Quantitative data were initially stored in an Access database (Microsoft Access, 2003). Statistics were generated using Epi Info and SPSS version 13.0. The crude prevalence of adherence was estimated and its 95% confidence interval calculated. The Chi-square test was used to compare adherence rates among two or more categories. Logistic regression models were used to determine predictors of adherence and to estimate the independent and multiple effects of selected factors on adherence. All hypotheses were tested using α = 0.05 level of significance.
Adherence rates, measured as the percentage of pill intake over a specified time, were estimated using three methods: two-day recall using a 'sun and moon chart', which depicted the sun at different times of the day and the moon at night; visual analogue (a one-month recall using an uncalibrated 10 cm line); and a one-month pharmacy pill count. In the visual analogue, respondents were requested to indicate, by marking on the line, how they perceived their adherence over the past month. The overall adherence rate was estimated as a composite measure (i.e. the average of the one-month visual analogue, pharmacy pill count (one-month) and the two-day recall.
Pill count (one-month)
Pill counts were calculated by subtracting the number of pills returned from the number of pills issued. This provided the amount of medication used by the patient during this period. The amount used is then divided by the expected amount and multiplied by 100 to determine the percentage adherence per participant.
Self-report (two-day recall)
In the two-day recall the patients were asked to recall the frequency and timing of medication as well as their food intake over the previous two days. The data were captured in the sun and moon chart.
Self-report (one-month recall)
Participants were asked to indicate their adherence rate using a visual analogue line measuring 10 cm. The distance from zero to the tick on the line multiplied by 10 was considered to be the estimated percentage adherence rate.
Most of the patients interviewed were on first-line regimens, which include: efavirenz, lamivudine and zidovudine. These do not have any food requirements. However, some patients may choose to take them after meals to reduce nausea. Therefore since most of the ARV combinations used for first-line regimen in Botswana do not necessarily require that they be taken with food, the variable timing of taking medication and whether the drugs were taken with food or not were dropped in the analysis.
The qualitative data collected were analysed with a view to gaining understanding of the factors that influenced adherence to ART. The data analysis process included a four-day workshop, with technical assistance provided by the University of Amsterdam. The work involved reading through the data from the qualitative research tools - which included the semi-structured interviews with health workers and ARV users, and the FGDs with ARV users and the community - in order to identify key themes. Initially, 28 themes were identified. The quotes were then manually pasted onto theme cards for easy perusal. A general thematic analysis was then conducted, focusing on similarities and differences of perspective between different groups of respondents. Further analysis revealed that the themes appeared to be linked, and these were then analysed together. Information was analysed to capture the different perspectives of the different actors: ARV users, health workers and community members. Where there were agreements or conflicting views, these were shown.