Neurocognitive Functioning is Associated with Self-Reported and Performance-Based Treatment Management Abilities in People Living with HIV with Low Health Literacy

Abstract

Objective

People living with HIV (PLWH) are at higher risk for poorer neurocognitive functioning and health literacy than uninfected persons, which are associated with worse medical outcomes. Aging research suggests that the effect of neurocognitive functioning on health outcomes may be more pronounced in those with low health literacy. We aimed to determine whether low health literacy might amplify the adverse effects of neurocognitive functioning on treatment management outcomes in 171 PLWH aged 40+.

Method

In this cross-sectional, observational study, participants completed a well-validated battery of neurocognitive, health literacy, and treatment management measures. A binary health literacy variable (low vs. adequate) was determined via established cut points on the well-validated health literacy tests. Treatment management outcomes included biomarkers of HIV (i.e., CD4 counts and viral load), self-management of HIV disease (i.e., self-reported medication adherence and self-efficacy for HIV disease management), and performance-based health-related decision-making.

Results

Forty-seven percent of the sample met the criteria for low health literacy. Multivariable regressions adjusting for clinicodemographic (e.g., race, socioeconomic status) covariates revealed significant interactions for self-efficacy for HIV disease management and health-related decision-making, such that neurocognitive functioning was associated with these outcomes among those with low, but not adequate health literacy.

Conclusions

Findings suggest that low health literacy may increase the vulnerability of PLWH to the adverse effects of neurocognitive impairment on health outcomes, or conversely that adequate health literacy may provide a buffer against the health risks associated neurocognitive impairment. Interventions targeting health literacy in PLWH may mitigate the effects of neurocognitive impairment on health outcomes.

Introduction

People living with HIV (PLWH) are at a higher risk for both poorer neurocognitive functioning (Heaton et al., 2010) and lower health literacy (Kalichman & Rompa, 2000) than the general population, both of which are independently associated with worse medical outcomes (Becker, Thames, Woo, Castellon, & Hinkin, 2011; Jacks et al., 2015; Kordovski, Woods, Avci, Verduzco, & Morgan, 2017; Rebeiro et al., 2018) including nonadherence to HIV medications. Low health literacy and poorer neurocognitive functioning are also often comorbid in PLWH (Morgan et al., 2015). Lower health literacy demonstrated by PLWH compared to HIV seronegative counterparts is likely a product of risk factors that are more common in PLWH (e.g., low socioeconomic status), as well as greater neurocognitive dysfunction in this population. For example, the deleterious effects of HIV-associated neurocognitive disorders (HAND) on both fundamental (e.g., basic knowledge and numeracy) and critical (e.g., ability to comprehend, appraise, and apply health information) aspects of health literacy have been demonstrated (Morgan et al., 2015). This is consistent with work in the larger aging literature, suggesting that health literacy declines with increasing age and that higher neurocognitive function may protect against such decline (Geboers et al., 2018; Kobayashi, Wardle, Wolf, & von Wagner, 2015), particularly in the domains of memory (Federman, Sano, Wolf, Siu, & Halm, 2009; Geboers et al., 2018; Kobayashi et al., 2015) and executive functions (Federman et al., 2009; Kobayashi et al., 2015). Further, the association of low health literacy with poorer health outcomes such as quality of life, depression, and mortality is well-established in the broader aging literature, and there is evidence that neurocognitive function may explain some or all of this association between low health literacy and outcomes (Serper et al., 2014).

The process-knowledge model of health literacy (Chin et al., 2011) offers a framework to explain the overlapping yet distinct pathways of poor neurocognitive function and low health literacy, and ultimately why not all individuals with cognitive impairment exhibit poor health literacy (and vice versa). This model posits two fundamental components of health literacy: processing capacity (fluid cognitive abilities such as processing speed and working memory) and knowledge (crystalized abilities such as general knowledge [e.g., vocabulary] and health-specific knowledge). Higher knowledge may counter deficits in comprehension of health information associated with poorer processing capacity. Several studies support this model (Chin et al., 2017  , 2011), showing that the influence of processing capacity on performance on health-related tasks is greater for those with low knowledge, whereas those with high knowledge perform well regardless of processing capacity level. Thus there is a reason to suspect that level of health literacy could modulate the adverse influence of neurocognitive impairment on health outcomes (e.g., treatment management) in PLWH.

Little is known about the association of neurocognitive functioning on treatment outcomes in the context of low health literacy in PLWH. In injecting drug users with HIV, low health literacy and neurocognitive impairment showed a combined negative effect on medication nonadherence (Waldrop-Valverde, Jones, Weiss, Kumar, & Metsch, 2008). Another study by this group highlighted the interrelatedness of health literacy and higher-order cognitive abilities, showing that health literacy measures shared common traits with neurocognitive measures (e.g., executive skills), which together were associated with medication self-management in PLWH (Waldrop-Valverde, Jones, Gould, Kumar, & Ownby, 2010). Work has shown that PLWH with HAND performed poorer on performance-based tasks of health-related decision-making than those without HAND, and further, that health literacy and neurocognitive functioning were independent predictors of this important HIV outcome (Doyle et al., 2016). Finally, low health literacy was shown to partially mediate the effects of neurocognition in using online health resources (Woods & Sullivan, 2019). At present, however, no prior studies have examined whether low health literacy modulates the established effects of poor neurocognitive functioning on treatment management in PLWH.

Assessment of everyday health functioning (e.g., medication adherence, engagement and retention in healthcare) is an essential component of a clinical neuropsychological evaluation of persons with chronic medical conditions in the modern era (e.g., Bilder, 2011). Indeed, deficits in both lower- (e.g., motor skills) and higher-order (e.g., memory, executive functions) neurocognitive functions are risk factors for impairment in a variety of different aspects of everyday health functioning in various clinical populations (e.g., Gorman, Foley, Ettenhofer, Hinkin, & van Gorp, 2009). However, the pathway from neuropsychological deficits as measured by clinical tasks to successfully accomplishing health-related tasks in daily life is highly complex and multi-determined. This pathway may be a function of a variety of potentially critical mediating and moderating factors, including functional capacity, the demands of the everyday functioning tasks themselves, motivation, insight, psychiatric comorbidity, and the use of compensatory factors (see Casaletto, Weber, Iudicello, & Woods, 2017). In this study, we examine the potential moderating role of health literacy, which describes one’s ability to obtain, process, understand, and use health-related information (US Department of Health and Human Services, 2010) and may be an important determinant in whether neurocognitive deficits affect everyday health functioning in persons with chronic medical conditions.

Thus, given the demonstrated risk for low health literacy and poorer neurocognitive functioning in PLWH, evidence of an association between these two domains, and their association with health outcomes in HIV and in the larger aging literature, the purpose of this study was to examine whether there is a synergistic association between neurocognitive functioning and low health literacy on HIV treatment management outcomes. In other words, does the contribution of neurocognitive function to health outcomes differ for those with adequate versus low health literacy? Given the previously reviewed literature, it is plausible that high levels of health literacy could provide a buffer against the adverse effects of neurocognitive impairment on treatment management and outcomes in PLWH; or viewed differently, low levels of health literacy could amplify the negative impact of neurocognitive dysfunction on treatment management and outcomes. We examined a broad range of treatment management outcomes in the current study including biomarkers of HIV (i.e., CD4 counts and viral load), self-management of HIV disease (i.e., self-reported medication adherence, self-efficacy for HIV disease management), and performance-based health-related decision-making.

Methods

Participants and Procedure

The current cross-sectional, observational study included 174 PLWH aged 40 years and older (range 40–73) recruited from a University HIV/AIDS clinic in the Southeastern United States between September 2016 and November 2018. Primary HIV risk factors for the sample included men who have sex with men (37%), heterosexual sex (50%), and injection drug use (13%). Exclusion criteria were determined via a self-report telephone screen and included major conditions that may affect neurocognitive functioning (e.g., schizophrenia, Alzheimer’s disease), significant visual or auditory deficits, head injury with loss of consciousness over 30 m, and current chemo- or radiation therapy. After providing written, informed consent at the study visit, participants completed a comprehensive assessment including measures of neurocognitive functioning, treatment management, and health literacy. Table 1 displays the sample’s clinicodemographic characteristics. Three participants were missing item-level data for the health literacy measures, thus the final analyzed sample was 171 (see Table 1).

Measures

Independent variables. Neurocognition. Global neurocognitive functioning was determined with a comprehensive, well-validated battery of clinical measures that is routinely used in the NeuroAIDS field (Carey et al., 2004; Heaton et al., 2010). This battery assesses the domains of verbal fluency (Controlled Oral Word Association Test and Animals and Action Fluency), executive functions (Wisconsin Card Sorting Test, Trail Making Test B), learning and delayed recall (two domains [Hopkins Verbal Learning Test and Brief Visuospatial Memory Test]), speed of information processing (Trail Making Test A, Digit Symbol Task, Symbol Search), working memory (Letter Number Sequencing, Paced Auditory Serial Addition Test), and fine motor skills (Grooved Pegboard Test dominant and nondominant hands). Raw test scores were demographically adjusted using the most appropriate normative corrections for age, gender, education, and race/ethnicity when available and then grouped and converted to global and domain deficit scores (Carey et al., 2004). These scores reflect the number and degree of impaired test performances, with higher scores indicating poorer neurocognition. Global deficit scores greater than or equal to 0.50 are indicative of neurocognitive impairment (Carey et al., 2004).

Health literacy. Health literacy was ascertained by four widely used measures that have well-validated cut points for identifying persons with low health literacy: the Single-Item Literacy Screener (SILS) (Morris, MacLean, Chew, & Littenberg, 2006), the Newest Vital Sign (NVS) (Weiss et al., 2005), the Test of Functional Health Literacy in Adults (TOFHLA) (Parker, Baker, Williams, & Nurss, 1995), and the Rapid Estimate of Adult Literacy in Medicine (REALM) (Davis et al., 1993). These measures were also chosen because they capture both fundamental and critical aspects of health literacy, including self-efficacy (i.e., SILS), word knowledge and understanding (i.e., TOFHLA and REALM), numeracy (i.e., TOFHLA and NVS), and appraisal (i.e., NVS). The SILS includes one item (“How often do you need to have someone help you when you read instructions, pamphlets, or other written material from your doctor or pharmacy?”), with scores ranging from 1 to 5 and scores > 2 (Morris et al., 2006) reflecting low health literacy. The NVS is a six-item measure with an ice cream label stimulus to which participants answer questions such as “Pretend that you are allergic to the following substances: Penicillin, peanuts, latex gloves, and bee stings. Is it safe for you to eat this ice cream?” and “If you eat the entire container, how many calories will you eat?”. Scores on the NVS range from 0 to 6, with scores <4 (Weiss et al., 2005) reflecting low health literacy. The TOFHLA includes reading (50 items) and numeracy (17 items) subtests, with numeracy scores weighted to a 50-point scale such that total scores range from 0 to 100 and scores < 75 (Colbert, Sereika, & Erlen, 2013; Navarra, Neu, Toussi, Nelson, & Larson, 2014) reflect low health literacy. In the reading subtest, participants are given passages of text about medical topics with missing words in which they must select the word most appropriate to the context of the passage from a multiple-choice list of options. In the numeracy subtest, participants are shown stimuli with medical information or instructions (e.g., instructions on a prescription label) to which they answer questions testing their understanding of the information in the scenarios. The REALM is a 66-item list of medical words that participants read aloud, with scores ranging from 0 to 66 (cut score < 61; (Colbert et al., 2013).

Given that there is no consensus on how to operationalize low health literacy and most studies use a single measure with a single cut point, which is suboptimal from a psychometric standpoint (e.g., Brooks & Iverson, 2010), we adopted a battery approach to measuring health literacy that included a diverse set of well-validated measures of both verbal and nonverbal health literacy, as well as both basic, lower-order and integrative, higher-order tasks. Consistent with the Frascati criteria for diagnosing HAND (and other diagnostic systems), participants with two or more scores in the impaired range based on established cut points were classified with low health literacy (Low HL), whereas those who did not meet these criteria were classified as having adequate health literacy (Adequate HL). This approach avoids the Type I error base rate problem that accompanies approaches that only use one impaired test. At the same time, it helps to minimize Type II error by acknowledging that not all test in a given domain need to be impaired for a domain to be considered deficient. Forty-seven percent of the current sample met the criteria for Low HL (49%, 35%, and 16% of which scored below established cut points on 2, 3, and 4 HL measures, respectively) leaving 53% who were classified as having Adequate HL (41% of whom scored below established cut points on 1 HL measure).

Covariates. Given the complexity of the clinical factors under study, several variables conceptually related to the constructs under study in this manuscript were considered as covariates for the primary analyses. Sociodemographics were determined via a self-report questionnaire and included age, sex, race, years of education, and yearly household income (i.e., 1 = $0–$10,000 to 11 = $100,001 and above). The oral reading subtest of Wide Range Achievement Test 4th edition (WRAT (Wilkinson & Robertson, 2006)) provided an estimated verbal IQ and is often considered a proxy for quality of education (Manly, Jacobs, Touradji, Small, & Stern, 2002). Given the conceptual and statistical (average Spearman’s rho = 0.45) overlap between education, WRAT, and income, an SES composite was created as the sample-based Z-score for all three indices for use in subsequent analyses to yield more parsimonious models, with higher values reflecting higher SES. The Center for Epidemiological Studies Depression Scale (CES-D; Radloff, 1977) assessed the frequency of depressive symptomology over the past week. CES-D scores greater than or equal to 16 are considered to represent clinically relevant symptoms of depression. Finally, urine samples for toxicology (TransMed®) were obtained in the morning of the study visit and tested to determine if participants were positive for the following illicit substances: amphetamines, methamphetamine, opiates, and cocaine.

Dependent variables . Biomarkers of HIV disease. HIV disease markers were derived from clinic records and included CD4 T-lymphocyte count (current and nadir) and plasma viral load (detectable > 20 copies/milliliters vs. undetectable ≤ 20 copies/milliliters) values that were collected closest to the current study visit (i.e., within 3 months prior or within 1 month after).

Self-management of HIV disease. Participants completed four questionnaires assessing HIV disease self-management. The first self-report measure was a visual analog scale (VAS) of medication adherence (Finitsis, Pellowski, Huedo-Medina, Fox, & Kalichman, 2016; Giordano, Guzman, Clark, Charlebois, & Bangsberg, 2004; Walsh, Mandalia, & Gazzard, 2002), commonly used in PLWH. Participants are presented with a horizontal line with a 0% (“I never took my medication in the past month”) at the beginning and a 100% (“I always took my medication in the past month”) at the end and are instructed to place a mark on the line to reflect how adherent they have been to their HIV medications over the last 30 days.

The Beliefs Related to Medications (BERMA) (McDonald-Miszczak, Maris, Fitzgibbon, & Ritchie, 2004) survey is a 53-item measure assessing self-efficacy for Memory for Medications (e.g., “I am good at remembering to take my medications.”) and Dealing with Health Professionals (e.g., “I am good at asking questions about my medical conditions.”), as well as Attitudes About Medications (e.g., “The medications my doctor prescribed to me are appropriate for my medical conditions.”). Participants rate each item on a Likert scale from 1 (“strongly disagree”) to 5 (“strongly agree”) such that higher scores reflect better abilities.

The HIV Treatment Adherence Self-Efficacy Scale (ASES) (Johnson et al., 2007) is a 12-item measure assessing self-efficacy for adhering to one’s “HIV treatment plan,” which includes—but is not limited to—antiretroviral medications. ASES items reflect two main themes: integration of treatment into one’s daily life (e.g., “In the past month, how confident have you been that you can: Integrate your treatment into your daily routine?”) and perseverance in the face of treatment challenges (e.g., “In the past month, how confident have you been that you can: Continue with your treatment even when you are feeling discouraged about your health?”). Participants are asked to rate their confidence in each of the items over the past month on a scale ranging from 0 (“cannot do at all”) to 10 (“completely certain can do”), such that higher scores reflect greater self-efficacy.

Finally, the Perceived HIV Self-Management Scale (PHIVSMS) (Wallston, Osborn, Wagner, & Hilker, 2011) is an eight-item measure of disease self-management self-efficacy (e.g., “I succeed in the projects I undertake to manage my HIV infection.”). Participants rate each item on a Likert scale from 1 (“strongly agree”) to 5 (“strongly disagree”), with reverse scoring of positive items such that higher scores reflect better abilities.

Given the conceptual and statistical overlap between these three self-efficacy (BERMA, ASES, PHIVSMS) measures (mean Spearman’s rho = 0.54, ps < 0.001), a composite was created for analyses to reflect the construct of self-efficacy of HIV disease management. Specifically, average scores for each of the three measures were divided by the total possible score for each test and multiplied by 100. These individual scores were then averaged, resulting in a composite score on a 100-point scale, with higher values reflecting greater levels of self-efficacy for HIV disease management.

Health-related decision-making. The Modified University of California San Diego (UCSD) Brief Assessment for Capacity to Consent (UBACCT) (Burton et al., 2012; Doyle et al., 2016) is a performance-based task used to assess health-related decision-making abilities. In this task, participants read a hypothetical scenario in which their friend is terminally ill and diagnosed with pneumonia. Participants are instructed to apply information from the scenario (e.g., side effects) to decide whether or not to treat the pneumonia with antibiotics. Participants are administered with 10 questions related to both assessing (e.g., “How will the choice affect Pat’s health?”) and comprehending (e.g., “Please describe some of the risks or discomforts of the treatment.”) the health information. Scores range from 0 to 19, with higher scores reflecting better performance. Although successful performance on the UBACCT inherently requires better health literacy, this medical decision-making task captures a higher-order health behavior that is conceptually downstream from—though ostensibly related to—critical health literacy. Medical decision-making is a multi-faceted construct, similar to constructs such as medication adherence, which although reliant on fundamental and critical health literacy, also involves nonliteracy factors such as values, attitudes, and neuropsychological functions (Doyle et al., 2016).

Statistical Analyses

We first performed collinearity diagnostics on several conceptually related study variables, as well as examining normality of the continuous dependent variables. Next, to yield parsimonious models, a series of exploratory analyses were conducted to determine data-driven covariates for subsequent main analyses from a pool of clinicodemographic variables from Table 1 which are conceptually related to both health literacy and our outcomes. Specifically, we used t-tests, chi-square tests, or correlations to examine associations between these potential confounders and the HL group variable as well as the dependent variables (i.e., HIV biomarkers [current and nadir CD4, viral load detectability], self-management of HIV disease [VAS medication adherence and the self-efficacy for HIV disease management composite], and health-related decision-making [UBACCT]). We also examined bivariate associations between HL group and neurocognition and the dependent variables using the same approach. Table 2 includes Spearman’s rho associations among study variables for visualization purposes. Those variables associated at p < .05 with either HL group or the dependent variable were entered as covariates in the respective multiple regression models. Thus, six multiple regressions were conducted, including relevant covariates and the independent variables of interest: neurocognitive functioning, health literacy group (Low vs. Adequate), and their interaction. A critical alpha of 0.05 was used, and all analyses were conducted in JMP Pro version 14.

Results

Exploratory Analyses for Determination of Covariates

Given the conceptual overlap between some of the individual health literacy measures comprising our HL grouping variable (e.g., numeracy, reading) and our neurocognitive domains as well as WRAT, we examined correlations between these measures to determine multicollinearity. For neurocognition, the highest correlations were between the global deficit scores and TOFHLA reading (rho = −0.35) and verbal fluency and REALM (rho = −0.35), suggesting that there was not multicollinearity and these measures were indeed capturing distinct constructs. Although WRAT had higher associations with these measures (average rho = 0.66), particularly with TOFHLA reading (rho = 0.79) and REALM (rho = 0.81), our approach to including WRAT in our models (as part of the SES composite) is both conceptually relevant and a conservative approach allowing for better examination of the role of health literacy versus basic nonhealth literacy. Furthermore, post hoc analyses of our main analyses confirmed that using an SES composite without WRAT still yielded the same pattern of results.

Examination of distributions for our continuous dependent variables showed that although all variables had significant (p < .05) Shapiro-Wilks tests, the skewness and kurtosis values were all in an acceptable range (−1 to 1) for the self-efficacy for HIV disease management composite, health-related decision-making, current and nadir CD4. Given that VAS was particularly skewed (51% of the sample reported 100% adherence), we created a dichotomous VAS variable, reflecting those with 100% adherence versus < 100% adherence, and our pattern of results subsequently was not changed. Therefore, given that regression is robust to violations of normality (Schmidt & Finan, 2018), all continuous dependent variables were not transformed.

Demographic, clinical, and health literacy characteristics of the study sample are displayed in Table 1. Participants with Low HL were significantly more likely to be African–American (p < .01), to have a positive urine drug screen (p < .01), and to have fewer years of education, lower estimated verbal IQ, less income, as well as lower SES composite scores (ps < 0.01) than those with Adequate HL. At the bivariate level, PLWH with Low HL obtained significantly poorer scores across global (p < .001, Cohen’s d = 0.50) and all domain-level neurocognitive deficit scores (ps < 0.05, mean range of Cohen’s d = 0.31 [learning] – 0.64 [verbal fluency]), except recall (p = .58) and motor (p = .59). Likewise, PLWH with Low HL had significantly lower scores on the self-efficacy of HIV disease management composite (p < .0001, d = 0.82) and the health-related decision-making task (p < .0001, d = 1.32). The VAS for medication adherence (p = .86, d = 0.03) was not significantly different by HL group. Analysis of HIV biomarkers showed that PLWH with Low HL had lower nadir CD4 at the trend level (p = .05, d = 0.32) but did not differ from PLWH with Adequate HL on current CD4 (p = .40, d = 0.15) or the frequency of detectable HIV RNA (p = .46, odds ratio = 1.30).

Hypothesis-Driven Analyses

Finally, we used multiple regression to examine the main effects of health literacy group, neurocognition, and their interaction on each of the HIV disease management outcomes (see Table 3). Each model adjusted for confounders that were associated with either health literacy group (i.e., race, SES, urine screen positivity) or the HIV disease management outcome. The omnibus regression models for nadir CD4 and viral load detectability were not significant, nor were there any significant covariates or a health literacy group × neurocognition interaction (all ps > 0.05). For current CD4, the omnibus model was significant (p = .046), and a health literacy group × neurocognition interaction did not emerge; however sex emerged as a significant predictor, with men having lower current CD4 counts (B = −81.35, p = .02). For VAS the omnibus model was also significant (p = .042), and a health literacy group × neurocognition interaction did not emerge; however SES emerged as a significant predictor, with higher SES associated with higher VAS scores (B = 1.50, p = .03).

The multiple regression model predicting the self-efficacy of HIV disease management composite was significant F(7,158) = 9.51, p < .001, adjusted R² = 0.27. Main effects were observed for health literacy group and depression (ps < 0.001), but not for SES (p = .21), race (p = .77), urine screen positivity (p = .40), or neurocognition (p = .39). However, these main effects were tempered by a significant interaction between health literacy group and neurocognition (β = −0.14, B = −3.19, 95% confidence interval = –0.11 to –6.27, p = .04). To interpret this interaction, a follow-up analysis of the prior model was conducted stratified by health literacy group, which revealed that poorer neurocognitive functioning was associated with poorer self-efficacy for HIV disease management in persons with Low HL (β = −0.24, B = −4.82, p = .04) but not in those with Adequate HL (β = 0.08, B = 1.72, p = .42).

A similar pattern of results was obtained for health-related decision-making, which yielded an overall significant model F(6,160) = 18.27, p < .001, adjusted R² = 0.38. In this model, main effects were observed for health literacy group and SES (ps < 0.001), but not race (p = .55), urine screen positivity (p = .81), or neurocognition (p = .09). However, the interaction of health literacy group and neurocognition was significant (β = −0.14, B = −1.07, 95% confidence interval = –0.08 to –2.06, p = .03). Follow-up analysis stratified by health literacy group revealed that poorer neurocognitive functioning was associated with poorer health-related decision-making in PLWH with Low HL (β = −0.27, B = −1.73, p = .02), but not in those with Adequate HL (β = 0.04, B = 0.27, p = .70) (Figs. 1 and 2).

Discussion

It is well documented that PLWH are at greater risk for both low health literacy and neurocognitive dysfunction than the general population. Although some work in HIV and in the larger aging literature suggests these risk factors may interact to disrupt health outcomes, little work has focused on this topic in PLWH across diverse outcomes using well-validated neurocognitive functioning and health literacy measures. The current study aimed to examine whether there was a synergistic association between low health literacy and neurocognitive functioning on several treatment outcomes in PLWH. These included clinical biomarkers, subjective treatment management abilities, and performance-based health-related decision-making.

Consistent with prior work in PLWH, our preliminary analyses showed that those with low health literacy were more likely to be African–American, active substance users, have lower SES, and have poorer neurocognitive functioning (e.g., Morgan et al., 2015; Woods & Sullivan, 2019). In fact, five of seven neurocognitive domains as well as global neurocognitive functioning were associated with low health literacy at the bivariate level with small-to-medium and medium-to-large overall effect sizes, which speaks to the multidimensional nature of health literacy, particularly the measures and operationalization we used to define low health literacy. Low health literacy also mapped onto several diverse and important treatment management outcomes at the bivariate level, including self-management of HIV disease and performance-based health-related decision-making, with large effect sizes.

Results from our multivariable models revealed that in the context of clinicodemographic factors, low health literacy remained a significant correlate of a self-management of HIV disease index and performance-based health-related decision-making task, whereas global neurocognitive functioning was not an independent predictor. However, in both models, a synergetic effect of neurocognitive functioning and low health literacy emerged, such that neurocognitive functioning was associated with both outcomes in those with low health literacy but not among those with adequate health literacy. This finding suggests that in PLWH with low health literacy, poor neurocognitive functioning may be more detrimental to treatment management. On the other hand, adequate health literacy may act as buffer against cognitive limitations. Findings are consistent with the Process-Knowledge Model of Health Literacy (Chin et al., 2017,, 2011), which posits that higher knowledge may counter deficits in comprehension of health information associated with poorer processing capacity. In other words, possessing better health literacy and knowledge may make comprehension processes more efficient and less reliant on fluid neurocognitive resources. Yet for those with low health literacy, they may rely more on neurocognitive skills to achieve better health outcomes.

These results have several implications for research and practice. First, findings suggest several potential avenues of intervention. Interventions targeting health literacy in PLWH may mitigate the effects of poor neurocognitive functioning on treatment outcomes in this vulnerable population. For example, health literacy education and communication techniques that make minimal demands on processing capacity may be most beneficial (e.g., designed with the use of salient headers that reduce need for effortful search and organization processes). Indeed, work in PLWH (Waldrop-Valverde et al., 2010) suggests that merely lowering readability in such interventions are not likely to be successful, given the complex cognitive and metacognitive processes that underlie navigating health-related scenarios. On the other hand, cognitive interventions (e.g., cognitive remediation therapy, compensatory strategies) may bolster neurocognitive functioning and thus may buffer the negative effects of low health literacy on outcomes. However, the most effective intervention approach may be one that directly targets both risk factors. For example, psychoeducational approaches targeting health literacy that incorporate mnemonic techniques grounded in applied cognitive psychology paradigms may offer a valuable avenue for intervention (Avci et al., 2017).

Second, this study lends support for incorporating brief neurocognitive functioning and health literacy screenings into scheduled medical visits. Despite the increased risk of impairment in both of these interrelated domains in PLWH and their association with HIV-related health outcomes, impaired neurocognitive functioning and low health literacy can often be overlooked during medical visits unless the patient exhibits gross impairments. The information gained from screening for such impairments can be used by the medical providers to further assess patients including asking direct questions about neurocognitive functioning and medication adherence, for example. Health literacy in particular may be assessed with brief screeners such as the NVS, which is performance-based, has validated cut scores, and takes less than 3 m to administer and score. The NVS shows evidence of validity in PLWH (Kordovski et al., 2017) and has demonstrated associations with neurocognitive functions (Morgan et al., 2015), performance-based functional tasks (Woods et al., 2016), and HIV visit adherence (Fazeli, Woods, Gakumo, Mugavero, & Vance, 2019). Thus, given the increased risk of neurocognitive functioning in the aging population living with HIV and the synergistic effects of neurocognitive functioning and health literacy, the practice of screening is warranted.

Although this study has many strengths, including a comprehensive and well-validated battery of neurocognitive and health literacy measures, a diverse pool of HIV-related health outcomes, and a diverse sample representative of the epicenter of the HIV epidemic (i.e., the Deep South), several limitations should be noted. Although our sample was reflective of the demographic most burdened by the modern HIV/AIDS epidemic (i.e., African–American men in the Deep South), the findings may not nevertheless generalize to other populations of PLWH with regard to factors influencing treatment management. There were missing data for several of the HIV biomarker variables that reduced analyzable sample size and thus reduced power in our multivariable models to detect an interaction between low health literacy and neurocognitive functioning on these outcomes. Our cross-sectional design limits causal inference on the protective or deleterious role of health literacy on HIV outcomes over time, as well as in the temporal (and likely bidirectional) associations between health literacy and neurocognitive functioning, as well as HIV disease management and neurocognitive functioning. Urine toxicology data were available from the day of the study visit, but we did not administer a structured clinical interview to gather substance use diagnoses, which limited our ability to understand the influence substance use disorders may have on the study findings. Another limitation was our selection of HIV treatment management outcome variables. Although we did include both subjective (i.e., treatment self-efficacy, VAS medication adherence) and performance-based measures (i.e., health-related decision-making), examining other objective treatment outcomes (e.g., Medication Event Monitoring System [MEMS] caps, pharmacy refill data) would have enhanced the validity of our findings.

Conflict of interest

None declared.

Funding

This study was supported by National Institutes of Health (NIH) grants K99/R00-AG048762, R01-MH106366, P30-AG022838, and R24-AI067039.

References