This study adds to the literature assessing the impact of the scale-up of ART services in Africa by providing a method for estimating retention in care that accounts for the large fraction of patients initially lost to follow-up (i.e., who have unknown outcomes). Using this method, we found that between 78.9% and 86.2% of the clinic patient population starting ART during this time were retained in care two years after ART initiation. This paints a different picture of the effectiveness of ART delivery than our naïve estimate of 68.9% or previous work which suggested at as few as 50% were retained at that time 
. Although there is no clear threshold at which retention in care is considered adequate, these figures are disparate enough to potentially represent “failure” on the one hand and “success” on the other.
This study also proposes a metric termed “connection to care” that may further the epidemiologic discourse on engagement with ART services in Africa. Given a lack of sophisticated diagnostic or therapeutic capabilities, some patients (such as those with smear negative tuberculosis or malignancies) will likely die despite engagement with the package of interventions centered on ART. The metric of “retention in care” groups deaths in patients who were essentially in care at the time of death with living patients who cease to access care, many of who die after disengagement. As such, it is a measure of the total effect of ART services including the success of clinical care as well as continued access to care. The estimate of “connection to care,” in contrast, captures the fraction of patients who are in care even if they died while in care (i.e., shortly after a clinic visit) and is therefore a more precise measure of continued access to public health services in resource limited settings. Although deaths shortly after the last clinic visit may be due to a limitation of the diagnostic technologies or treatment modalities available, these deaths are less likely attributable to a problem with access to the ART care system once initial linkage was established. We found approximately half of patients who were not retained in care died before a long interval of absence from clinic (3 months) while the other half “disconnected” from care while still alive. This implies both improvement of clinical services (at the last clinic visit) and outreach (to re-engage those out of care) in equal measure are needed to optimize patient outcomes.
Although large implementing organizations and national programs are decentralizing ART services 
, few studies have documented the movement of patients from older centralized to newer and lower level ART clinics and considered the implications of this movement on estimates of retention 
. We found that among those lost from the ISS Clinic in Mbarara, each additional calendar year of last visit conferred over a 4-fold rise in the odds of being retained in care while each 10 kilometers of distance from residence to clinic conferred almost a 50% rise in the odds of retention in care at another site. In the setting of southwestern Uganda, these associations are unsurprising: while in 2000 the ISS Clinic was the only ART delivery site in the region, by 2009 over 60 ART delivery centers had opened in southwestern Uganda. These data are consistent with the hypothesis that high loss to follow-up is in part due to “silent transfers” during the decentralization of ART services. An alternative interpretation of these data is that losses to follow-up at a particular clinic are due to a large extent to incomplete penetration of ART delivery into the community.
This study has several important limitations. First, we cannot be sure that the patients whom we deemed to be retained in care after transferring to another clinic did not also experience a break in ART with its attendant virologic and immunologic consequences 
. Research on the “safety” of transfers (whether silent or documented) is urgently required to answer this question. Second, it is possible that patients who have truly stopped ART may nonetheless report ART use due to perceived social desirability. The tracker, however, was not a medical provider and hence have had a more limited influence on reporting and all patients who reported a “silent transfer” also provided names of legitimate ART providers in the area – providing some additional support for the veracity of their reports. Third, we were unable to directly ascertain whether the 39% of patients reported to be alive by an informant were in HIV care. Sensitivity analysis, however, yielded figures that departed significantly from the naïve estimates, and hence the central inference that retention in care is underestimated remains unchanged. Finally, our sample was not formally random and we did not ascertain outcomes in 100% of the sample and so our estimates may be biased.
In summary, a sampling-based approach and competing risk analyses can provide estimates of retention in care and connection to care among HIV-infected patients on ART in Africa where resource limitations preclude ascertainment of outcomes in all patients. In estimates using both pessimistic and optimistic assumptions, we found retention in care and connection to care in a patient population of ART users in Southwestern Uganda to be higher than existing studies suggest. Among those lost, patients who lived farther from the clinic and who became lost later in calendar time were more likely to be retained in care elsewhere —an observation consistent with the presence of “silent transfers” during a period of decentralization of ART services in southwestern Uganda. Use of a sampling-based approach at “sentinel sites” or more widely within clinic based cohorts to estimate retention in care and connection to care across diverse setting can help to improve evaluation of the global effort to deliver ART.