Despite substantial improvements in the affordability and availability of ART in recent years, African health-care systems face enormous challenges in the context of exploding demand for HIV care . The main objective of this study was to identify factors influencing ART adherence and evaluate related outcome of therapy 6 months after ART initiation in a routine setting.
Treatment outcomes and adherence measurements
Our cohort showed a relative low patient retention rate: 70% of the patients who started ART still came to the pharmacy to take their prescription after six months of follow-up, 17% disappeared, 9% died and 4% were referred to other care centres. These results confirm that most losses to follow-up and deaths occur during the initial period after ART initiation.
A recent systematic review  of 33 earlier cohorts in developing countries reported a mean retention rate of 79% at six months. Our slightly lower retention rates may be due to the increasing challenge of managing growing numbers of patients treated.
Pharmacy-refill history gives no description of daily adherence to treatment, because patients may not take all prescribed medications. It could also be considered as a time-consuming monitoring tool for the pharmacy staff, owing to the rapidly growing number of patients in public ART programs. However, this is a simple, inexpensive approach and it was previously reported to be as accurate as CD4 counts for predicting virological response . We found the same correlation between pharmacy-refill adherence and virological outcome at 6 months. In our settings, pharmacy-refill adherence had even greater accuracy, with higher sensitivity and similar specificity to CD4 count changes at 6 months for predicting virological treatment failure. Moreover, data from refill charts in routine-care conditions were available for 95% of the participants, whereas only 23% of the cohort performed their first CD4 cell count follow-up, an analysis that patients had to cover at their own expense. The ability of adherence monitoring from the pharmacy to identify patients at risk of treatment failure may help health-care providers for early adherence counselling interventions. At the time of the study, data from refill charts were kept at the Central Pharmacy and were not communicated to clinicians. A practical implication of our findings is that systematic monitoring of pharmacy-refill adherence should be integrated into therapeutic score cards carried by every patient.
Self-reported adherence has been described in several studies as a rapid and inexpensive method, albeit subject to social-desirability and recall biases [26, 27]. In our cohort, self-reported adherence was not predictive of virological treatment failure. One earlier study in a South African cohort , found only a modest increase in risk (unadjusted OR: 2.35, 95% CI: 1.52-2.53) for patients reporting <100% at 6 weeks for virological failure at 12 months, which might not have been detectable in our smaller population. In addition, the proportion of our participants that reported 100% adherence to treatment during the past month decreased from 83% at one month to 57% at 6 months. Mannheimer  described a similar decline in self-reported adherence with time. It is likely that these decreases reflect decreases in true daily adherence with time. However, there may also have been an overestimation of self-reported medication in early stages due to desire to please when patients do not feel confident in a new environment. Further studies in lower-income settings are needed to verify the accuracy of self-reported medication adherence as a predictor of virological outcome.
Determinants of ART adherence
We examined determinants of adherence in two scenarios to reduce bias due to loss to follow-up. We identified female sex, middle monthly income and less ART related side-effects at one month as predictors of higher pharmacy adherence under both scenarios. Age and HIV clinical CDC stage correlated with pharmacy adherence only under best and worst-case scenarios, respectively. Gender remained of borderline significance after adjusting for potential confounders in the multivariate model.
To the best of our knowledge, the current study is the first to demonstrate that income may not be linearly associated with adherence: patients with monthly middle income had greater pharmacy adherence rates than both the poorest and the richest participants. A recently published meta-analysis  examined the association between socio-economic status and adherence to antiretroviral therapy: out of 8 studies, only 2 prospective studies identified low income as a predictor of non-adherence. All, except one , analysed income as a binary variable which could explain why none of them described our U-shaped association. Selection bias through restricted financial access to health care seems unlikely in our settings: two-thirds (67%) of patients reported no or occasional income, whereas only 25% of our population reported earning more than USD125 per month.
With the exception of a trend towards greater loss rates among men, we failed to demonstrate any other social or demographic association with loss to follow-up. The same gender association with both death and loss to follow-up was recently reported from a study in Malawi .
Limitations of the study areas of uncertainty
It is unclear how far our results can be generalised to other countries and healthcare systems. Compared with reports from patients starting ART in other treatment programmes in lower-income countries , our cohort had a lower proportion of men, fewer patients at clinical advanced disease stages and higher baseline CD4 cell counts.
The percentage of male patients in our cohort reflects the gender distribution of HIV prevalence in Cameroon, which indicates that women's access to health care for HIV is improving. We also observed discrepancies between biological (CD4 counts) and clinical (CDC stages) levels of disease in our participants: Only 14% of our patients started therapy at CDC clinical stage C, despite a median CD4 cell count of 107 cells/μl. The simplest explanation for this is an underestimation of CD4 cell count at the Day Hospital Laboratory. This conclusion is supported by our observation of a large variability of baseline CD4 cell counts for patients who were analysed at close intervals at different laboratories (data not shown). A lack of association between immunological and virological outcomes have been found in similar settings  suggesting that CD4 cell count follow-up should be interpreted with caution, particularly if performed in different laboratories. An alternative explanation would be a systematic clinical misclassification of patients. This is supported by the observation that 75 of 169 patients (44%) enrolled at the YCH Day hospital between 2001 and 2003 were classified as CDC clinical stage C . Such inconsistencies also reflect the constraints on hospital resources and pressure on staff generated by the rapidly increasing number of eligible cases to be evaluated by the Therapeutic Committee every week.
Only 15 patients in our study (5%) lived more than 4 hours of travel from the Day Hospital, which may not be representative of patients in ART programmes at other hospitals. A survey of patients initiating ART from 2002 to 2005 in Limbe Provincial Hospital, the only ART clinic serving the Southwest Province of Cameroon at the time of the study, showed that 44.3% of patients were living more than 40 km by inaccessible road from the clinic . As treatment scale-up programmes are currently attempting to shift ART delivery to health districts in remote areas, more research is needed on geographic access to ART.
Only 312 out of 440 originally eligible patients were included in our study cohort. Such high attrition rates before initiating ART treatment are a main source of selection bias in studies of retention rates in Africa. Data from different cohorts in other lower-income settings  suggest that about 50% of patients lost early to follow-up may have died. This phenomenon needs to be better understood to enable targeted interventions. Without an informed consent, we were not allowed to review all medical records in detail to assess if baseline characteristics of the initial eligible population differed from our cohort.
Our study focused on the first six months of follow-up after ART initiation, a period previously characterized by high attrition [5, 18, 19, 32] and thus considered to be a crucial phase for promoting adherence. As level of adherence and its predictors may vary over time, we strongly encourage ART programs to conduct long-term surveillance of these outcomes to fully understand the subtle variations of its dynamic behavioral process.