Patient?Clinician Relationships and Treatment System Effects on HIV Medication Adherence

AIDS and Behavior, Vol. 9, No. 1, March 2005 ( C© 2005)DOI: 10.1007/s10461-005-1684-1

Patient–Clinician Relationships and Treatment SystemEffects on HIV Medication Adherence

Karen S. Ingersoll1,2,3 and Carolyn J. Heckman1

Received April 21, 2003; revised June 24, 2004; accepted June 29, 2004

The study objectives were to determine the impact of the patient–clinician relationship onpatient adherence to HIV medication, to identify which aspects of the patient–clinician re-lationship and the treatment system influenced adherence, and to determine which of thesevariables remained important when the impact of mental distress and substance abuse wereconsidered. The design was a cross-sectional study using a sample of 120 HIV+ clinic patients.The Primary Care Assessment Survey (PCAS) assessed the clinician–patient relationship andthe treatment system. The Composite International Diagnostic Inventory—Short Form (CIDI-SF) screened for mental disorders, and the Brief Substance Abuse History Form measuredrecent and remote substance use. Patient adherence was assessed using five markers includ-ing 3 interview-elicited self-reports, 1 medical chart review, and 1 summary score. Logisticregression analyses identified independent predictors of each adherence behavior. PCASscores contributed to all five models, and their effects persisted when mental distress andsubstance abuse were considered. Adherence behaviors are explained by a variety of fac-tors and should be assessed using multiple methods. Further study to illuminate the mech-anisms of action of the clinician–patient relationship on adherence to HIV medication iswarranted.

KEY WORDS: HIV medication adherence; patient–clinician relationship; PCAS.

INTRODUCTION

The Patient–Clinician Relationship InfluencesGeneral Treatment Adherence

The relationship between patient and clinician isone source of the healing that can occur in chronicillness (Simpson et al., 1979, 1991). Quality patient–clinician relationships are associated with greater ad-herence in chronically ill patient groups (DiMatteo,1994; Kaplan et al., 1989; Sanson-Fisher et al., 1989)and better clinical outcomes (Stewart, 1995). Goodpatient–clinician relationships are central to primary

1Department of Psychiatry, Virginia Commonwealth University,Richmond, Virginia.

2Department of Internal Medicine, Virginia Commonwealth Uni-versity, Richmond, Virginia.

3Correspondence should be directed to Karen Ingersoll, PhD,Virginia Commonwealth University, Richmond, Virginia 23298-0109; e-mail: [email protected].

care (Institute of Medicine, 1994; Simpson et al., 1991).Research on the Primary Care Assessment Scale(PCAS) in a large sample (n = 7204 patients) indi-cated that physician knowledge of the patient and pa-tient trust in the physician were associated with ad-herence to physician advice regarding substance use,safe sex, diet, stress-management. These two relation-ship variables accounted for 14% of the variance inmedical adherence (Safran et al., 1998).

There is emerging evidence that the clinician–patient relationship may also be associated withpatient adherence to HIV medication. Martiniet al. (2002) found that patient satisfaction with theclinician–patient relationship was related to adher-ence (number of medication errors in 2 months) inoutpatients with HIV in an Italian multicenter study.In a correlational study of 707 outpatients, Bakkenet al. (2000) found that patients who were moreengaged with their providers evidenced better ad-herence to medications and appointments and better

89

1090-7165/05/0300-0089/0 C© 2005 Springer Science+Business Media, Inc.

90 Ingersoll and Heckman

immune health than their less-engaged peers. Theseresults suggest that patient–clinician relationshipsmay be related to adherence and clinical outcome ofHIV treatment.

Importance of High Medication Adherence in HIV

In HIV treatment, high medication adherence isassociated with slower progression to AIDS and lowermortality (Bangsberg et al., 2001; DeOlalla et al.,2001; Hogg et al., 2002). Adherence of at least 80%with potent combinations, and possibly 90–95%, is re-quired to avoid drug-resistant HIV and viral rebound(Flandre et al., 2002; Paterson et al., 2000). Unfor-tunately, adherence to HIV medications is uniquelychallenging. Rabkin and Chesney (1999) asserted that“combination therapy for HIV illness is perhaps themost rigorous, demanding, and unforgiving of anyoutpatient oral treatment ever introduced” (p. 61).Therefore, it is not surprising that adherence belowoptimal levels is common. Catz et al. (2000) found thata third of HIV/AIDS patients had missed doses in thepast 5 days, while 18% persistently missed doses ev-ery week over the past 12 weeks. Flandre et al. (2002)found that nonadherence ranged from 13 to 40% over3 months. Ingersoll (2004) found that 28% of clinicpatients reported taking fewer than 95% of proteaseinhibitor doses in the past week, 33% were noncom-pliant by medical records notations, 36% reportedthey did not always take medication as directed, and44% reported they had run out of their HIV medica-tion. Liu et al. (2001) also found that nonadherencerates varied by type of adherence behavior. Their es-timates of nonadherence for pill count were 6–24%,for electronically monitored pill caps (MEMS) were28–41%, and for a composite score was 24%. Patersonet al. (2000) reported a MEMS nonadherence rate of25.3%. Knobel et al. (2002) reported nonadherenceof 32.3–36.6% in a Spanish study. Spire et al. (2002)reported nonadherence of 26.7% in a French cohortof 445 patients. Taken together, these recent, well-designed studies suggest that nonadherence to HIVmedication varies by type of adherence behavior andranges between 6 and 44%; the majority of estimatesfall between 24 and 36%.

Patient Predictors of Nonadherence in HIV

Demographic variables such as education level,gender, and ethnicity are unrelated to adherencein most studies (Catz et al., 2000; Ingersoll, 2004;

Paterson et al., 2000; Singh et al., 1996; Wutoh et al.,2001). Factors that are associated with nonadherenceinclude depression, severity of side effects, poor treat-ment adherence self-efficacy, minimal social support(Catz et al., 2000; Spire et al., 2002), and anxiety(Ingersoll, 2004). Active alcohol or substance use issometimes associated with poor adherence. Cook andcolleagues found that problem drinkers were signifi-cantly more likely to report missing doses of antiretro-viral medications or to take them off schedule, andto attribute missed doses to forgetting, running outof medication, or consuming alcohol or drugs (Cooket al., 2001). Ever using heroin quadrupled the odds ofever running out of medication among clinic patients,while recent use of crack cocaine tripled the oddsof notations of noncompliance in medical recordsand sextupled the odds of taking fewer than 95%of protease inhibitor medications in the past week(Ingersoll, 2004). In two prospective studies, patientswhose alcohol or drug use increased showed reduc-tions in adherence (Lucas et al., 2002; Spire et al.,2002) and those ceasing substance use demonstratedimprovements in adherence (Lucas et al., 2002). How-ever, a few studies have not found substance abuseto be associated with adherence (Catz et al., 2000;Paterson et al., 2000; Singh et al., 1996).

No published studies have examined the impactof the clinician–patient relationship and treatmentsystem on medication adherence in HIV+ individualsusing a well-validated measure of the relationship andmultiple markers of adherence. The objectives of thisstudy were to determine whether the clinician–patientrelationship and aspects of the treatment system par-tially explained patient adherence to HIV medication,and to identify specific aspects of the clinician–patientrelationship that influenced adherence. A secondarypurpose was to examine whether clinician–patientrelationship variables maintained their independentpredictive power when mental distress and substanceabuse were considered.

METHODS

Participants

A volunteer sample of 120 patients attendingan urban university hospital infectious diseases clinicduring an 18-month period participated in the study.Eligibility criteria included able to give informed con-sent, no obvious cognitive impairment, adult, HIV+,and prescribed combination antiretroviral therapy.

Patient–Clinician Relationships and Treatment System 91

Measures

Demographics

Participants provided demographic informationabout themselves on a self-report form developedfor this study. Items included standard and HIV-specific demographic questions. Questions addressedemployment and insurance status, housing status,partner status, method of transmission of HIV, his-tory of selected transmission risk behaviors, and sex-ual orientation (see Table I).

Patient–Clinician Relationship

The patient’s perception of the patient–clinicianrelationship and treatment system was assessed usingthe Primary Care Assessment Survey (PCAS, Safranet al., 1998). The PCAS is a 51-item self-report mea-sure of primary care as defined by the Institute of

Table I. Demographic Characteristics of the Sample

Variable n %

SexMen 74 62Women 46 38

Ethnic/racialAfrican American/Black 100 83Caucasian 18 15Native American 2 2

EducationLess than high school 59 49High school or equivalence 29 24Some college 32 27

EmploymentDisabled 57 48Unemployed 37 31Part time 19 16Full time 7 6

Partner/marital statusSingle 66 56Divorced, separated, or widowed 28 24Living with spouse/partner 23 20

Sexual orientationHeterosexual 70 61Lesbian, gay, bisexual 44 39

Insurance typeGovernment 78 66None 30 25Private 10 8

Housing statusHome/apartment 61 51Another’s home/apartment 39 33None 17 14Institutionalized 3 2

Medicine Committee on the Future of Primary Care(IOM, 1994). The PCAS consists of seven constructsmeasured through eleven scales: accessibility (orga-nizational, financial), continuity (longitudinal, visit-based), comprehensiveness (knowledge of patient,preventive counseling), integration of care, clinicalinteraction (clinician–patient communication, thor-oughness of physical examinations), interpersonaltreatment, and trust (Safran et al., 1998). Responses tothe PCAS items are summed to yield raw scores thatare then converted to T scores, with higher scores in-dicating higher levels of patient-perceived quality inthat care dimension. Internal consistency reliabilitywas found to range from 0.81 to 0.95 in a sample ofover 6300 employees enrolled in 12 health plans inMassachusetts (Safran et al., 1998). Validity evidenceincludes an association between patient satisfactionand adherence to physician advice (Safran et al., 1996)as well as good convergent and discriminant validi-ties. In addition, the instrument was found to performconsistently across a number of diverse patient groups(Safran et al., 1998). The doctor–patient relationship,and the use of the PCAS in particular, has recentlybeen recommended as an important direction to im-prove understanding of adherence in chronic illnessbecause many studies have focused only on patientfactors (Goldstein, 2002). Patients rate their primaryclinician on items such as “caring and concern foryou,” “your doctor’s knowledge of your entire med-ical history,” and the “attention your doctor givesto what you have to say.” In this study, participantswere instructed to answer the questions in regardsto their primary HIV care provider. The preventivecounseling scale was not used because the patientshad already contracted HIV. PCAS psychometric databased on HIV+ samples have not been published,so these results provide unique information in thatregard.

HIV Treatment Regimen

Participants completed a 30-min, interviewer-guided self-report questionnaire, the MedicationAdherence Form, that elicited their current and pastmedication regimens using a series of branching ques-tions and visual aids such as colored cards depictingphotographs of HIV medications with their genericand trade names to enhance memory. This form wasdeveloped for this study based on a number of otherquestionnaires including the Adherence to Ther-apy Interview (The Measurement Group, 1997) and

92 Ingersoll and Heckman

included content-specific questions pertaining to HIVmedications that had newly become available in 1999and 2000. The interview covered protease-inhibitorregimens and PI-sparing combination regimens,experience with side effects, opportunistic infectionprophylaxis, attitudes about medication, barriers totreatment, and adherence behaviors. The interviewerasked whether a participant was now taking eachmedication, the current dose, dosing schedule,dosing requirements, and side effects related to thatmedication. Participants reported their reasons fornot taking medication. Summary statements weremade by the interviewer at several checkpoints todetermine completeness of the reported regimen.Following these summary statements, participantswere reminded of the number of protease inhibitorpills their regimen required, then were asked to recallhow many pills of each medication they missed duringthe past week. From these answers, the proportionof PI medications missed and taken was calculated.

Adherence

Because adherence can best be understood asa set of related behaviors, and due to the lackof a single “gold-standard” for adherence measure-ment, multiple markers of adherence should be usedto fully characterize the behaviors (Dunbar-Jacob,2002). Empirical data also support the predictive va-lidity of composite adherence measures for subse-quent viral load (Liu et al., 2001). The first markerwas proportion of PI medications taken during thepast week. Proportions of 95% or greater were cate-gorized as High, with 94% and below as Low, basedon Paterson et al.’s (2000) finding that those with atleast 95% adherence showed significantly greater vir-ulogical response than others. The second marker wasthe self-report, yes/no response to the question “Haveyou ever run out of your HIV medications?” Thethird marker was the self-report, yes/no response tothe question “Do you always take all of your med-ications as directed?” These questions were embed-ded in different sections of the Medication AdherenceForm. The last marker was a collateral report of ad-herence derived from the electronic medical record.Researchers queried the electronic medical record toidentify any provider notations of noncompliance inthe medical chart; if notations such as “noncompli-ant,” “having trouble taking medications,” “possiblenoncompliance,” or “long history of poor adherence”were present, the participant was categorized as non-adherent by the medical record. Lastly, it was of in-

terest to determine the extent of overlap of the fourindicators of adherence and to develop a summarymarker of adherence, as recommended by Liu et al.(2001). Therefore, an adherence score was devisedthat gave a point for each of the measures on whichthe person was categorized as adherent. Because aminority of participants were taking PI regimens andtherefore were not queried on proportion of PI dosestaken in the past week, their scores on the summarymeasure could range from 0 to 3, with 0 meaning theywere adherent on none of the measures and 3 indi-cating perfectly adherent by all of the measures. Forthose taking PI regimens, scores could range from 0to 4, with 4 indicating adherence on all four measures.

Health Status

The participant’s health status, including CDCclassification that indicated HIV or AIDS, historyof opportunistic infections, currently prescribed regi-men, three most recent CD4 and three most recent vi-ral load counts, diagnostic codes, and staff commentsregarding adherence were derived by querying theparticipant’s electronic medical record.

Substance Abuse

The Brief Substance Abuse History, a structured,interviewer-administered form, was adapted fromthe Drug History form (NIDA, 1993) and queriedwhether the participant had ever used a list of com-monly abused substances including nicotine, alco-hol, amphetamines, cocaine, heroin/opioids, hallu-cinogens, marijuana, drug combinations, and otherdrugs. In each case when the participant had everused/tried a substance, a series of branching questionswas used to determine the extent of use, recency ofuse, and number of days used in the past 30. Addi-tionally, participants answered whether they consid-ered themselves to have a “primary drug” or a “drugof choice” and whether any drug was causing prob-lems for them or had in the past. Participants alsoindicated whether they had undergone various typesof drug abuse treatment.

Mental Disorders

A range of DSM-IV Axis I acute mental dis-orders was screened for using a brief, interviewer-administered questionnaire, the Composite Interna-tional Diagnostic Interview—Short Form (CIDI-SF;

Patient–Clinician Relationships and Treatment System 93

WHO, 1998). The CIDI-SF takes 7 min to admin-ister and is convergent with the full-length CIDI(Kessler et al., 1998). The short form also hasgood sensitivity, specificity, and predictive validity(Kessler et al., 1998). For this study, CIDI-SF moduleswere used to assess symptoms of Major Depression,Anxiety Disorders (Panic Attacks, Agoraphobia,Generalized Anxiety Disorder, Social Phobia, andObsessive-Compulsive Disorder), and Drug andAlcohol Dependence in the past 12 months.

Procedures

The study utilized procedures that protected vul-nerable human subjects and that were approved bythe Institutional Review Board. Patients attendingthe infectious diseases clinic were approached in thewaiting room by a research assistant and asked toparticipate in the study. Patients were also invitedto participate via hospital flyers and personal recruit-ment by clinic staff. Patients were provided with infor-mation about the study and were scheduled to meetwith a research assistant at their convenience, typi-cally around their other scheduled hospital appoint-ments. The assessment required between 2 and 3 hr.Participants were offered frequent breaks and snacksin addition to transportation vouchers and compensa-tion. Although the procedures were rather long, par-ticipants did not seem bothered by the inconvenience,and some referred others for screening. Research as-sistants helped patients read and understand the sur-vey items if necessary. Interviewers were trained todevelop a good rapport with the participant, ask allquestions neutrally and nonjudgmentally, and correctinconsistencies that arose over the course of the inter-view before proceeding. These procedures were de-veloped to reduce potential response bias in whichparticipants would be tempted to “please” the in-terviewer by overreporting adherence. Participantsreceived $30 merchandise gift cards from a major dis-count department store as compensation for their par-ticipation.

Data Analyses

Means, standard deviations, and frequencieswere used to describe continuous and discrete de-mographic, mental disorder, substance use, and ad-herence characteristics of the sample. The relation-ship between adherence score and recent CD4 (aboveor below 200) and viral load (detectable vs. unde-

tectable) were examined using t tests. Internal con-sistency reliability (Cronbach’s coefficient alpha) wascalculated on the score items of the MedicationAdherence Form and the PCAS. Stepwise multivari-ate logistic regression was used to develop explana-tory models of all five types of adherence behaviorincluding running out of medication, not taking med-ication as directed, low proportion of protease in-hibitor pills taken, noncompliance noted by a clinicianin the medical chart, and overall adherence score.All PCAS scales were entered into these initial mod-els. After significant models and significant PCAS ex-planatory variables were identified, mental distressand substance abuse variables previously identified assignificantly related to the four adherence behaviors(Ingersoll, 2004) were added to the PCAS models.

RESULTS

Thirty-eight percent of the sample was female,slightly more women than among the clinic popula-tion in general. The mean educational level of thesample was 10th grade, with years of formal educationcompleted ranging from 0 to 18. The mean age was40.4 years. Participants’ average viral load was 77,493copies (SD 188,416), while their average CD4 (t cell)count was 378.3 (SD 313.8). Participants averaged 3.2(SD 1.6) medications for HIV and 6.2 (SD 3.6) med-ications total. The typical participant was an AfricanAmerican heterosexual single man who was disabledand diagnosed with AIDS (rather than asymptomaticHIV disease) for 4 years, had a detectable viral loadand CD4 count above 200, and was on a triple com-bination protease inhibitor-sparing medication regi-men, often due to failing previous regimens. Demo-graphic characteristics are provided in Table I.

A large proportion of the sample reported psy-chiatric and substance use problems within the previ-ous 12 months (Table II). Over 50% of the participantsreported a major depression, almost 45% reported ananxiety disorder, and 26% reported problems with al-cohol or drug dependence. In addition to substancedependence criteria, over half of the sample reportedhaving a “primary drug,” with most of these beingillicit, and 40% of participants admitted to illicit sub-stance use within the past 6 months.

Rates of nonadherence in the current samplewere high, ranging from 30 to 44%, depending onthe criterion employed (See Table II). The strictestmeasure of adherence was the composite score. Com-posite adherence rates (scoring as fully adherent by

94 Ingersoll and Heckman

Table II. Rates of Mental Disorders, Substance Abuse Problems, Health Status, and Adherence in 120 Patients With HIV

Measure Disorder/characteristic N %

CIDI-SF (past year) Major depression 60 52.6Anxiety disorder 50 43.9Alcohol or drug dependence 30 26.3

Brief substance abuse history Has a “primary drug” 65 55.1Primary drug is alcohol 21 17.5Primary drug is illicit 42 35.0Recent illicit drug use (within 6 months) 47 39.8

Medical record Noted to be a substance abuser by provider 47 39.8Medication adherence form a. Does not always take medication as directed 43 36.8

b. Has ever run out of medications 52 44.4c. Noted to be noncompliant in records 38 32.2d. Proportion of PI taken below 95% past weeka 14 30.4e. Sometimes skips protease inhibitor dosea 26 56.5

Composite Adherence Score (To how manyitems is patient scored as adherent of a, b,and c, above?)

0 10 8.62

1 29 25.02 44 37.93 33 28.5

Composite Protease Adherence Score (To howmany items is patient scored as adherent ofa, b, c, and d, above?)a

0 2 4.1

1 7 14.32 16 32.73 10 20.44 14 28.6

Disease status AIDS 71 65HIV 38 35

Most recent viral load by chart Detectable (>400) 70 59Undetectable 48 41

Most recent CD4 count by chart Over 200 70 59Below 200 48 41

Has regular HIV care provider Yes 112 96No 5 4

On OI Prophylaxis Yes 42 36Taken and discontinued a PI Yes 48 41Current PI None 62 53

Nelfinavir 25 21Ritonavir 13 11Indinavir 7 6Kaletra 5 4Amprenavir 4 3Saquinavir 2 2

aSample for these analyses consists of the 46 patients who were on a current protease inhibitor regimen from the total sampleof 120, which included patients on PI-sparing regimens.

always taking medication as prescribed, never run-ning out of medication, and not having noncompli-ance noted in the medical records) were very low;only 29% scored a 3 out of 3.

Adherence was related to immune health in thissample. Those with detectable viral loads (n = 67) hadlower mean adherence scores (1.69, SD 0.91) thanthose with undetectable viral loads (n = 48), whosemean adherence score was 2.19 (SD 0.68); this dif-ference was significant (t = 3.04, p = .003). Among

those with CD4 counts below 200, the mean adher-ence score was 1.64 (SD 0.86), while among those withCD4 counts above 200, the mean adherence score was2.06 (SD 0.90), and this difference was also significant;t = 2.45, p = .02.

Internal Consistency of the MAF and PCAS

The Medication Adherence Form (MAF) isan interview process that yields primarily discrete

Patient–Clinician Relationships and Treatment System 95

variables. The four behavioral measures of adherence,and the two composite scores, were converted fromcategorical yes/no to numeric variables to permit as-sessment of the item–scale correlations. Therefore,scores of 1 or 0 were assigned to each of the items: al-ways takes medication as directed, notations of non-compliance in the chart, proportion of protease in-hibitors taken is 95% or greater, and has ever run outof medication. The composite scores ranged from 0 to3 for those not taking protease inhibitors and 0 to 4 forthose taking PIs as previously described. The internalconsistency of these six score items of the MAF wasvery good, with a standardized Cronbach’s coefficientalpha of .81.

The internal consistency of the PCAS scale forthis sample of HIV+ patients was excellent, with astandardized Cronbach’s coefficient alpha of .88 forthe total scale. Scale coefficient alphas ranged froma low of .24 for the Trust scale to a high of .95 forintegration of care (see Table III). Two scales (Trustand Organizational Access) failed to achieve an alphaof .70 and therefore have inadequate internal consis-tency to yield reliable measurement of their purportedconstructs in this sample. This may be due to the lim-ited range of responses to these scales, with almost allparticipants rating both the provider and clinic veryhighly on these scales. However, we retained all scalesfor predictive analyses to allow comparison with pre-vious research.

Patients with HIV-generated PCAS scale scoresthat were consistent with those reported for patientsin primary care in terms of financial and organi-zational accessibility, visit continuity, integration of

Table III. Primary Care Assessment Scale t Scores and Scale Reliability in 120 PatientsWith HIV

SubscaleCorrelation coefficient

Scale name n Mean SD with total alpha

Financial accessibility 106 64.70 21.53 .33 .90Organizational accessibility 107 66.23 18.83 .74 .85Longitudinal continuitya 107 63.32 32.09 .09Visit continuity 107 82.99 25.22 .37 .84Comprehensive knowledge 106 74.16 19.75 .85 .66Integrationb 52 71.69 21.69 .85 .95Physical exama 107 81.68 20.21 .79Interpersonal treatment 107 81.12 18.65 .88 .94Communication skill 107 82.17 17.36 .83 .94Trust 107 92.60 15.90 .74 .24aThe Longitudinal continuity and physical exam scores are calculated based on oneitem each and thus interitem correlations and reliability cannot be calculated for thesescales.

bThe Integration of Care scale is only scored for those patients who indicated that theirprimary HIV care provider had referred them to a different specialist.

specialty care, quality of physician exam, interper-sonal treatment, and communication. In contrast, themean longitudinal continuity score in this samplewas 13 points lower than the norm group mean and12 points lower than the 50th percentile of the nor-mative group. In two areas, this sample of patientsscored higher than the normative sample. In contex-tual knowledge of the patient, this sample’s mean of74.16 was 20 points higher than the mean and 6 pointshigher than the 75th percentile of the normative sam-ple. In trust, this sample’s mean of 92.60 was 17 pointshigher than the mean of the normative group and5 points higher than the 75th percentile of the nor-mative group. In other words, while patients reportedbeing seen in the clinic or by their provider for a rel-atively short period of time, they reported very goodrelationships with their providers. PCAS Scale scoresare provided in Table III.

Explanation of Adherence

Explanation of Adherence Using Patient–ClinicianVariables Only

According to the logistic regression (seeTable IV), PCAS scores did not predict running outof medication and the model was not significant(LR�2

(2 df) = 4.07, p < .09). In contrast, the modelof not taking medication as directed was significant,explaining 10% of the variance (LR�2

(4 df) = 12.53,p < .02). The PCAS interpersonal treatment vari-able increased the risk of nonadherence, whereas

96 Ingersoll and Heckman

Table IV. Final Logistic Regression Models

Measure Variable B SE OR CL Wald �2

Model 1: Predicting running out of medicationPCAS Knowledge of patient −0.03 .01 0.97 0.94–0.99 6.54∗PCAS Length of continuity of care 0.01 .01 1.01 0.99–1.02 1.26CIDI-SF Depression 0.52 .22 2.85 1.20–6.77 5.67∗

Model 2: Not taking medication as directedPCAS Organizational access −0.03 .02 0.97 0.94–1.00 3.45

Longitudinal continuity 0.02 .01 1.02 1.00–1.03 3.60Interpersonal skill 0.06 .02 1.06 1.02–1.11 7.24∗∗Knowledge of patient −0.04 .02 0.94 0.90–0.98 4.16∗

CIDI-SF Any anxiety disorder −0.60 .26 0.25 0.08–0.74 5.48∗Substances Recent illicit drug use 0.26 .25 1.66 0.64–4.36 1.07

Model 3: Taking below 95% of protease inhibitor pillsPCAS Financial access −0.09 .04 0.91 0.84–1.00 4.29∗

Scientific knowledge −0.11 .06 0.90 0.80–1.01 3.36Longitudinal continuity 0.06 .03 1.07 1.00–1.14 3.51Communication skill 0.12 .08 1.13 0.96–1.33 2.19Knowledge of patient 0.08 .06 1.09 0.97–1.22 2.16

CIDI-SF Any anxiety disorder −0.72 .62 0.24 0.02–2.74 1.33Drug or alcohol dependence 1.38 .80 15.65 0.68–361.61 2.95

Substances Recent illicit drug use 0.71 .58 4.10 0.42–40.12 1.47Model 4: Predicting noncompliance (per chart)

PCAS Interpersonal skill −0.03 .02 1.05 0.99–1.11 1.22Communication skill 0.05 .03 1.05 0.99–1.11 2.33

Substances Recent illicit drug use 0.41 .23 2.27 0.93–5.53 3.27Model 5: Predicting poor adherence score (≤1 of 3)

PCAS Organizational access −0.03 .02 0.97 0.94–1.01 2.65Communication 0.04 .02 1.04 1.00–1.09 3.93∗Knowledge of patient −0.03 .02 0.97 0.94–1.01 2.48

Substances Recent illicit drug use 0.39 .25 2.18 0.82–5.76 2.45Primary drug identified −0.36 .25 0.48 0.18–1.30 2.07

Note. The confidence limits of some significant Odds Ratios appear to contain 1.00 due to rounding.∗ p < .05. ∗∗ p < .01.

organizational access predicted better adherence. Inaddition, PCAS continuity of care was retained inthe model but was not an independent predictor.The model of low proportion of protease inhibitorpills taken was also significant, with five PCAS vari-ables explaining a moderate amount of the variance,29% (LR�2

(5 df) = 15.56, p < 0.01). Two variablescontributed independently to low proportion of pillstaken, financial accessibility, which was a protectivefactor, and continuity of care, which was a risk factor.Other variables in the model included comprehensiveknowledge of the patient, communication skills, andphysical exam (see Table IV).

The model of chart noncompliance was notsignificant, with PCAS variables explaining only3% of the variance (LR�2

(2 df) = 3.28). In contrast,the predictive model of poor adherence score wassignificant, and PCAS variables explained 9% of thevariance (LR�2

(4 df) = 11.39). It had two independentpredictors, PCAS interpersonal treatment, a risk

factor, and organizational access, a protective factor.PCAS scales that had no relationship to adherenceas measured by any of the four behavioral or thecomposite indicators included visit-based continu-ity, integration of care, quality of physical exam,and trust.

Explaining Adherence With Both Patient–Clinicianand Patient Mental Distress and Substance Use

The final models that include both PCAS andpatient variables (mental distress and substance use)are presented in Table IV. Some PCAS variables re-mained in these models as independent predictors.The model of running out of medication was sig-nificant and included three variables that accountedfor 13% of the variance (LR�2

(2 df) = 13.81, p <

.01). Variables that contributed independently tothis model were contextual knowledge of the pa-tient, which was slightly protective, and depression,

Patient–Clinician Relationships and Treatment System 97

which nearly tripled the risk of running out ofmedication.

The model of not taking medication as di-rected was also significant, explaining 16% of thevariance (LR�2

(6 df) = 17.35, p < .01). Variables thatcontributed independently to not taking medicationas directed were one risk factor, PCAS interpersonaltreatment, and two protective factors, PCAS contex-tual knowledge of the patient, and any CIDI-SF anx-iety disorder. Anxiety disorder decreased the risk ofnot taking medication as directed more than fourfold.

The model of low proportion of protease in-hibitor pills taken was significant, with its eight vari-ables explaining a large amount of the variance, 45%(LR�2

(8 df) = 25.04, p < .01). However, only one vari-able contributed independently to low proportion ofpills taken, PCAS financial accessibility, which was aprotective factor.

The model of chart noncompliance was stillinsignificant, explaining only 6% of the variance(LR�2

(3 df) = 5.80), and had no independent predic-tors. Similarly, the overall predictive model of pooradherence score remained insignificant, explainingonly 8% of the variance (LR�2

(5 df) = 8.65). It had oneindependent risk factor, PCAS communication.

DISCUSSION

Medication regimens for HIV/AIDS are com-plex and demanding, and these results confirm thatmany individuals prescribed antiretroviral medica-tion are not adequately adherent. Rates of nonadher-ence were high, ranging from 30 to 44% depending onthe measure. Using the composite measure of adher-ence, or consistently scoring as adherent across threebehaviors, the rate of consistent adherence across be-haviors was very low, 29%. These figures indicate thatpatients are willing to self-report various nonadherentbehaviors.

These data demonstrate that the patient–clinician relationship and treatment system influencemedication adherence in HIV, even when patient fac-tors such as substance abuse and mental health vari-ables are considered. Scales from the PCAS were im-portant in all five models predicting different aspectsof adherence. PCAS scales better explained adher-ence behavior than past year depression, substanceuse, or anxiety. In addition, some clinician–patientrelationship factors served as protective, and someserved as risk factors for adherence to HIV med-ication. Of special interest are the counterintuitive

findings that some of the “positive” relationship vari-ables conferred risk to adherence, while recent anxi-ety symptoms provided protection.

Contextual knowledge of the patient as a personincludes perceptions that the clinician knows one’shistory, understands one’s responsibilities at work,home, or school, and understands one’s core valuesand beliefs. Patients who perceived their clinicians toknow them well were less likely to report they hadrun out of medication or to report they did not alwaystake it as directed. It is notable that the study sample’smean on this scale was much higher than the norma-tive group’s mean; this likely indicates that the HIVcare providers of this sample of patients are perceivedto understand their patients’ life situations in muchgreater depth than other primary care providers.

Financial accessibility, or the fairness/value of thecost of care, was an independent and protective pre-dictor of the proportion of PI medication taken. Fi-nancial considerations may play a more importantrole in adherence to the most costly regimens, al-though many participants in this study were likely el-igible for treatment at no or low cost either throughthe charity care of the institution or the state’s AIDSDrug Assistance Program.

Interpersonal treatment was an independentrisk factor for not taking medication as directedand contributed to the prediction of noncompliancenotations in the chart. A related area, communi-cation skill, was an independent risk factor for apoor adherence score, again conferring only a slightincrease in the risk. These results initially seemcounterintuitive. Why might communication andinterpersonal skill be detrimental to medicationadherence in HIV patients? Providers who discussadherence issues but are not confrontational may beperceived as more interpersonally skilled. Patientsmay misconstrue these providers’ empathy for per-mission to be less adherent. Alternatively, providerswho are good communicators and have higher levelsof interpersonal interest may discuss a broad rangeof issues with their patients, rather than focusing onlyon adherence. Patients may then fail to grasp thecritical nature of adherence because it did not standout enough from other topics of discussion. It is alsoimportant to recognize that interpersonal treatmentwas rated by patients, whereas professionals mightrate “skill” level differently. Patients might rateinterpersonal treatment based on kindness alone,as opposed to a balance of appropriate challengeand support that might better foster adherentbehaviors.

98 Ingersoll and Heckman

Continuity of care, or the length of the patient’srelationship with the clinician, contributed to threemodels but was not an independent predictor. Thisscale mean was lower in this sample than in the nor-mative group, and the lower mean may indicate thatpatients completing the original PCAS experiencedlonger relationships with their PCPs than HIV+ pa-tients had with their HIV care providers. This makessense given that in this adult sample, HIV had notbeen a lifelong illness, and this result may reflect theshorter duration of their HIV care. Longer continu-ous care could be related to nonadherence as it con-veys a longer duration of living with HIV. Maintainingexcellent adherence over longer periods may be par-ticularly challenging and remains an area in need offurther study.

Trust has previously been identified as an im-portant predictor of patient outcome and adherence(Safran et al., 1998). Carr (2001) conducted qualita-tive interviews with 14 HIV+ patients to explore trust.Several patients explicitly identified trust as an impor-tant relationship factor that provided curative bene-fits. Trust appeared to be based on a variety of fac-tors such as the length of the relationship, a generalfeeling of comfort on the part of the patient, and sev-eral provider variables such as their apparent clinicalskills and knowledge, willingness to impart informa-tion, apparent enthusiasm, nonjudgmental attitude,understanding of the patient’s personal situation, re-assuring behaviors, and general willingness to collab-orate in care. The absence of trust as a predictor in thisstudy is most likely due to its unusually low reliabilityin this sample due to a restriction of range in the re-sponses to this scale. The participants uniformly ratedtheir trust in their clinician as extremely high. Thus,trust did not vary by level of adherence in this sam-ple. This finding should not be misconstrued as mean-ing that trust is unimportant to HIV treatment adher-ence. In fact, we did find that some of the elements oftrust suggested by patients (longitudinal continuity,comprehensive knowledge of patient) were, in fact,predictive of adherence in this study. In the setting ofmore variation in trust and a more reliable measureof trust, the role of trust in facilitating adherence inHIV might emerge more clearly.

Although mental disorders (Major Depressionand Anxiety) and substance use contributed to theprediction of adherence, substance use was not an in-dependent predictor in any model. Depression hasbeen identified in numerous other studies as an im-portant predictor of adherence (e.g., Catz et al., 2000;Spire et al., 2002), and although it contributed to pre-

dicting running out of medication, it did not predictother types of nonadherent behavior. A patient withactive major depression may experience lethargy anddemoralization that result in failing to obtain a pre-scription refill on time, leading to running out of med-ication. Moreover, he/she may be passively suicidaland therefore, not take medications for this reason.Anxiety was a significant protective factor in not tak-ing medication as directed, but was not independentin any other model. A patient with an anxiety disor-der may be vigilant about his or her health, and thisanxiety helps him/her to adhere closely to taking med-ication as directed. Substance disorders did not inde-pendently explain any aspect of adherence measuredin this study. While previous studies have demon-strated such effects (Cook et al., 2001; Lucas et al.,2002; Spire et al., 2002) as might be expected given thecognitive, emotional, and social impairments causedby substance problems, the current study did not ob-tain such results. It may be that such discrepancieswith existing literature are related to measurementtimeframe. This study screened for disorders duringthe past year, but perhaps recent psychiatric and sub-stance problems (e.g., past 30 days) would be morerelated to adherence. Alternatively, it is possible thatpatient–clinician relationship factors and a range ofpsychiatric disorders are more strongly related to ad-herence than is substance use. In summary, aspects ofthe doctor–patient relationship were important acrossthe broad behaviors that comprise adherence, whilepatient mental health factors contributed only to spe-cific types of nonadherent behavior in this sample.

Clinical Recommendations

These findings should encourage clinicians car-ing for people with HIV to enhance their relation-ships with patients to improve adherence. Resultssuggest that this enhancement might include bothsupport and challenge, as some of the results arecounterintuitive with regard to the clinician–patientrelationship. Relationship building can be challeng-ing, as HIV/AIDS can raise issues for clinicians thatmay impact their relationship with patients nega-tively due to its severity and prognosis, contagion is-sues, homophobia, substance abuse, as well as iso-lation and stigmatization (Battegay et al., 1991).Developing a deeper knowledge of the patient in con-text may be an important bridge to improved adher-ence. Clinicians should also examine their methodof discussing adherence. Physicians providing HIV

Patient–Clinician Relationships and Treatment System 99

care vary widely as to the timing, content, and ex-tensiveness of their communication about adherence(Roberts and Volberding, 1999). Those clinicians withsmoother interpersonal skills, who enjoy open com-munication, may cover a range of issues; therefore,their patients may not perceive the critical nature ofadherence.

Providers may be able to improve patient ad-herence by facilitating the treatment of depressionand fostering healthy “anxiety” about the risk of nottaking medications as directed. This is not to sug-gest that providers leave anxiety disorders untreated.Rather, they can help patients understand the poten-tial clinical outcomes of sporadic adherence and ex-plore their ambivalence about taking medication. Inaddition, while substance use may have negative con-sequences for patients’ lives in general, its impact onmedication adherence may not be so important as pre-viously thought, especially when other variables aretaken into consideration.

Limitations of the Study

This study’s primary limitation was its conve-nience sampling strategy among volunteer partici-pants who were patients of an infectious diseasesclinic, which could introduce biases into the study thatlimit generalizability. Factors that moderate this con-cern include a sample that reflected the demographiccharacteristics of the larger clinic population and ade-quate variability in most predictor and outcome vari-ables. Future research should use representative sam-pling strategies among patients with HIV.

A second limitation is the use of self-reportscreening instruments rather than structured clini-cal interviews to generate hits for mental health andsubstance abuse disorders. However, several authors(Chesney et al., 2000; Duong, et al., 2001) foundthat self-report measures, despite their risk of under-estimating nonadherence, provide satisfactory con-current and predictive validity (e.g., predicting viralload). While the CIDI-SF has excellent psychome-tric properties that compare well to structured in-terviews, its diagnoses are not definitive. Thus, therates of recent Major Depression, Anxiety Disorder,and Alcohol or Drug Dependence identified in thisstudy may vary from rates ascertained using other in-struments. Therefore, the relationship of these men-tal disorders, or a lifetime rather than recent his-tory of symptoms, to adherence should be examinedfurther.

A third limitation is that without a “gold stan-dard” for adherence, it is difficult to know the truerates of adherence in the sample. Although some stud-ies have used electronic monitoring, that approach isexpensive, not feasible in most clinical settings, andpresents practical concerns about requiring patientsto use individual pill bottles rather than pillboxes, thatcan serve as a memory aid.

Future Directions

Methodology improvements in adherence re-search remain a pressing need. A recent meta-analyticreview of the relationship of adherence to clinical out-comes revealed that method of measuring adherencewas the largest source of variance in the relationshipbetween adherence and outcomes (DiMatteo et al.,2002). We have attempted to address this need by de-veloping separate models for each indicator of ad-herence. The use of multiple self-report measures,behavioral measures, pill counts, electronic monitor-ing, pharmacy reporting, and other technology-basedreporting methods should be pursued to maximizethe reliability, validity, and practicality of adherencemeasurement, and researchers should avoid report-ing these using composites alone. Future studies thatreport a range of adherence behaviors will facilitategreater understanding of the facilitators and barriersof each type of adherence.

These data suggest that adding measures of thepatient–clinician relationship into future adherencestudies could yield improved understanding of adher-ence. Although this study provides some hypothesesabout the nature of the relationships between adher-ence and patient–clinician variables, these merit fur-ther study, using both self-report and observationalmethods such as rating relationship processes. In addi-tion, characterizing the nature of the patient–clinicianrelationship may be fruitful. There may be differ-ent patterns of adherence behaviors in relationshipsthat are paternalistic, mentoring, collaborative, or au-tonomous in nature (Balint and Shelton, 1996). Thegoal is to specify the mechanism of action of thepatient–clinician variables on subsequent adherencebehaviors. To that end, prospective studies that ex-amine the patient–clinician relationship over time,and observe changes in adherence and outcomes,are highly desirable. Finally, intervention researchersshould examine the most effective ways to facilitateadherence through improving the patient–clinicianrelationship.

100 Ingersoll and Heckman

ACKNOWLDGMENTS

The authors thank Stephanie Lundgren, KristinaHash, Jan Ivery, and Raphael Mutepa for datacollection and the VCU Health System InfectiousDiseases Clinic staff and patients for participation.Support was provided by NIMH #K01MH01688 andby NIDA #T32DA07027-26.

REFERENCES

Bakken, S., Holzemer, W. L., Brown, M. A., Powell-Cope, G. M.,Turner, J. G., Inouye, J., Nokes, K. M., and Corless, I. B.(2000). Relationships between perception of engagement withhealth care provider and demographic characteristics, healthstatus, and adherence to therapeutic regimen in persons withHIV/AIDS. AIDS Patient Care STDS, 14, 189–197.

Balint, J., and Shelton, W. (1996). Regaining the initiative: Forginga new model of the patient–physician relationship. Journal ofthe American Medical Association, 275, 887–891.

Bangsberg, D. R., Perry, S., Charlebois, E. D., Clark, R. A.,Roberston, M., Zolopa, A. R., and Moss, A. (2001). Non-adherence to highly active antiretroviral therapy predicts pro-gression to AIDS. AIDS, 15, 1181–1183.

Battegay, M., Weber, R., Willi, J., Eich, D., Siegenthaler, W., andLuthy, R. (1991). Exploring the doctor–patient relationshipreduces staff stress and enhances empathy when caring forAIDS patients. Psychotherapy and Psychosomatics, 56, 167–173.

Catz, S. L., Kelly, J. A., Bogart, L. M., Benotsch, E. G., andMcAuliffe, T. L. (2000). Patterns, correlates, and barriers tomedication adherence among persons prescribed new treat-ments for HIV disease. Health Psychology, 19, 124–133.

Chesney, M. A., Morin, M., and Sherr, L. (2000). Adherence toHIV combination therapy. Social Science and Medicine, 50,1599–1605.

Cook, R. L., Serelka, S. M., Hunt, S. C., Woodward, W. C., Erlen,J. A., and Conigliaro, J. (2001). Problem drinking and medica-tion adherence among persons with HIV infection. Journal ofGeneral Internal Medicine, 16, 83–88.

DeOlalla, P., Knobel, H., Carmona, A., Guelar, A., Lopez-Colomes,J., and Cayla, J. (2001). Impact of adherence on highly activeantiretroviral therapy on survival in HIV-infected patients.Journal of Acquired Immune Deficiency Syndrome, 30, 105–110.

DiMatteo, M. R. (1994). Enhancing patient adherence to medicalrecommendations. JAMA, 271, 79–83.

DiMatteo, M. R., Giordani, P. J., Lepper, H. S., and Croghan, T. W.(2002). Patient adherence and medical treatment outcomes: Ameta-analysis. Medical Care, 40, 794–811.

Dunbar-Jacob, J. (2002, April). Adherence: Perspective on the in-dividual. Master lecture presented at the Society of Behav-ioral Medicine 23rd annual meeting and scientific sessions,Washington, DC.

Duong, M., Piroth, L., Grappin, M., Forte, F., Peytavin, G.,Boisson, M., Chavanet, P., and Portier, H. (2001). Evaluationof the patient medication adherence questionnaire as a toolfor self-reported adherence assessment in HIV-infected pa-tients on antiretroviral regimens. HIV Clinical Trials, 2, 128–135.

Flandre, P., Peytavin, G., Meiffredy, V., Saidi, Y., Descamps, D.,Delagnes, M., Brun-Vezinet, F., Raffi, F., and the TrilegeStudy Team. (2002). Adherence to antiretroviral therapyand outcomes in HIV-infected patients enrolled in an induc-

tion/maintenance randomized trial. Antiviral Therapy, 7, 113–121.

Goldstein, M. G. (2002, April). Building relationships and promot-ing partnerships: Creating bonds to strengthen adherence. Pres-idential address given at the Society of Behavioral Medicine23rd annual meeting and scientific sessions, Washington, DC.

Hogg, R. S., Heath, K., Bangsberg, D., Yip, B., Press, N.,O’Shaughnessy, M. V., and Montaner, J. S. (2002). Intermittentuse of triple-combination therapy is predictive of mortality atbaseline and after 1 year of follow-up. AIDS, 16, 1051–1058.

Ingersoll, K. (2004). The impact of psychiatric symptoms, drug use,and medication regimen on non-adherence to HIV treatment.AIDS Care, 16, 199–211.

Institute of Medicine. (1994). Defining primary care: An interimreport. Washington, DC: National Academy Press.

Kaplan, S. H., Greenfield, S., and Ware, G. E. (1989). Assessing theeffects of physician–patient interactions on the outcomes ofchronic disease. Medical Care, 27(Suppl.), S110–S127.

Kessler, R. C., Andrews, G., Mroczek, D., Ustun, B., and Wittchen,H.-U. (1998). The World Health Organization CompositeInternational Diagnostic Interview Short Form (CIDI-SF).International Journal of Methods in Psychiatric Research, 7,171–185.

Knobel, H., Alonso, J., Casado, J. L., Collazos, J., Gonzalez, J.,Ruiz, I., Kindelan, J. M., Carmona A., Juega, J., and Ocampo,A. (2002). Validation of a simplified medication adherencequestionnaire in a large cohort of HIV-infected patients: TheGEEMA study. AIDS, 16, 605–613.

Liu, H., Golin, C. E., Miller, L. G., Hays, R. D., Beck, C. K.,Sanandaji, S., Christian, J., Maldonado, T., Duran, D., Kaplan,A. H., and Wenger, N. S. (2001). A comparison study of multi-ple measures of adherence to HIV protease inhibitors. Annalsof Internal Medicine, 134, 968–977.

Lucas, G. M., Gobo, K. A., Chaisoon, R. E., and Moore, R. D.(2002). Longitudinal assessment of the effects of drug and al-cohol abuse on HIV-1 treatment outcomes in an urban clinic.AIDS, 29, 767–774.

Martini, M., Parazzini, F., and Agnoletto, V. (2002). Adherence toHIV treatment: Results from a 1-year follow-up study. HIVMedicine, 3, 62–64.

National Institute on Drug Abuse (NIDA). (1993). Drug HistoryForm. In How good is your drug abuse treatment program? Re-source manual (pp. RS3-40–RS3-41). Rockville, MD: NIDA.

Paterson, D. L., Swindells, S., Mohr, J., Brester, M., Vergis, E.,Squier, C., Wagener, M. M., and Singh, N. (2000). Adherenceto protease inhibitor therapy and outcomes in patients withHIV infection. Annals of Internal Medicine, 133, 21–30.

Rabkin, J. G., and Chesney, M. (1999). Treatment adherence toHIV medications: The Achilles heel of the new therapeu-tics. In D. Ostrow and S. Kalichman (Eds.), New HIV thera-pies: Psychological and public health implications (pp. 61–82).New York: Plenum Press.

Roberts, K. J., and Volberding, P. (1999). Adherence communi-cation: A qualitative analysis of physician–patient dialogue.AIDS, 13, 1771–1778.

Safran D. G., Taira D. A., Rogers W. H., et al., (November 1996).Primary care performance in traditional and managed care sys-tems. Presented at Annual Meeting of the American PublicHealth Association, New York.

Safran, D. G., Kosinski, M., Tarlov, A. R., Rogers, W. H., Taira, D.,Lieberman, N., and Ware, J. E. (1998). Primary Care Assess-ment Survey: Test of data quality and measurement perfor-mance. Medical Care, 36, 728–739.

Sanson-Fisher, R. W., Campbell E. M., Redman, S., and Hennikus,D. J. (1989). Patient–provider interactions and patient out-comes. Diabetes Education, 15, 134–138.

Simpson, M., Buckman, R., Stewart, M., Maguire, P., Lipkin, M.,Novack, D., and Till, J. (1991). Doctor–patient communication:The Toronto consensus statement. BMJ, 303, 1385–1387.

Patient–Clinician Relationships and Treatment System 101

Singh, N., Squier, C., Sivek, C., Wagener, M., Hong Nguyen, M., andYu, V. L. (1996). Determinants of compliance with antiretrovi-ral therapy in patients with HIV: Prospective assessment withimplications for enhancing compliance. AIDS Care, 8, 261–269.

Spire, B., Duran, S., Souville, M., Leport, C., Raffi, F., Moatti,J.-P., and the APROCO cohort study group. (2002). Adher-ence to HAART in HIV-infected patients. Social Sciences andMedicine, 54, 1481–1496.

Stewart, M. A. (1995). Effective physician–patient communicationand health outcomes: A review. Canadian Medical AssociationJournal, 152, 1423–1433.

Stewart, M. A., McWhinney, I., and Buck, C. W. (1979). Thedoctor–patient relationship and its effect upon outcome.

Journal of Royal College of General Practitioners, 29, 77–82.

The Measurement Group. (1997). Module P1: Adherenceto Therapy Interview. Retrieved from Http://www.themeasurementgroup.com/resource/protease int.htm.

World Health Organization (WHO). (1998). The WHO CIDI-SF, v1.0, November 1998. Retrieved from http://www.who.int/msa/cidi/cidisf.htm

Wutoh, A. K., Brown, C. M., Kumoji, E. K., Daftary, M. S., Jones,T., Barnes, N. A., and Powell, N. J. (2001). Antiretroviral ad-herence and use of alternative therapies among older HIV-infected adults. Journal of the National Medical Association,93, 243–250.