Physical and emotional health outcomes after 12 months of public-sector antiretroviral treatment in the Free State Province of South Africa: a longitudinal study using structural equation modelling

This study is part of an ongoing cohort study of patients enrolled in the public-sector ART programme in the Free State Province of South Africa. The research was approved by the Ethics Committee of the Faculty of Humanities of the University of the Free State.

The sampling frame consisted of a list of names obtained from the Provincial Department of Health of adult patients certified as medically ready to commence ART (CD4 < 200 cells/μL and/or WHO stage IV) within two months of the first patient having received his/her treatment. The list distinguished those patients who were receiving treatment at baseline ('treatment' patients) and those who were certified as ready to commence treatment but were not yet receiving it ('non-treatment' patients). Eighty patients were sampled randomly from this list for each of the five districts of the province (i.e., Lejweleputswa, Motheo, Thabo Mofutsanyana, Fezile Dabi, and Xhariep), proportional to the number of patients per clinic and per treatment group (treatment and non-treatment patients). Because there were less than 80 eligible patients in Xhariep, a census of all treatment and non-treatment patients was conducted.

A total of 371 study participants were recruited into the study (268 treated and 103 untreated patients). Only patients who were undergoing treatment at baseline (n = 268) were included in this study, because it assesses the impact of ART on the QoL of patients. This study used two waves of panel data to examine the impact of ART on QoL in the medium term. At baseline (< 6 months ART), trained enumerators conducted face-to-face interviews with 268 treatment patients using a standard questionnaire, after the written consent of all the patients had been obtained. Approximately six months later (< 12 months ART), 234 of the original cohort of patients were re-interviewed using an updated version of the questionnaire.

Figure 1 shows the conceptual model to be tested. The only exogenous variable is the length of treatment time. We measured the adverse effects of ART using a construct that categorized patients as having no adverse effects, mild adverse effects, or disruptive adverse effects. Incorporating the severity of the adverse effects is important because many adverse effects are related to mild toxicity, which can be treated easily. A wide variety of good measures of quality exist for use in HIV/AIDS populations when physical criteria are examined [29, 36]. To assess physical QoL in our study, a latent variable was constructed using EuroQoL 5D – a standardized, widely validated and tested instrument developed by the EuroQoL Group to assess five generic aspects of current health (mobility, self-care, limitation of activities, pain, and mood). This instrument has shown good construct validity in AIDS trials in resource-poor settings [6, 37‐41]. Respondents could indicate one of three levels of impairment in each health domain: none, some/moderate, or extreme. The five items (and the entire questionnaire) were translated into Afrikaans, Xhosa, Zulu, and Sotho. The Afrikaans, Xhosa, and Zulu versions of the EQ-5D are recognised as official translations by the EuroQoL Group and have shown equivalent reliability and validity [30, 40, 42]. The Sotho instrument was translated for the purpose of this study and further research is required to ensure its measurement validity. The EQ-5D is a generic instrument developed to describe and evaluate overall health-related QoL. However, in this study, a distinction was drawn between the physical and emotional outcomes of ART. The "mood" item in the EQ-5D assesses an aspect of emotional health and was not included as an indicator of physical QoL. Therefore, a latent variable measuring the physical QoL of patients was constructed using only the first four dimensions of the EQ-5D.

Measures of overall emotional or psychological well-being are generally based on self-reports. This is called subjective well-being (SWB) [43‐45]. SWB is an important indicator of QoL because it describes a person's multi-dimensional evaluation of his/her life, including affective evaluations of moods and emotions, and cognitive judgements of life satisfaction [45‐49]. There is currently a consensus that SWB consists of three primary components: life satisfaction, pleasant affect, and low levels of unpleasant affect [44, 47, 48, 50, 51]. SWB has previously been used successfully to evaluate people's well-being in South Africa [30, 43, 49]. Patient responses to a standard five-point scaled item regarding overall life satisfaction were used to measure the cognitive part of SWB. An item asking about global happiness was used as a measure of positive emotions. The fifth EuroQoL item, measuring anxiety and depression, was used to measure the third component of SWB, the lack of unpleasant affect. This instrument has shown good reliability and construct validity in earlier studies of QoL in ART patients in South Africa [30].

Structural equation modelling (SEM) was used to assess firstly, the influence of ART on physical and emotional QoL, and secondly, the impact of physical QoL on emotional QoL. This model also included the impact of adverse treatment effects on the patients' physical and emotional QoL. SEM is a combination of factor analysis and path analysis, and is particularly well suited to addressing longitudinal analytical problems. Unlike multiple regression models, SEM permits the simultaneous assessment of multiple dependent variables in a single model. SEM also allows the examination of both the direct and indirect effects of one variable on another. Furthermore, with SEM, a given construct can be treated as both an independent and a dependent variable, providing a model that more closely parallels the complex nature of most social and clinical phenomena (path analysis). Finally, SEM allows the examination of the relationships among both measured and latent variables (factor analysis), which often provides a more precise measure of complex social phenomena than do single measurements [52]. All data analysis was performed with the statistical software package LISREL Version 8.72.

The mean age of this sample of people living with HIV/AIDS was 37.9 years (SD = 8.6). Of the 268 ART patients included in this study, 66.8% were women. Overall, the patients described their physical QoL as good or excellent, with a relatively small improvement over time. The vast majority of patients had no problems with walking about (81.3% at baseline, 83.2% at follow-up), with washing or dressing themselves (94.0% at baseline, 94.8% at follow-up), or with performing their usual activities (80.6% at baseline, 85.8% at follow-up). Pain and discomfort were more common, and approximately four of every 10 respondents (39.6% at baseline, 37.5% at follow-up) reported at least some pain or discomfort. Generally, the respondents' emotional QoL was also high: 70.1% of respondents reported no anxiety or depression and 77.6% of respondents were fairly happy to very happy with their life in general, showing high levels of positive affect. The patients' life satisfaction was slightly lower, insofar as 44.0% of patients were dissatisfied to very dissatisfied with the way their life was going. At follow-up, patients emotional QoL was slightly better: 74.1% of respondents were neither anxious nor depressed, 72.4% were fairly happy to very happy, and only 38.8% were dissatisfied with their life in general (additional file 1 displays the descriptive statistics of the sample).

After six moths of ART, 55.1% of patients experienced some adverse effects of the antiretroviral medication. The most reported adverse effects were nausea, skin problems, and dizziness. When we measured the severity of these adverse effects, 58.9% of patients who reported adverse effects experienced them as very disruptive. Approximately six months later, drug substitution and symptomatic treatment had resulted in only 27.3% of patients reporting any adverse effects of treatment. However, 66.7% of these patients experienced these adverse effects as very disruptive. Between baseline and follow-up, 17.7% of the patients changed treatments. Of the patients evaluated, 56.0% started treatment with stavudine, lamivudine, and efavirenz, and 41.8% started treatment with stavudine, lamivudine, and nevirapine. Approximately six months later, 62.3% of patients were using the efavirenz-based regimen and only 36.4% were receiving nevirapine-based first-line ART. A recent clinical study by van Leth et al. (2005) indicated that both regimens are effective in patients with advanced HIV disease and that possible differences, if they exist, are small [53].

The attrition rate of 13% (34 patients) between the two interviews was mainly caused by death (n = 15), refusal to be interviewed (n = 8), and our failure to determine the current whereabouts of the patient (n = 5). Attrition analysis of the study variables (physical and emotional QoL, adverse effects) revealed that responders had significantly better physical and emotional QoL at baseline than those of non-responders and reported fewer adverse effects than did non-responders.

Because SEM is a hybrid of factor analysis and path analysis, a two-step approach is recommended. Separate assessments of the measurement and structural models prevent the good fit of one model compensating for (and potentially masking) the poor fit of the other (additional file 2 displays the correlations among all study variables). Table 1 shows the final measurement model, which has acceptable practical fit indices. As in our previous study, the composite reliability shows the excellent consistency of the indicators in measuring both latent variables: physical and emotional QoL. The discriminant validity of the constructs is supported by the χ² difference test and the variance extracted test. Combined, these findings support the reliability and validity of the two constructs and their indicators.

Table 2 summarizes the parameter estimates of the structural model. The root mean square error of approximation (RMSEA) indicates the closeness of fit of the overall model, with reasonable errors of approximation in the population. Other goodness-of-fit statistics suggest that the structural model not only fits adequately, but also withstands the tests of parsimony (RMSEA = 0.0312, comparative fit index [CFI] = 0.996, parsimony normed fit index [PNFI] = 0.630).

Unlike previous studies, this model included the adverse effects of treatment as a dependent variable on treatment duration and a predictor of physical and emotional QoL. Firstly, treatment duration was significantly associated with the experience of adverse effects (β = -0.38, P < 0.001). Patients at follow-up reported significantly fewer adverse effects than did patients at baseline. The amount of explained variance in the experience of adverse effects was 0.20. Secondly, the variable measuring adverse effects had a strong negative impact on physical QoL. The analysis shows that the respondent's physical QoL decreased by 0.27 standard deviations when the patient reported adverse effects (P < 0.01). The amount of explained variance in physical QoL was 0.15. Finally, these results indicate that adverse effects were also negatively associated with the emotional well-being of ART patients. The significant path coefficient of -0.15 implies that their emotional QoL increased by 0.15 standard deviations in the absence of adverse effects. The squared multiple correlation coefficient, R², for the structural equation predicting emotional QoL was 0.40, indicating that 40% of the variation in emotional QoL can be explained by ART, the associated physical health benefits, and the absence of adverse effects.

The strengths of this study include its longitudinal character and the availability of information on an understudied population. However, it has some limitations. Because of the limited generalizability of the study findings, we can only ascribe the outcomes to patients enrolled in a public-sector ART programme and, more specifically, to patients enrolled in South Africa's Free State Province ART programme. Secondly, although the analysis focused on the impact of treatment duration and adverse effects on physical and emotional QoL, the structural equation model offers an incomplete explanation of the patients' QoL. Patient characteristics (age, sex) and socio-economic traits (education level and income) were tested as predictors of the patients' QoL but did not improve the model significantly. Other potentially relevant predictors of QoL (e.g., ART adherence, symptomatic interventions, and co-treatments) were not available in the data set.