Factors associated with patterns of plural healthcare utilization among patients taking antiretroviral therapy in rural and urban South Africa: a cross-sectional study

As part of detailed case studies of healthcare access and equity in four sub-districts of South Africa, we conducted cross-sectional studies among ART users (See Table 1). We purposefully selected four sub-districts, two rural and two urban, located in four provinces following consensus between research partners and policy-makers. While such a strategy cannot generate a representative sample, the settings reflect some of the diversity of context typical of South Africa. Table 1 outlines the study sites in more detail, demonstrating that both urban sites and rural sites varied in population size. The two rural sites are also existing demographic surveillance sites, known for high levels of circular labor migration. The urban site in Western Cape and the rural site in Kwa-Zulu Natal are considered to have decentralized ART programs, whereas the other two have highly-centralized ART programs. Since ART is primarily delivered through accredited national public sector roll-out facilities, only users of these programmes were recruited in each of the four settings. A total of 13 ART facilities were selected for this study. In each sub-district, all accredited facilities were included where possible, and where multiple facilities existed, self-weighting stratified or probability proportional to size methods were used to select facilities using data on the total number of users in each facility. A total sample of 1266 users of ART was obtained across the four sites using a systematic random sampling method, ranging from 300 to 331 in each sub-district. Only adult ART patients, 18 years and older, with a minimum of two weeks since ART initiation were included in this study, and their ART clinic records were reviewed for biological markers. The sampling procedure aimed to ensure a representative sample in each of the four sites. Following selection of facilities, ART clinic appointment and attendance registers were used to estimate the total number of patients scheduled to attend the ART clinic and to select a random position in the queue of patients on the day of data collection. The fixed interval (nth position) in the queue was chosen depending on the size of the facility and the total numbers expected on the day of data collection. The selection process was also spread out through the entire day in order to capture patients presenting at different times of the day. In each site, once data collection commenced, it was continued on every ART clinic day until the desired sample had been achieved. The data collection occurred between April 2008 and March 2009.

Structured questionnaires were used to collect data during exit interviews (see Additional file 1). Data were collected on socio-demographic factors, availability and affordability of ART services, psychosocial support for ART, acceptability and quality of ART care, utilization and clinical outcomes, and household assets were used to estimate socioeconomic status (SES). Participants were interviewed on the day of their clinic visit to the ART service point. Trained research assistants conducted face-to-face exit interviews in XiTsonga, Afrikaans, IsiXhosa, IsiZulu, Sepedi or English depending on the participant’s preference. Exit interviews were linked by unique identifiers to an HIV clinic record data extraction form that obtained additional clinical data such as CD4 count and viral load levels on the same day as the interview.

Participants were asked two key questions in relation to additional health care sought in the month prior to their clinic visit. The first question was: “Apart from visits to this clinic for your ARVs, have you used this clinic or any other health service in the last four weeks? If so, how many visits did you have, and how much did you have to pay the provider for each?” Participants were prompted with a series of choices, including the ART clinic itself (not for ARVs), public primary healthcare (PHC) clinic or hospital, a private chemist, doctor or hospital, and Tuberculosis (TB) or antenatal (ANC) clinics. The second question was “Have you spent any other money on health care in the past month (e.g. traditional medicines, spaza shops, special food, etc.)? If so, how much have you spent?” The first question relates to the concurrent use of additional providers, and represents plurality of health service and provider usage, and the second question to purchase of health care products and substances, representing a form of self-care behavior.

Data were double-entered into EpiData v3.1 and imported into STATA v10 (Statacorp, College Station, Texas) for analysis. Descriptive statistics of plural healthcare patterns and constructs, stratified by urban and rural residence, were generated and bivariate analysis conducted with the chi-squared test and logistic regression. Multiple logistic regression analyses were used to identify predictors of utilization patterns, and we present adjusted odds ratios and 95% confidence intervals. The factors assessed included a socio-economic index, access barriers (including perceptions of quality) and biological outcomes such as baseline and follow-up CD4 counts and viral load suppression. An index for SES was constructed using multiple correspondence analysis, described elsewhere [20]. Catastrophic healthcare expenditure was computed as mean monthly healthcare-related expenditure of more than 15% of mean monthly household expenditure. ART-related knowledge was generated using a combination of three questions: knowledge of recent CD4 count, and correct responses to two questions, do you stop ART when weight improves, and does ART cure HIV/AIDS.

Provider pluralism represents use of any additional provider, irrespective of additional self-help practices. Self-care pluralism refers to purchase of any product or substance for health care purposes, and excluding additional provider utilization. A third dependent variable was generated by a combination of plural provider and self-care variables, referred to as plural healthcare utilization. Plural healthcare is used in particular way in this study, whereby use of physicians and private chemists is measured alongside traditional healers as a reflection all healthcare service modalities available in this context. Multivariate logistic models were fitted for provider pluralism, self-care pluralism and the combined variable of plural healthcare utilization. Participants who sought additional care from their regular ART clinic were excluded from provider pluralism. For purposes of analysis, TB and ANC users were included in the initial description of plural healthcare patterns, but excluded in the models for determinants and implications, as there are clearly defined indications for use of TB and ANC services.

Ethical approval was obtained from Universities of Witwatersrand, Kwa-Zulu Natal, Cape Town and the respective Provincial Health Research Committees. Written informed consent was obtained separately for the interview, and for the HIV clinic record review.

As shown in Table 2, participants were distributed approximately equally between urban (52%) and rural (48%) sites, and were predominantly Black African (98%). In keeping with the gender distribution of most ART clinics, 74% of participants were female. Most participants were aged below 40 years (61%), never married (55%) and unemployed (78%). While the majority of participants did not possess medical aid insurance (98%), 42% were enrolled on the temporary disability grant intended for financing of healthcare needs. The SES measure allowed for equally-sized groupings across all quintiles (20%), indicating lack of clumping or truncation [21]. At initiation of ART, 78% of participants had a baseline CD4 count above 50 cells/ul.

Just under one-fifth (18.9%) of participants visited additional healthcare providers over and above their regular ART clinical care. As shown in Figure 1, provider pluralists visited chemists, doctors and hospitals in the private sector, primary healthcare (PHC) clinics and hospitals in the public sector, and traditional healers in the indigenous sector. A further 14.9% of participants purchased products or substances for purposes of healthcare, which we refer to as self-care pluralism. The two variants of plural healthcare pluralism, provider and self-care, occurred in a combined 33.8% of all respondents.

The types of provider used varied between urban and rural sites (Figure 1). Urban residents made greater use of private chemists (30% vs. 19%), while rural areas were more likely to use traditional healers (7% vs. 1%). Usage of private doctors was slightly higher among rural residents (30% vs. 23%). A slightly greater proportion of urban residents used public hospitals (13% vs. 7%), whereas a greater proportion of rural residents used PHC facilities (20% vs. 13%) relative to urban residents. Traditional healers were used more by rural residents than their urban counterparts (7% vs. 1%). Within a period of one month, an average single visit to an additional provider was reported by the participants. Mean and range of visits were identified as follows among plural provider users: private chemists (01, 01–03), private doctors (01, 01–06), private hospitals (01, 01–01), PHC clinics (01, 01–03), public hospitals (02, 01–10) and traditional healers (01, 01–03).

Participants incurred expenses related to both provider and self-care pluralism, measured by reported direct healthcare costs only, as shown in Figure 2. Although utilization of traditional healers (N = 11) was uncommon, the highest amount of expenses reported (USD125) was paid to these providers. Private chemists (Inter-quartile range [IQR] USD3-USD13, N = 86) and doctors (IQR USD19-USD30, N = 88) were also relatively costly for participants, with median costs of USD6 and USD19 respectively. The costs of self-care practices (N = 154) were also high, with a median of USD9 (IQR USD3-USD25). However, usage of PHC clinics (IQR USD1-USD3, N = 53) was less costly with a maximum amount of USD7 spent.

Despite the likelihood that some of the indirect and direct costs may not have been fully captured, and for a largely unemployed source population, the patients surveyed incurred large costs to finance their healthcare. Nearly one-fifth (18.7%) of the total sample reported having to borrow money (18.7%) and at times sell their belongings (4.9%) in order to finance healthcare. Borrowing was more common among plural care users (P < 0.001), and rural residents were 6.8 times more likely to borrow than urban residents (Adjusted Odds Ratio [aOR] 6.75, 95% Confidence Interval [95%CI]: 4.73–9.64). Amongst those borrowing money, plural care users (63.0%) were more likely to report difficulty incurring the costs of healthcare use than non-plural users (37.0%) (P = 0.006).

When the degree of catastrophic healthcare expenditure was measured, 34.8% of plural healthcare users showed catastrophic levels as opposed to 6.9% among patients not using additional healthcare (P < 0.001). Catastrophic level of healthcare expenditure was identified in 77.2% of plural healthcare users with rural origin compared to 22.8% of urban residents (P < 0.001). Using the socioeconomic distribution of participants by quintiles ranging from poorest to richest shown in Table 3, urban plural care users were found to be of a higher SES in comparison to their rural counterparts (P < 0.001).

The use of plural healthcare was higher in rural participants (aOR 1.97, 95%CI: 1.12–3.44). As shown in Table 4, higher rates of pluralism occurred among those who possessed medical aid insurance (aOR 6.15, 95%CI: 2.31–16.33). Having a disability grant was associated with 35% increased odds of plural healthcare. Although SES was not statistically significant when adjusted for other factors, creating debts by borrowing money (aOR 2.68, 95%CI: 1.87–3.84) to finance healthcare were strongly associated with plural healthcare usage. Borrowing money was also associated with catastrophic household expenditure (aOR 3.26, 95%CI: 2.25–4.73).

The presence of social support in the form of treatment buddies increased the odds of plural healthcare by 44%. Finally, several variables related to the quality of care experienced by participants in the ART clinic showed association with use of plural healthcare. Perceived lack of privacy in the ART consultation rooms (aOR 1.50, 95%CI: 1.08–2.08) and disrespect by ART providers (aOR 2.07, 95%CI: 1.54–2.79) was associated with plural healthcare use, whereas participants who encountered language barriers with ART providers had 43% reduced usage of plural healthcare as opposed to those who did not.

We constructed separate regression models for the two variants of plural healthcare utilization, namely, provider pluralism and self-care pluralism, and the results are presented below as well as in Table 5.

Lower SES was a predictor of reduced provider pluralism (aOR 0.67, 95%CI: 0.45–0.98). Ability to finance health care, such as seen among those who used medical aid insurance (aOR 7.59, 95%CI: 3.06–18.78), was predictive of provider pluralism. However, debt creation was higher among provider pluralists by 90% (Table 5).

xParticipants who on any previous occasion had to leave the ART clinic without receiving help were 2.7 times more likely to use additional providers. In addition, perceptions of ART provider disrespect showed 43% increased odds of provider pluralism. However, experiences of language barriers (aOR 0.55, 95%CI: 0.32–0.96) were associated with reduction in provider pluralism. Therefore, factors representing poor quality of care were predictive of provider pluralism.

Rural-dwelling (aOR 5.91, 95%CI: 1.90–18.40) and temporary disability grant enrolment (aOR 1.93, 95%CI: 1.33–2.80) were found to be associated with self-care pluralism. The self-care category was 2.4 times more likely to borrow money in order to finance healthcare. Those engaging in self-care were 58% more likely to have a treatment supporter, a form of social support, although the level of statistical significance was marginal (Table 5).

Self-care utilization was associated with 83% and 92% increased odds in lack of privacy during consultation and perceived disrespect for patients by clinic staff, respectively. Participants who preferred to be seen by a nurse provider (aOR 0.52, 95%CI: 0.31–0.89) on their regular clinic visit were less likely to engage in self-care pluralism, as opposed to those who preferred a doctor. Surprisingly, participants who perceived facilities to be dirty were also less likely to engage in self-care pluralism (aOR 0.57, 95%CI: 0.34–0.96). However, dirty facilities predicted plural provider utilization in bivariate logistic regression (OR 1.53, 95%CI: 1.10–2.14).

CD4 count level at ART initiation was not associated with plural healthcare (aOR 0.93, 95%CI: 0.70–1.25), adjusted for age, sex, education, SES, closeness to ART facility and ART knowledge. Among participants with recorded recent CD4 count results (59%), improvement in CD4 results was not associated with plural healthcare utilization (aOR 1.06, 95%CI: 0.72–1.57). When viral loads were retrievable (66%), viral suppression (70%) was not associated with plural use of healthcare (aOR 0.97, 95%CI: 0.83–1.58). However, in bivariate but not multivariate logistic regression models, self-care pluralism was associated with poor levels of ART-related knowledge (OR 1.77, 95%CI: 1.30–2.43) and missed doses of ART (OR 2.40, 95%CI: 1.33–4.32), but not with missed ART clinic visits (OR 0.43, 95%CI: 0.17–1.10). Some patients (8.0%) had previously received ART from facilities other than their regular clinic, a practice associated with provider pluralism (OR 1.76, 95%CI: 1.01–3.09).