In our study we found a relatively low proportion of virological failure (6.0%) following 2010 WHO definition (VL >5000 copies/ml) in patients on ART for more than 12 months as reported in other LMICs, supporting the fact that ART can be provided in resource-poor settings with favorable outcomes 
In this study, the PPV of clinical or immunological monitoring for detecting virological treatment failure was relatively low. Mee et al reported a PPV of CD4 count of 36.8% while Kaiser et al recorded PPV ranging from 9.5%–28.7%. 
, which could result in patients with adequate viral suppression being incorrectly identified as failing treatment and being unnecessarily switched to second-line therapy 
. This would not only reduce treatment options for patients but also potentially increase costs and make follow-up of patients receiving protease inhibitors more difficult. An algorithm for determining treatment failure based on clinical history, hemoglobin level, and CD4 cell count has recently been proposed, but it has not been validated in routine clinical care 
Low sensitivity of clinical and immunological criteria to define treatment failure highlights the need for improved methods to detect treatment failure in the absence of VL testing. In our study, only 8.1% of patients with clinical failure and 24.5% of those with immunological failure were found to have virological failure.
Only 3/55 (5.45%) of patients with confirmed virological failure met both clinical and immunological criteria for treatment failure, and 35/55 (63%) of the patients with virological failure did not meet both clinical and immunological definitions of failure, showing that patients with VL >5,000 copies/mL might not meet any of the currently used criteria to detect treatment failure. Many treatment failures may therefore be missed using only clinical and immunological criteria, which could lead to accumulated resistance in patients who continue on failing regimens.
An evaluation from resources limited countries found no evidence of improved mortality in programs with viral load test, though follow-up was short. 
. On the other hand, several studies have concluded that clinical indicators and CD4 cell count are not favorable predictors of virological failure and routine laboratory monitoring is associated with improved health and survival when compared with clinical monitoring alone. 
In HIV high-prevalence, resource-poor settings, where task shifting takes place to scale up ART, sensitive models are needed to accurately detect treatment failure when VL testing is unavailable.
These results from a Kenyan ART program also illustrate the difficulties faced by other African countries in implementing the new WHO recommendations for ART initiation 
, moving to improved first-line regimens containing tenofovir (TDF) or zidovudine (AZT) in patients who have already been treated with d4T-based regimen and might have treatment failure. In the ideal scenario of universal access to VL testing, every patient could be assessed before being switched from a first- to second-line regimen, but because this is not the case in most resource-limited settings, many patients might be switched to a regimen which is the only available second-line therapy in LMICs.
Follow-up of patients with VL measurements seems to be the only way to adequately monitor the patients, and VL appears to be the most reliable tool for deciding when to switch failing regimens for patients 
. In programs with access to VL monitoring, patients tended to switch treatment earlier and at higher CD4 cell counts than at sites without VL 
. Despite the evidence, VL testing is not yet widely available for monitoring of patients on ART in resource-poor settings, and no other simple tools exist for treatment failure detection. New VL assays meeting specifications for use in resource-constrained settings are urgently needed to tackle the current needs of ART monitoring and clinical assistance for treatment decision-making.
A strength of our study was in the assessment of clinical outcomes since they were systematically collected in patient’s files, allowing us to examine the correlation between clinical and immunological criteria together with viral load measurement. Another strength of the study was that it was done in a routine ART program in a resource-limited setting including decentralized rural clinics, which reflects the reality of other sub-Saharan African countries.
A limitation of the study lay in not being able to analyze adherence despite the data obtained through questionnaires since it was impossible to find standard definitions using the current self-reported and visual-analogue scale. Another study weakness was the limited diagnostic capacities for the main opportunistic infections seen in our program, which could bias some of the clinical events registered.
This study builds on existing literature and builds the case that clinical and immunologic criteria, given low sensitivity, allow for individuals to switch to expensive second-line who may not have true virological failure.
In conclusion, these data illustrate the urgent needs for new or improved algorithms for measuring clinical or immunological treatment failure and wider access to VL monitoring in low-resource settings. Using current WHO immunological and clinical criteria to determine virological treatment failure is inadequate in a setting were VL is not widely available and second-line ART options are limited.