There is growing recognition that adolescents have greater challenges to adherence on ART compared to adults. However, published studies from cohorts of this age group from resource-limited settings are limited and little consistency in the age categories used in published studies, making comparisons difficult 
. In addition the largely unrecognised epidemic of late progressors in the young and older adolescent age groups is evidenced by the almost equal sex distribution in these age groups in our setting. Across Southern Africa an increasing pool of survivors from mother-to-child transmission programmes are entering into adolescence, with data suggesting that up to one third of HIV infected infants, likely to have been infected during the breast feeding period, are “slow progressors”” with a median survival of greater than ten years. In Zimbabwe, mathematical modelling estimates that deaths among untreated slow progressors will increase from 8000 per year in 2008 to a peak of 9700 per year in 2014. Hence, identifying these children and addressing their needs once on treatment is an essential challenge for antiretroviral therapy (ART) programmes in the coming years 
In this study, from a rural primary health care programmes in a resource-limited settings, the age group with the worst outcomes were the young adults (19.1–24 years). Cited challenges to medication adherence in this group include side effects, appointment schedules that interfere with daily schedules, depression, stigmatisation, fears about disease transmission, transfer to adult HIV services 
, and disclosure of HIV status 
. In this setting partial disclosure is aimed for between the ages of seven to nine with full disclosure completed no later than age twelve. While no formal qualitative surveys have been done in this programme, feedback from experienced counsellors working in these sites suggested that context specific challenges to retention in care for the young adult group also highlighted during informal focus included: the transition out of education, the need to move away in order to seek work (moving to an urban setting or moving across a border to South Africa or Botswana); and entering into more serious relationships and marriage where new disclosure was needed.
Other factors that may have impacted on rates of retention rates during the period analysed included the fact that decentralisation to the network of primary care clinics was gradually being implemented with patients having to travel up to 3 hours to reach a clinic. In addition, the economic and political instability experienced in Zimbabwe during this period may have led to patients defaulting due to inability to pay for transport or due to forced movement out of the district.
In a survey carried out between 1998 and 2005 in Manicaland, Zimbabwe (the same province as this study site) the median age at first sex and first marriage was 18.5 years and 21.4 years for men and 18.2 years and 18.5 years for women respectively 
. Marriage and giving birth, particularly to a first born, may be a reason for patients to move away from their original clinic. Many women will travel to be near a parent or in laws depending on cultural norms and hence be lost within the ART cohort.
Various counselling and service delivery models of care have been developed in western settings targeting adolescents. These have also included the development of “transitioning protocols” from specialised adolescent clinics to the standard adult services 
. However, in resource-limited settings where counsellors are often lay people with relatively low levels of education, it may not be feasible to implement specialised counselling techniques.
Simple steps to addressing the major challenges to retention in care should be incorporated into existing counselling tools and service delivery models. One example being employed in one South African adolescent project is incorporating the theme of “life steps” into routine counselling that aims to address the major challenges around having to seek work and entering into new relationships. Whether peer support for this age group is helpful is questionable as when asked, many stated that they would prefer to be counselled by someone older than themselves.
Adapting both provision of ART and monitoring and evaluation tools to the challenge of mobility of this age group should also be integrated into programme implementation. Counsellors and nurses should proactively and routinely ask about travel plans for work, education or social reasons and allow for provision of longer drug supplies and a degree of flexibility in often rigid follow up schedules. Being proactive in this way should allow formal transfers out to be organised to alternative ART sites when needed. In addition it is essential to provide a clearly documented HIV history to the patient, the so called patient passport, enabling the health professional at the new site to continue care in an appropriate manner.
In addition to adapting approaches to the socioeconomic challenges in this age group, simplification of drug delivery may also be another route to aid adherence. Already the introduction of guidance to use a once-daily, fixed-dose combination of tenofovir combined with lamivudine and efavirenz from the age of twelve is expected to simplify adherence and reduce side-effects 
. In countries where such optimized regimens are still unavailable for the general population due to cost concerns it may be reasonable to consider prioritizing their use in this high risk age group. Long acting formulations that are in development could also be ideal to pilot for these age groups. The use of mobile health technologies and pre-existing social network platforms may be another attractive means of giving additional support mechanisms to this vulnerable age group.
As with any observational study there were several limitations to the analysis. Incomplete baseline CD4 data did not allow for appropriate adjustment. Adjustment for baseline CD4 count using a separate category for missing values might have introduced bias. This is particularly true in situations where the percentage of missing data is high 
. Restricting the analysis to records with complete CD4 count data will also result in bias except when the variable is missing completely at random. The proportion of missing CD4 counts was similar across age groups and outcomes (death and loss to follow-up). A high proportion (32.8%) of CD4 counts were missing for operational reasons, as the country was going through a period of civil unrest with breakdown of infrastructure. Therefore CD4 count data were likely to be missing completely at random and analysis of the complete case scenario can be expected to result in unbiased results. Incomplete outcome verification for patients lost to follow up may lead to possible misclassification of true death as loss to follow up 
. Nevertheless, when deaths and losses were combined as a single adverse outcome (attrition), young adults still did worse. Finally, the fact that a proportion of adolescents may have been perinatally infected may reflect a survival bias in this group.
In conclusion, this analysis highlights the young adult group as being at highest risk of being lost from care compared to adolescent and adults in a resource poor rural setting. Adapting adherence support and service delivery models for this group should be a priority to avoid treatment interruptions, development of resistance and increased morbidity within this group.