Education and HIV treatment adherence

There are numerous reasons to believe that education could have a positive influence on chronic HIV treatment in sub-Saharan Africa. First, education may increase an adult’s employment and wealth, in turn providing individuals with material resources that can be used to overcome logistical barriers to chronic care services [23]. For example, for people to effectively adhere to their HIV treatment they must return to the health clinic on a monthly, quarterly, or semi-annual basis to obtain their medication and be monitored for medication effectiveness. These routine clinic visits necessitate money for transportation and the ability to negotiate and compensate for time off of work [24,25]. Expanded access to HIV treatment in the region means that these routine visits can now be carried out at local clinics. However, even local clinic visits can involve travel of an hour or more and long wait times [26]. To the extent that more education provides greater access to financial resources, transportation, and/or more flexible work arrangements, more educated PLWH may find these financial and time costs easier to manage.

Second, education can facilitate the development of key skills (e.g., numeracy and literacy), motivation, and self-efficacy that can positively influence an individual’s treatment adherence [27–29]. First, those who can read and write may be more effective in maintaining their HIV care and treatment. Appointment dates are often written down in a client’s personal health record (i.e., health passport)—necessitating literacy skills to remember appointments [30,31]. Other tools for remembering appointments, like calendars and cell phones, require basic literacy skills. Second, education may be protective, as more educated HIV clients may be more inclined to advocate for themselves due to greater self-efficacy and feel more empowered to return to care rapidly after a missed appointment [32]. Further, those with more education may more accurately assess the risk of treatment non-adherence, may have greater exposure to information about HIV treatment via different social circles, and have greater motivation to maintain HIV care and treatment, even in the face of other obstacles [33].

Although conventional wisdom and prior literature generally suggest a positive relationship between education and treatment adherence, there are also several reasons to expect the relationship in the sub-Saharan African context may be attenuated. First, limited employment opportunities in many areas mean that individuals across the education spectrum may only have access to agricultural or ad-hoc/informal work, often with low pay. [34]. Second, despite remarkable gains in access to primary schooling in the region, the quality of education has not seen commensurate gains, often resulting in limited literacy and numeracy skills gained from attending school [35]. Thus, even with more education, individuals may not possess the material or cognitive resources needed to support their HIV treatment adherence. Third, because HIV treatment has become increasingly simple and widely accessible—it is now free and requires just one pill a day and may be even less frequent via injectables in the future. Thus, it is plausible that education could play a less important role in treatment adherence than would be expected for more expensive or complex chronic care regimens, or where treatment access is more limited [36]. Fourth, many other socio-cultural factors may operate independent of education to facilitation HIV treatment adherence. For example, many people in the region now know about and highly value HIV treatment, leading to increased social support for treatment adherence [37].

Gender differences

Gender differences in education, health care access, and power dynamics suggest that the association between education and HIV treatment adherence may be distinct for men and women. In this study, I move beyond well-known gender disparities in both education and HIV care [14,38], to examine how educational attainment may benefit men’s and women’s chronic HIV treatment adherence in different ways.

Men’s greater access to employment and income suggests that the benefits of education for HIV treatment adherence via employment may be more pronounced for men. Even when women obtain higher levels of education, they are less likely to work outside of the home and earn less money when compared to similarly educated men [39]. Thus, it is not clear that having more education confers a substantially greater material advantage to women—for example, for managing frequent trips to the HIV clinic. In contrast, more educated men may have comparatively greater access to employment and material resources to facilitate their routine HIV care than less educated men [34].

Men and women may exhibit different associations between education and HIV treatment adherence as a consequence of their different levels of comfort with using health services in general and historically different experiences with accessing HIV care. Health care systems in sub-Saharan Africa have largely centered on women of reproductive age, with a focus on antenatal care services and under-five vaccinations and check-ups [40]. Women were the first recipients of universal HIV treatment for life, at least five years before universal treatment was expanded for men [40]. As such, women across the education spectrum are likely to have more experience and comfort with managing their HIV treatment. Whereas men with more education may have had greater opportunities to seek HIV treatment in private clinics prior to the expansion of universal HIV treatment, and may generally have greater self-efficacy and experience seeking health services than their less educated counterparts. These gendered dynamics pertaining to financial resources and comfort attending health clinics suggest that positive education-HIV treatment adherence associations will be stronger for men and weaker for women.

On the other hand, HIV treatment adherence is not only a matter of clinic access—it must also be managed inside the home, where gender dynamics are particularly entrenched. More educated women may be better equipped to negotiate gendered power dynamics and stigma in the home, which are frequent barriers to women’s treatment adherence and HIV clinic attendance [41]. For example, women in the region report hiding their ART from other household members [42]. More educated women likely face less precarity due to HIV stigma for a variety of reasons, including greater financial resources outside of marriage, wider social networks, and their individual skill sets [43]. This suggests the alternative possibility that there could be a strong positive association between education and HIV treatment adherence for both men and women.

Age differences

Across the life spectrum, it is plausible that education could be a resource to help individuals navigate a variety of life obstacles that impede HIV care and treatment. There are well documented differences in HIV outcomes by age in the region [15], with worse outcomes among younger adults, suggesting that older and younger individuals may experience different barriers to HIV care [44]. Life demands such as work and child rearing are common in younger and middle-aged adult populations, which likely create competing priorities with the demands of on-going HIV care and treatment [24]. Older individuals may have other complications such as greater mobility issues and more chronic health concerns that complicate their HIV treatment adherence [44]. While each set of challenges are unique, it is likely that education could be an important resource for overcoming the HIV treatment barriers faced by both older and younger populations—suggesting positive education-HIV treatment adherence associations for younger and older adults alike.

On the other hand, a cohort perspective suggests that people of different ages likely obtained different sets of resources from their schooling, influencing the extent to which their education supports HIV treatment adherence. Given extensive educational reforms in the region over the past 40 years—increases in primary education enrollment coinciding with declines in learning outcomes—those in older age groups enjoyed less access to education, but those who did access education may have had better quality schooling [45–47]. Indeed, more recent cohorts in sub-Saharan Africa show dramatically lower literacy levels than older cohorts [46]. Thus, being more educated in an older age group is potentially more differentiating than it might be within younger age groups, for whom attending school was more normative but also lower quality [48]. As a consequence, older adults with more education may have gained more income-earning opportunities from their schooling credentials and more skills for managing their HIV treatment adherence [49,50]. This suggests that in older age groups, those with more education may be better able to adhere to their HIV treatment, but that education may be less differentiating for younger age groups.

The aim of this study is to examine whether education is associated with viral load suppression (as an indicator for HIV treatment adherence) in southern and eastern Africa, and whether this association varies by gender and age. Because the theoretical and empirical literature suggests competing hypotheses for the existence or direction of these associations, I do not offer specific hypotheses. I will examine and discuss how my results contrast or support prior theoretical suppositions and empirical information.

Ethics statement

To conduct this study I used secondary, publicly available Population HIV Impact Assessment (PHIA) data, and I carefully reviewed and followed all documentation guidelines for appropriately managing and analyzing PHIA data. All PHIA survey protocols, consent forms, screening forms, refusal forms, referral forms, recruitment materials and questionnaires were reviewed and approved by in-country ethics and regulatory bodies and the institutional review boards of Columbia University Medical Center, Westat, and the U.S. Centers for Disease Control and Prevention. All adults provided written or verbal informed consent that was recorded on electronic tablets before beginning the survey. All respondents were informed of the voluntary and confidential nature of the survey and separate consent was obtained for laboratory samples. All publicly available data is free of identifiers and numerous measures are in place to protect respondent confidentiality. PHIA ethical protocols, including procedures to ensure privacy and confidentiality, are outlined in detail in country final reports that can be found here: https://phia.icap.columbia.edu/resources/. The data for all seven countries included in this study were accessed by the author upon approval from the PHIA study team, and were downloaded in August of 2022. All data is publicly available upon request from the PHIA study at https://phia-data.icap.columbia.edu/. All statistical codes are available upon request from the author.

Data

To answer the question of whether education is associated with viral load suppression, I draw on the first round of data collection (cross-sectional) from the Population HIV Impact Assessment (PHIA) study [51]. All PHIA data is publicly available, with permission from the PHIA study team, via this website: https://phia-data.icap.columbia.edu/datasets. I focus on seven countries in southern and eastern Africa for which data collection began between 2015 and 2017, was completed by July 2018, and for which the data was made publicly available by June 2021: Lesotho, Malawi, Namibia, Tanzania, Uganda, Zambia, and Zimbabwe. These countries were chosen based on their location within the southern and eastern African region, similar nature and history of the HIV epidemic, their similar education and HIV policy histories, and the similar timing of data collection and availability [4–8]. It should be noted that there are important differences between and within each country’s context, leading me to include both separate and pooled approaches in this study.

Study sample

My final analytic sample included 12,198 PLWH with complete data who had initiated HIV treatment by the time of the survey, as documented either via self-report or by evidence of ART in survey blood tests.

The PHIA samples are designed to be population-representative for men and women, ages 15–64, following a two-step process [51]. First, enumeration areas were selected with a probability proportional to size procedure—wherein the probability of an enumeration area being sampled is proportional to the size of the area. Second, households within selected enumeration areas were chosen at random. All eligible individuals who lived in or slept in each selected house the prior night were eligible to participate. Household data include information on household resources/wealth and number of household members; individual data include education, age, and information to construct dependent variables, including self-reported engagement with HIV services. Bio-marker data include laboratory test results for HIV diagnosis, viral load, immune system health (e.g., CD4 count), the presence of ART in the body, and ART resistance. Additional information on the PHIA study design can be found here: https://phia-data.icap.columbia.edu/.

Among the seven countries included in my analysis, there were 17,187 PLWH before limiting the analysis to those who had initiated ART. Among the full sample of PLWH, I excluded 778 respondents from my analyses who had incomplete data on self-reported variables (apart from education) and bio-marker values that were included in my multivariable tests (<1% of my N) or outliers who reported >26 years of education. For those 5% of cases who were only missing data on the key independent variable of education, I imputed the years of education using a single imputation nearest neighbor approach [52]—wherein the missing education for an observation was replaced with the education of the ‘nearest neighbor’. Nearest neighbors were statistically matched based on the greatest similarity of characteristics within each country (using the control variables in the multivariate model 2, listed below).

Measures

I obtained information on socio-demographic information, including years of education, from linked household and individual datasets. I then linked this information with bio-marker datasets, which contained information on viral load and the presence of ART in the body.

For the dependent variable, I used the internationally agreed upon limit of a viral load of < 1000 copies/mL [53] to create a dichotomous measure of viral suppression, conditional on having started ART. Viral loads were obtained via blood tests collected as a routine component of the PHIA study among willing participants. 1000 copies/mL is the minimum consistently detectable viral load count available via most viral load lab tests in sub-Saharan Africa, and there is very low risk of disease progression or onward transmission below this level [54]. Thus, this is the most clinically relevant cut-off used in health care practice and research throughout the region.

For the main independent variable of interest, I use a continuous measure of years of education, which I constructed by combining two measures of self-reported education—‘highest level of school you attended’ (e.g., primary, secondary) and ‘highest grade you completed at that level’ (i.e., years of schooling completed at that level)—a common practice for population-level surveys from the region [55]. For example, someone who reported completing four grades at the secondary level would have a total of 12 years of education assuming primary schooling ends with the completion of grade 8. Because the number of grades in each level and the number of levels of education varied between countries, I used country-specific calculations.

My use of a continuous measure of years of education contrasts with other commonly used ordinal or categorical measures of primary, secondary, and tertiary level attendance and/or completion. A continuous measure provides an opportunity to assess the value of education independent of the credentialing offered by completing a given level of education [56], while preserving the variation within a given level of schooling. Further, using a continuous measure of education allowed me to explore any potential curvilinearity in observed relationships, and simplifies comparisons between countries with different education level structures. For consistency in comparing my findings with other studies, I provide sensitivity analyses using the different categorical measures of education levels (i.e., none, some primary, completed primary, some secondary, completed secondary, and tertiary) (see S1 Table in Supporting Information).

To adjust for key socio-demographic characteristics, I included individual-level covariates in my multivariable regression models. I measured age in years using a continuous measure centered at the minimum age of 15 (e.g., in the model age 0 = age 15), which facilitates the interpretation of the coefficients. I use a self-reported variable for gender as a binary measure (men vs. women). I controlled for enrollment in school among the adult survey sample at the time of the survey, using a binary measure of currently in vs. out of school, to control for the possibility that some respondents would not yet have finished their education. I used an integer measure to control for the number of children living in the home, which may suggest differential care-taking and resource burdens within a household. To account for the influence of being in a partnership (e.g., via social support and/or fear and stigma), I adjusted for a mutually exclusive, categorical measure of marriage: never married, currently in a married/partnered (married/partnered/co-habitating), and previously married/partnered (widowed, divorced, separated). To account for other measures of socio-economic status, I used a binary measure of employed vs. not employed in the past year (i.e., any work for cash or in-kind payment), and a five-level index measure of household wealth—from poorest to richest—created by the PHIA study using a principle components analysis and relative quintile breakdown [51]. I accounted for geographic differences by controlling for a binary measure of urban vs. rural. Finally, I adjusted for key socio-political and socio-economic differences between countries (e.g., GDP, literacy level, economic development, etc.) —which can influence education and health care access and outcomes—by including country-level fixed effects as dummy variables for each of the seven countries in my models with all countries pooled together.

Analysis

I conducted all analyses using Stata 18 [57]. Analyses include descriptive statistics along with bivariate and multivariable logistic regression models. I applied the PHIA individual-level base weights for blood test participation (the level of measurement with the lowest level of participation). All weights were calculated by PHIA to account for individual selection in the PHIA sampling process, which includes adjustments for non-response, the multi-stage stratification process, and under-coverage by age and gender [51]. Following guidance from PHIA, I applied weights using jack-knife replication, including appropriately weighting data when pooling across multiple countries [51].

I developed two multivariable logistic models. Model 1 included all covariates listed above, with the exclusion of the potentially mediating socio-economic variables of employment and wealth. Subsequently, in Model 2, I tested the unique influence of education on viral load suppression, net of wealth and employment, including all other covariates from Model 1. This allowed me to assess the association between education and viral load suppression, with (Model 2) and without (Model 1) controlling for the influence of employment and wealth—providing insight about the value of education beyond material resources— while still controlling for other covariates.

I ran the final Models 1 and 2 individually for each country and for the pooled sample with all countries combined (with country-level fixed effects). Country-specific findings for Model 2 are provided in S2 Table in Supporting Information.

For Model 3, I tested whether the relationship between education and viral load suppression is moderated by gender by including an interaction term between the binary measure of gender and education. Country-specific findings for Model 3 are provided in S3 Table in Supporting Information. For Model 4, I tested whether age moderated the relationship between education and viral load suppression by including interactions between the continuous measure of age and years of education. Country-specific findings are provided in S4 Table in Supporting Information. Finally, I adapted Model 2 using a different categorical education measure—entry into or completion of schooling levels (i.e., primary, secondary, tertiary)— that align with some prior research in sub-Saharan Africa (see Table 1 in Supporting Information).

Education and viral load suppression associations

In Table 3, I provide the estimates from multivariable models testing the associations between education and viral load suppression in the pooled sample. Model 1 does not control for employment or wealth, whereas Model 2 does adjust for these factors. In both models, for each additional year of education there was no difference in the likelihood of achieving viral load suppression—with statistically and practically small effect sizes and narrow 95% confidence intervals (CI) that crossed 1.00 (Model 1: AOR = 1.01, 95% CI = 0.99,1.03; Model 2: AOR = 1.00, 95% CI = 0.98, 1.03).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Associations between years of education and viral load suppression (VLS) among people living with HIV who initiated antiretroviral therapy, adjusted odds ratios (AOR) with 95% confidence intervals (CI).

https://doi.org/10.1371/journal.pone.0348947.t003

Women were approximately 60% more likely to be virally suppressed comparted men, and for each additional year of age, there was a 3% greater odds of being virally suppressed in both models, controlling for other factors. In Model 2, neither employment nor wealth were associated with viral load suppression when adjusting for other variables.

As shown in Fig 1, the lack of association between education and viral load suppression persisted across all countries. The lack of statistical association between education and viral load suppression was robust across sensitivity analyses—I observed similar findings when I used an ordinal measure of education level (e.g., primary, secondary, tertiary), a binary measure of any vs. no education (instead of years of education), and when I included all PLWH regardless of their ART initiation.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Adjusted odds ratios of viral load suppression for each additional year of individual education among people living with HIV who initiated antiretroviral therapy, with 95% confidence intervals (fromTable 3, Model 2).

https://doi.org/10.1371/journal.pone.0348947.g001

Gender moderation results

Table 4, Model 3 demonstrates that the interaction between gender and education was statistically insignificant at the 95% confidence level, suggesting that there was no moderation in education-viral load suppression associations by gender. Fig 2 more clearly demonstrates the lack of any education-viral load suppression association for both men and women--showing the probability of viral load suppression across all years of education in the pooled model, with two separate lines for men and women. Across the education spectrum, the probability of viral load suppression for men was between 80% and 85%, whereas for women it was between 85% and 90%. In my country-specific sensitivity analyses, gender did not modify educational associations with viral load suppression in any country.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Gender and age moderation of associations between years of education and viral load suppression (viral load suppression) among people living with HIV who initiated antiretroviral therapy, adjusted odds ratios (AOR) with 95% confidence intervals (CI).

https://doi.org/10.1371/journal.pone.0348947.t004

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Probability of viral load suppression for each year of individual education with 95% confidence intervals, for men and women living with HIV who initiated antiretroviral therapy (fromTable 4, Model 3).

https://doi.org/10.1371/journal.pone.0348947.g002

Age moderation results

In Model 4, the interaction term between education and age was statistically significant at the 95% confidence level, suggesting that age moderates the relationship between education and viral load suppression (see Table 4). For each additional year of education there was a 5% decrease in the odds of viral load suppression when controlling for other factors, but the odds of viral load suppression in relation to education increase by only.2% for each 1-year increase in age—a small effect size. These relationships are more clearly visualized in Fig 3, which displays the probability of viral load suppression across all years of education, with each line representing a different age group, from 25–60 (or 10–45 using the age centered variable) in the pooled model. Among the 60-year-old age group, there was a positive association; among middle aged groups, there was a statistically significant (at the 95% confidence level) positive association at age 45—reflecting a difference between those with no education vs. those who had 12 or more years of schooling—and there was no association at age 35; and among young adults (age 25) there was a statistically significant negative association (at the 95% confidence level), which reflects a lower probability of viral load suppression among those with no schooling vs. those with 12 or more years of schooling. In sensitivity analyses, these patterns were consistent for both men and women in the pooled sample. However, the education-age interaction term was statistically not significant at the 95% confidence level in all country-specific models, suggesting that age moderation is only detectable in the larger, pooled sample size.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Probability of viral load suppression for each year of individual education among people living with HIV who initiated antiretroviral therapy with 95% confidence intervals, for ages 25, 35, 45, and 60 (fromTable 4, Model 3).

https://doi.org/10.1371/journal.pone.0348947.g003

Strengths and limitations

The PHIA is the first study to consistently collect viral load and education measures at the population level across multiple countries in sub-Saharan Africa. Historically, viral load measures have been obtained through clinical data, which rarely includes education measures and does not include the population disengaged from HIV care. Using this novel data, I am able to examine education-viral load suppression associations in a consistent manner at the population level across countries. Further, in contrast to many studies that categorize education by level of completion, I measure education continuously allowing me to examine the association within primary and secondary schooling, where the majority of the variance occurs in these contexts.

These findings should also be considered in light of several limitations. First, my use of a cross-sectional viral load measures fails to capture perfect adherence on a day-to-day basis—it is possible to have good, but imperfect adherence, and to still demonstrate viral load suppression on a blood test. Additionally, I am unable to measure sustained viral load suppression over time, limiting my ability to speak to the influence of education on successful, on-going maintenance in HIV care. Nevertheless, viral load suppression remains a critical indicator for individual and population health, and provides a clear marker of effective, if imperfect, treatment adherence at the population level. As my measure of education is based on self-report, rather than school records, responses may be influenced by recall or social desirability biases. Finally, it is possible that these findings are downwardly biased due to other unobserved factors or selection issues—for example, educational differences in HIV infection rates, entry into HIV care, or HIV-related mortality prior to the survey. Though, in separate analyses with this same population, I did not identify any relationship between educational attainment and HIV diagnosis or treatment initiation, which suggests that this form of selection bias will be less relevant for the present study [65].