed with hypertension. J Clin Nurs 2007;16:698–706.

31] Lowry KP, Dudley TK, Oddone EZ, et al. Intentional and unintentionalnonadherence to antihypertensive medication. Ann Pharmacother2005;39:1198–203.

32] Gazmararian JA, Kripalani S, Miller MJ, et al. Factors associated withmedication refill adherence in cardiovascular-related diseases: a focuson health literacy. J Gen Intern Med 2006;21:1215–21.

33] Horne R, Weinman J, Hankins M. The beliefs about medicinesquestionnaire: the development and evaluation of a new method forassessing the cognitive representation of medication. Psychol Health1999;14:1–24.

34] Wilke T, Moock J, Müller S, et al. Nonadherence in outpatientthrombosis prophylaxis with low molecular weight heparins aftermajor orthopedic surgery. Clin Orthop Rel Res 2010;468:2437–53.

35] Mann DM, Ponieman D, Leventhal H, et al. Predictors of adherence todiabetes medications: the role of disease and medication beliefs. JBehav Med 2009;32:278–84.

36] Turan B, Osar Z, Molzan Turan J, et al. The role of coping with disease inadherence to treatment regimen and disease control in type 1 andinsulin treated type 2 diabetes mellitus. Diabetes Metab 2002;28:186–93.

[37] Brown C, Battista DR, Bruehlman R, et al. Beliefs about antidepressantmedications in primary care patients: relationship to self-reportedadherence. Med Care 2005;43:1203–7.

[38] Gatti ME, Jacobson KL, Gazmararian JA, et al. Relationship betweenbeliefs about medications and adherence. Am J Health-Syst Pharm2009;66:657–64.

[39] Lee YY, Lin JL. The effects of trust in physician on self-efficacy,adherence and diabetes outcomes. Soc Sci Med 2009;68:1060–8.

[40] Wroth TH, Pathman DE. Primary medication adherence in a ruralpopulation: the role of the patient-physician relationship andsatisfaction with care. J Am Board Fam Med 2006;19:478–86.

[41] Krueger KP, Berger BA, Felkey B. Medication adherence andpersistence: a comprehensive review. Adv Ther 2005;22:313–56.

[42] Hess LM, Raebel MA, Conner DA, et al. Measurement of adherence inpharmacy administrative databases: a proposal for standard definitionsand preferred measures. Ann Pharmacother 2006;40:1280–88.

[43] Julius RJ, Novitsky MA Jr, Dubin WR. Medication adherence: a reviewof the literature and implications for clinical practice. J Psychiatr Pract2009;15:34–44.

[44] Cramer JA. A systematic review of adherence with medications fordiabetes. Diabetes Care 2004;27:1218–24.

[45] Bae JW, Guyer W, Grimm K, et al. Medication persistence in thetreatment of HIV infection: a review of the literature and implicationsfor future clinical care and research. AIDS 2011;25:279–90.

[46] Garber MC, Nau DP, Erickson SR, et al. The concordance of self-reportwith other measures of medication adherence: a summary of theliterature. Med Care 2004;42:649–52.

[47] Gao X, Nau DP. Congruence of three self-report measures ofmedication adherence among HIV patients. Ann Pharmacother 2000;34:1117–22.

[48] Rolley JX, Davidson PM, Dennison CR, et al. Medication adherence self-

report instruments: implications for practice and research.J Cardiovasc Nurs 2008;23:497–505.

[49] Pearson CR, Simoni JM, Hoff P, et al. Assessing antiretroviraladherence via electronic drug monitoring and self-report: anexamination of key methodological issues. AIDS Behav 2007;11:161–73.

[50] Hollo G, Kothy P, Geczy A, et al. Personality traits, depression, andobjectively measured adherence to once-daily prostaglandin analogmedication in glaucoma. J Glaucoma 2009;18:288–92.

[51] DiMatteo MR. Variations in patients’ adherence to medicalrecommendations: a quantitative review of 50 years of research. MedCare 2004;42:200–9.

[52] Morisky DE, Green LW, Levine DM. Concurrent and predictive validity ofa self-reported measure of medication adherence. Med Care 1986;24:67–74.

[53] Mahler C, Hermann K, Horne R, et al. Patients’ Beliefs about Medicinesin a primary care setting in Germany. J Eval Clin Pract 2010 Nov 18[Epub ahead of print].

[54] DiMatteo, MR. The Psychology of Health, Illness, and Medical Care: AnIndividual Perspective. Pacific Grove, CA: Brooks/Cole, 1991.

[55] Bradley MM, Lang PJ. Measuring emotion: the Self-AssessmentManikin and the Semantic Differential. J Behav Ther Exp Psychiatry1994;25:49–59.

[56] Hardin JW, Hilbe JM. Generalized Estimating Equations. London:Chapman and Hall/CRC, 2003.

[57] DiMatteo MR, Haskard KB, Williams SL. Health beliefs, disease severity,and patient adherence: a meta-analysis. Med Care 2007;45:521–8.

[58] Nicklas LB, Dunbar M, Wild M. Adherence to pharmacologicaltreatment of non-malignant chronic pain: the role of illnessperceptions and medication beliefs. Psychol Health 2010;25:601–15.

[59] Barron J, Wahl P, Fisher M, et al. Effect of prescription copayments onadherence and treatment failure with oral antidiabeticmedications. P T 2008;33:532–53.

[60] Maciejewski ML, Bryson CL, Perkins M, et al. Increasing copaymentsand adherence to diabetes, hypertension, and hyperlipidemicmedications. Am J Manag Care 2010;16:e20–34.

[61] Wang PS, Benner JS, Glynn RJ, et al. How well do patients reportnoncompliance with antihypertensive medications?: a comparison ofself-report versus filled prescriptions. Pharmacoepidemiol Drug Saf 2004;13:11–19.

[62] Shalansky SJ, Levy AR, Ignaszewski AP. Self-reported Morisky score foridentifying nonadherence with cardiovascular medications. AnnPharmacother 2004;38:1363–8.

[63] Zeller A, Ramseier E, Teagtmeyer A, et al. Patients’ self-reportedadherence to cardiovascular medication using electronic monitors ascomparators 2008;31:2037–43.

[64] Hawkshead J, Krousel-Wood MA. Techniques for measuringmedication adherence in hypertensive patients in outpatient settings.Dis Manag Health Outcomes 2007;15:109–18.

[65] Rickles NM, Svarstad BL. Relationships between multiple self-reportednonadherence measures and pharmacy records. Res Social AdmPharm 2007;3:363–77.

[66] Mateo JF, Gil-Guillén VF, Mateo E, et al. Multifactorial approach andadherence to prescribed oral medications in patients with type 2diabetes. Int J Clin Pract 2006;60:422–8.

[67] Hughes CM. Medication non-adherence in the elderly: how big is theproblem? Drugs Aging 2004;21:793–811.

[68] Clifford S, Barber N, Horne R. Understanding different beliefs held byadherers, unintentional nonadherers, and intentional nonadherers:application of the Necessity-Concerns Framework. J Psychosom Res2008;64:41–6.

[69] Rees G, Leong O, Crowston JG, et al. Intentional and unintentionalnonadherence to ocular hypotensive treatment in patients withglaucoma. Ophthalmology 2010;117:903–8.

[70] Poluzzi E, Strahinja P, Vaccheri A, et al. Adherence to chroniccardiovascular therapies: persistence over the years and dosecoverage. Br J Clin Pharmacol 2007;63:346–55.

[71] Wilke T, Müller S, Groth A, et al. Requirements of an effectiveadherence-persistence program using the example of oral diabeticsfor diabetes mellitus type 2 patients: a systematic review. Gesundhökon Qual Manag 2011; doi:10.1055/s-0031-1273282.

[72] Vermeire EIJJ, Wens J, Van Royen P et al. Interventions for improvingadherence to treatment recommendations in people with type 2diabetes mellitus. Cochrane Database Syst Rev 2005;2:CD003638.

[73] Lewis D. Computer-based approaches to patient education: a review ofthe literature. JAMA 1999;6:272–82.

[74] Higgins N, Regan C. A systematic review of the effectiveness ofinterventions to help older people adhere to medication. Age Ageing2004;33:224–9.

[75] Norris SL, Engelgau MM, Narayan KM. Effectiveness of self-

management training in type 2 diabetes: a systematic review ofrandomized controlled trials. Diabetes Care 2001;24:561–87.