AIDS and Behavior

, Volume 11, Issue 3, pp 385–392

Medication Beliefs as Mediators of the Health Literacy–Antiretroviral Adherence Relationship in HIV-infected Individuals


  • Joseph Graham
    • Department of Family Medicine and Community HealthUniversity of Pennsylvania School of Medicine
  • Ian M. Bennett
    • Department of Family Medicine and Community HealthUniversity of Pennsylvania School of Medicine
  • William C. Holmes
    • Division of General Internal MedicineUniversity of Pennsylvania School of Medicine
    • Center for Clinical Epidemiology and BiostatisticsUniversity of Pennsylvania School of Medicine
    • Center for Health Equity Research and PromotionPhiladelphia VA Medical Center
    • Division of Infectious Diseases, Department of MedicineUniversity of Pennsylvania School of Medicine
    • Center for Clinical Epidemiology and BiostatisticsUniversity of Pennsylvania School of Medicine
Original Paper

DOI: 10.1007/s10461-006-9164-9

Cite this article as:
Graham, J., Bennett, I.M., Holmes, W.C. et al. AIDS Behav (2007) 11: 385. doi:10.1007/s10461-006-9164-9


Identifying modifiable barriers to antiretroviral adherence remains an important aim. We hypothesized that mistaken beliefs regarding taking HIV medications mediated the relation between low literacy and poor adherence. We studied 87 HIV-infected individuals on standard antiretroviral regimens for ≥ 3 months. Adherence was assessed using pharmacy refill records. Medication beliefs, including an individual’s norm for acceptable adherence, were measured using questions developed by expert panel. Literacy was associated with ≥95% adherence (64% for ≥9th grade level vs. 40% for <9th grade level). Participants with <95% adherence had a lower threshold of acceptable adherence than those with ≥95% adherence [80% adherence (interquartile range 70–90%) vs. 90% adherence (interquartile range 80–90%)]. However, the effect was independent of literacy. No other beliefs assessed were associated with adherence. Although the beliefs assessed do not mediate the relation between literacy and adherence, we identified low adherence norms as a potentially modifiable belief associated with adherence.


HIV-1Antiretroviral therapyAdherenceHealth literacy


The importance of medication adherence in HIV care has been well characterized; individuals with poor adherence have lower rates of virological suppression and higher rates of adverse events and death (Bangsberg et al., 2001; Gross, Bilker, Friedman, & Strom, 2001; Paterson et al., 2000; Wood et al., 2003). Furthermore, lapses in adherence have been postulated to be a major cause of developing resistance (Wainberg & Friedland, 1998). While several risk factors have been proposed and supported by empirical research (Reynolds, 2004), identifying more potentially modifiable factors would broaden the potential targets for successful interventions under development.

Low literacy is one of the factors associated with poor adherence in HIV (Kalichman, Ramachandran, & Catz, 1999). In general, patients with low literacy have less knowledge of the management and treatment of their own chronic diseases and have poorer disease outcomes (Berkman et al., 2004; Estrada, Martin-Hryniewicz, Peek, Collins, & Byrd, 2004; Hope, Wu, Tu, Young, & Murray, 2004; Schillinger et al., 2002; Wilson, Tchetgen, & Spiegelman, 2001). In addition, low literacy has been linked with incorrect perceptions about the goals of HIV medications and about HIV transmission risks (Kalichman & Rompa, 2000). Although rates of low literacy have not been established in people living with HIV/AIDS, up to 35% of patients under age 65 receiving care at urban public hospitals have inadequate literacy skills to read materials provided in the clinical setting (Bennett, Robbins, Al-Shamali, & Haecker, 2003; Davis et al., 1994; Gazmararian et al., 1999; Williams et al., 1995).

Erroneous beliefs regarding the use of HIV medications and their side effects may explain observational studies’ findings of low adherence rates in individuals with low literacy. Mistaken beliefs of how the medications should be taken could result in individuals making clinically significant errors in pill taking (Nielsen-Bohlman, Panzer, & Kindig, 2004). We hypothesized that individuals with mistaken beliefs about HIV medications would adhere at lower rates and that these beliefs would contribute to the relation between low literacy and poor adherence.


Participants and Procedures

We conducted a cross-sectional study of a non-random sample of HIV-infected individuals ≥18 years of age on highly active antiretroviral therapy for ≥3 months and cared for in one of two University of Pennsylvania HIV clinics in Philadelphia, PA between February and June 2003. Study participants were recruited sequentially by study team members on their arrival for regularly scheduled clinic appointments. A consent form was read to all participants and the form was then signed if the individual agreed to participate. Data collection was then conducted via standardized individual face-to-face questioning by interviewers who were not part of the clinical care team. Study participants were compensated with $10 for their time.


Adherence Measurement

Adherence was assessed via a validated time to pharmacy refill surrogate measure (Grossberg, Zhang, & Gross, 2004). Adherence to a single index drug over the prior 3 months was assessed as in prior work (Gross, Zhang, & Grossberg, 2005; Grossberg et al., 2004). Three months was chosen since it was period of adherence previously shown to be related to virological response and was a clinically relevant interval since individuals typically visit their physicians quarterly. The protease inhibitor (PI) or non-nucleoside reverse transcriptase inhibitor (NNRTI) component of the multidrug regimen was utilized as the index drug since there is a high correlation between adherence to one drug in a regimen with the other drugs (McNabb, Nicolau, Stoner, & Ross, 2003). For patients taking both a PI and NNRTI, the PI was the index drug. For patients taking two PIs including ritonavir, ritonavir was the index drug. For patients taking an abacavir-containing triple nucleoside regimen, abacavir was the index drug.

Adherence over the previous 3 months was defined as: (days supply dispensed/# days between refills) × 100% and could range from 0 to >100%. For 30-day prescriptions, three consecutive months were assessed. For 2-week prescriptions, seven refills covering 3 months were assessed. Therefore, the days’ supply was approximately 90 and only the time to refill varied substantially.

Literacy Measurement

The Rapid Estimate of Adult Literacy in Medicine (REALM) was administered to estimate the literacy level of participants (Davis et al., 1993). This standard instrument for assessing reading skill in the health context has been validated in populations with socioeconomic and demographic characteristics similar to this cohort (Berkman et al., 2004; Davis, Michielutte, Askov, Williams, & Weiss, 1998). The REALM is well accepted by patients and takes approximately 2–3 min to administer. Below 9th grade literacy (REALM ≤ 61) is lower than the level of most medical brochures used in the clinical settings (Berkman et al., 2004).

Medication Beliefs

We identified a series of potential medication beliefs using an expert panel approach by interviewing 12 University of Pennsylvania affiliated HIV clinicians and pharmacists who reported hearing mistaken beliefs raised by clinic patients. We constructed questions to assess for these beliefs, then piloted and refined the questionnaire among patients at the Philadelphia Veterans’ Affairs Medical Center HIV clinic, a site affiliated with the University of Pennsylvania but not included in the study. The population used for developing this instrument was socio-demographically comparable to that of the study population.

One of the beliefs of interest was whether participants’ had an inappropriately low threshold of acceptable adherence. Therefore, we asked them how many of ten doses they could miss before they got worried about it (see Appendix 1). We then calculated the percent adherence corresponding to the number of missed doses using the following formula: [1 − (# doses/10)] × 100%. We term this concept an individual’s “adherence norm” and results could range from 0 to 100%. In addition, we addressed seven other beliefs in two categories: (A) specific beliefs about how to take medications appropriately when routine dose-taking is disrupted and (B) general beliefs about medication effects and side effects (see Appendix 2). Category A (questions 2, 5, and 8) addressed mistaken beliefs that would directly result in errors in medication taking. Category B (questions 3, 4, 7, and 10) addressed mistaken beliefs that might be a marker of other beliefs not included in category A that would nonetheless result in errors in medication taking. Other items were included in the questionnaire (questions 1, 6, and 9) to decrease the likelihood of participants detecting the goals of the study. Patients who reported “agree” or “strongly agree” to the mistaken beliefs were categorized as having the belief.

Other Measures

Demographic variables assessed for possible confounding of the relation between beliefs and adherence included age, race, history of drug and alcohol use, cognitive function (Jones, Teng, Folstein, & Harrison, 1993), level of schooling completed, income, insurance type, and social supports. Medical factors assessed for possible confounding included current HIV viral loads and CD4 counts as well as prior and current psychiatric diagnoses (depression, psychotic disorders, etc.).

Data Analyses

Adherence was included as a continuous variable (0–100%) and dichotomized at ≥95% or not (Paterson et al., 2000). The proportion of participants with undetectable viral loads (i.e., <50 copies/ml) was compared between participants with good and poor adherence using the chi-squared test. We assessed the association between literacy and adherence with chi-squared tests using a REALM score cutoff of 61, representing a 9th grade reading level. We compared the adherence norm between those with and without ≥95% adherence using Wilcoxon rank sum tests. Similarly, we compared percent adherence between participants with and without the other beliefs, again using Wilcoxon rank sum tests. We then compared the REALM scores of participants with and without each of the beliefs individually using Wilcoxon rank sum tests. Stratified analyses were performed to test for potential confounding by baseline characteristics. If these characteristics were associated with either the exposure or outcome with P < 0.1, multivariable analyses were carried out using linear regression to test for and control for potential confounding.

We tested whether beliefs mediated the relation between literacy and adherence using logistic regression models. We first constructed a model for the association between the outcome of adherence and the exposure of low literacy. We then compared this model with ones that included as the mediating exposure variable beliefs that in bivariate analyses were associated with both low literacy and suboptimal adherence. We used a reduction in the odds ratio of >10% to determine whether the association between literacy and adherence was dampened by addition of potentially mediating variables to conclude that they were, in fact, mediators.

Power Considerations

The target sample size was calculated based on our goal of demonstrating whether differences in adherence exist between individuals with and without medication beliefs. Based on clinical experience and a survey of local providers, it was estimated that 20–33% of participants would exhibit at least one of the medication beliefs. Based on a prior study at the Philadelphia VA Medical Center that used the adherence measurement technique we used in this study (Gross et al., 2005), we estimated that the standard deviation of adherence would be 0.2. We targeted the enrollment of 100 individuals over the duration of this study so as to be able to detect between a 12 and 14% difference in percent adherence between participants with and without mistaken beliefs with 80% power and an alpha level of 0.05.


We enrolled 87 individuals into the study and closed recruitment thereafter due to slowing of accrual. The study population was comprised largely of low income African Americans and predominantly unemployed men. Participants with a ≥ 9th grade reading level were significantly more likely to be white, men who have sex with men, high school graduates, employed, and from higher income strata (Table 1). Slightly more than half of both the lower and higher literacy groups had undetectable viral loads. The higher literacy group had higher CD4 counts, but this difference did not achieve statistical significance.
Table 1

Participant characteristics by adherence


<95% adherence N = 42

≥95% adherence N = 45

Test statistic

Median age (interquartile range)

44 years (37–48)

46 years (37–53)

z = −1.12

Male sex

32 (76%)

33 (73%)

χ= 0.09, df = 1, N = 87



37 (88%)

31 (69%)

χ= 4.70,* df = 1, N = 87


5 (12%)

14 (31%)

Men who have sex with men

11 (26%)

16 (36%)

χ= 0.89, df = 1, N = 87

Injection drug use history

8 (19%)

7 (16%)

χ= 0.19, df = 1, N = 87

High school graduate

25 (60%)

31 (69%)

χ= 0.83, df = 1, N = 87


11 (26%)

14 (31%)

χ= 0.26, df = 1, N = 87

Income/year < $10,000/year

27 (64%)

21 (47%)

χ= 2.73, df = 1, N = 87

Median CD4 count (interquartile range)

303 cells/cm3 (163–537)

363 cells/cm3 (248–470)

z = −0.19

Undetectable viral load (<50 c/ml)

19 (45%)

33 (73%)

χ= 7.13,** df = 1, N = 87

P < 0.05

** P < 0.01

aOne individual was self-described “other race”

Relation between Literacy and Adherence

Medication adherence was a median of 95% (IQR 67–101%) over the 3-month period prior to the study date. Participants with ≥95% adherence were significantly more likely to have undetectable viral loads than participants with <95% adherence [33/45 (73%) vs. 19/42 (45%), χ2 = 7.13, df = 1, N = 87, P < 0.01].

Adherence was higher for participants with higher literacy. Twenty-eight (64%) individuals with ≥9th grade reading level had ≥95% adherence while 17 (40%) individuals with lower literacy exhibited this high level of adherence, χ= 5.06, df = 1, N = 87, P < 0.05. No confounding was identified.

Relation between Beliefs and Adherence

There was an association between a participant’s adherence norm and actual adherence. The 45 (48%) participants ≥95% adherent reported a median of 90% adherence before getting worried (IQR 80–100%) while the 45 (52%) participants <95% adherent reported a median of 80% adherence before getting worried (IQR 70–100%), Wilcoxon rank sum z = 2.64, P < 0.01. No confounding was identified.

Although adherence often was higher in participants without the other beliefs, there was no definitive and/or significant relationship between adherence and other beliefs (Table 2). In fact, for two of the questions, participants harboring the mistaken beliefs had higher adherence rates. None of the observed differences in adherence were statistically significant.
Table 2

Comparison of REALM scores and adherence between subjects with and without misconceptions


Misconception absent n (%)

Misconception present n (%)

Wilcoxon rank sum


REALM scorea

REALM scorea





Any mistaken belief

23 (26%)

64 (74%)


63 (60–64)

59 (43–64)

z = 1.76

99% (62–101%)

94% (68–101%)

z = 0.17

When I drink alcohol, beer, or wine close to the time I am supposed to take my dose of HIV medication, it is best for me to skip that dose

53 (61%)

34 (39%)


61 (48–64)

60 (51–63)

z = 0.70

98% (68–101%)

87% (67–101%)

z = 0.98

My HIV medication is more effective in fighting the virus if I take a break from taking it every once in a while

64 (74%)

23 (26%)


61 (46–64)

59 (54–65)

z = −0.53

97% (68–102%)

82% (67–101%)

z = 0.75

If my HIV medication is causing me to have side effects, it means that the medication won’t be effective in fighting the virus

64 (74%)

23 (26%)


62 (58–64)

53 (34–59)

z = 3.36**

95% (66–101%)

99% (77–101%)

z = −1.53

When I miss my dose of HIV medication by a few hours, it is best to skip it and wait for the next dose

56 (64%)

31 (36%)


61 (50–64)

59 (47–64)

z = 0.55

96% (67–101%)

95% (67–101%)

z = −0.28

If my HIV medication is not causing me to have side effects, it means that the medication is not effectively fighting the virus

78 (90%)

9 (10%)


61 (53–64)

53 (32–60)

z = 2.31*

96% (67–101%)

88% (71–101%)

z = 0.46

When I miss my dose of HIV medication completely, it is best to double the next dose

82 (94%)

5 (6%)


61 (53–64)

36 (22–53)

z = 1.86

95% (67–101%)

99% (88–102%)

z = −0.99

If my HIV medication is causing me to have side effects, it is best to decrease the amount of medication I take that day

77 (89%)

10 (11%)


61 (53–64)

54 (22–61)

z = 2.17*

95% (67–101%)

94% (73–105%)

z = −0.95

P < 0.05

** P < 0.01

aValues represent median and interquartile (25–75%) ranges

Relation between Literacy and Beliefs

Participants with higher literacy had higher adherence norms (median 90%, IQR 80–90%) than those with lower literacy (median 80%, IQR 70–90%), though the difference only approached statistical significance (Wilcoxon rank sum z = 1.87, P = 0.06). Table 2 displays the relation between literacy and the other beliefs and the relation between these beliefs and adherence. Mistaken beliefs addressed in the questionnaire were common with 64 (74%) having at least one mistaken belief. REALM scores were lower for individuals with each of the mistaken beliefs. These differences were statistically significant for three of the mistaken beliefs: (1) presence of side effects signifies lack of efficacy, (2) absence of side effects signifies lack of efficacy, and (3) if side effects occur, decreasing the dose is an appropriate action; and the difference in REALM score approached statistical significance for a fourth belief: after missing a dose, doubling the next dose is an appropriate action.

Assessment of Mediation of Literacy–Adherence Relationship

Given the marginal association between literacy and adherence norms as well as the stronger association between adherence norms and actual adherence, we tested whether adherence norms mediated the relation between low literacy and poor adherence. Table 3 displays the unadjusted and adjusted models. When both higher literacy and higher adherence norms were included together in the model, both remained associated with <95% adherence with only minimal differences in the point estimates compared to the unadjusted models and the confidence intervals remained nearly the same, although now marginally crossing unity. However, the association between the adherence norm and observed adherence decreased by approximately 11% and therefore we conclude that the adherence norm contributes only minimally to the mediation of the association between literacy and adherence. Other beliefs were not assessed given their lack of association with adherence and thus lack of potential mediating effect.
Table 3

Regression models assessing adherence norm as mediator of literacy:adherence association

Variable tested for association with ≥95% adherence

Unadjusted model (95% confidence interval)

Adjusted modela (95% confidence interval)

≥9th grade literacy

2.67 (1.13–6.37)

2.38 (0.98–5.79)

Adherence norm (per 10% increase)

1.46 (1.02–2.08)

1.40 (0.98–2.00)

aThe adjusted model includes literacy and adherence norm as the independent variables


This study demonstrates an important link between an individual’s personal norm for acceptable adherence and actual adherence. This finding is important not only as a potential screening tool for identifying individuals at increased risk of non-adherence, but also as a potentially modifiable risk factor that may be amenable to interventions. While we found that several other mistaken beliefs about HIV medications were common, neither the specific behavior-related questions (category A) nor the general beliefs questions (category B) were predictive of adherence behavior.

This study again confirms the strong relation between lower literacy and suboptimal adherence to medications for HIV infection (Kalichman et al., 1999). Additionally, we further confirmed the validity of pharmacy refill data as a measure of adherence given the relation between the pharmacy measure and viral load in this study. Importantly, this study was conducted using ad hoc direct contact of local pharmacies rather than an automated pharmacy database. This demonstrates the feasibility of contacting pharmacies to obtain refill data for research, at least on the relatively small scale of this study.

Our findings that medication beliefs are associated with low literacy also are consistent with those of Kalichman et al. who found that knowledge was lower in those of lower literacy in HIV (Kalichman & Rompa, 2000; Wolf et al., 2005). This phenomenon of inadequate adherence to medical regimens has been described in other disease settings as well (Davis et al., 1998; Schillinger et al., 2002). However, despite the association between literacy and adherence and literacy and beliefs, we only found minimal evidence that the adherence norm somewhat mediates the relation between literacy and adherence.

Other potential, but untested mediators of this association include less detailed information provided to individuals with lower literacy by providers and therefore less knowledge of importance of adherence (Kalichman et al., 1999), lack of acceptance of information furnished by providers (Wilson, Hutchinson, & Holzemer, 2002), lack of basic organizational skills in individuals with low literacy, and more chaotic lifestyles due to daily stresses associated with insufficient literacy. Low literacy confers a number of individual characteristics which create obstacles to direct patient–physician communication and could in turn increase the likelihood of harboring mistaken beliefs (Bennett, Switzer, Aguirre, Evans, & Barg, 2006; Wolf et al., 2005). Adults with low literacy are also at increased risk of depressive symptomatology (Gazmararian et al., 1999) which has been shown to be associated with reduced adherence to medical therapy (DiMatteo, Lepper, & Croghanet, 2000). However, no studies have assessed whether these factors mediate the relation between low literacy and adherence. Therefore, the mediators of the relation between literacy and adherence remain to be determined.

This study has several potential limitations. Pharmacy refill data can only be considered a surrogate for true adherence behavior. However, several studies to date have demonstrated the validity of such measures (Grossberg et al., 2004; Wood et al., 2003). Selection bias is possible in that we only enrolled individuals who were prescribed antiretroviral therapy for at least 3 months, but had not been tested over other periods. It is possible that an association between the drug side effects beliefs and adherence truly exists, but that we excluded the individuals with both side effects and side effect beliefs who would have demonstrated the relation between mistaken beliefs and poor adherence that we hypothesized. Therefore, our results should only be generalized to individuals who are already tolerating the medications over their initial 3-month period. The study was also potentially limited by not achieving our target sample size and associations that nearly achieved statistical significance may in fact represent true associations that we were underpowered to detect. However, given the observed standard deviation of 0.27, for mistaken beliefs present in 25% of participants, we had sufficient power to detect at least a 19% difference in adherence, only 5% higher than we designed the study to detect. Of course, if the effect of these mistaken beliefs was more modest, we would be less likely to detect them. All observational studies such as this one are potentially limited by residual confounding by factors not considered such as other barriers to adherence (e.g., lack of access to transportation for picking up refills).

We did not test an exhaustive battery of potential mistaken beliefs. The beliefs we did assess were derived from providers and pharmacists, which may not have reflected the full range of mistaken beliefs that patients have. It is possible that other mistaken beliefs of which we are unaware mediate the relation between lower literacy and adherence. Further qualitative research is warranted to explore more fully the nature and influence of patient beliefs on medication use.

In conclusion, assessing an individual’s personal adherence norm holds promise for identifying individuals at higher risk of non-adherence. Because these adherence norms may be modifiable, they should be considered for inclusion in interventions to increase antiretroviral adherence. Further studies of the mechanism by which low literacy increase an individual’s risk of non-adherence are needed so as to better inform future interventions.


This project was supported (or supported in part) by an Agency for Healthcare Research and Quality Centers for Education and Research on Therapeutics cooperative agreement (grant # HS10399), the PENN AIDS Clinical Trials Unit (U01-AI32783), the PENN Center for AIDS Research (P30-AI45008), and National Institutes of Mental Health grant MH-01854. Dr Holmes is supported by a Research Career Development award from the Health Services Research and Development Service of the Department of Veterans Affairs.

Copyright information

© Springer Science+Business Media, LLC 2006