Early detection of virological failure is important for optimal management of HIV-infected patients receiving ART. Patients who continue to receive a failing regimen are at risk of immunological failure, morbidity and death. Moreover, accumulation of multiple antiretroviral drug resistance mutations may compromise the response to future drugs and fuel the spread of primary drug resistance within communities. Since VL monitoring is not available in most resource-limited settings, we investigated the utility of CD4 cell count measurements for predicting virological failure in a cohort of South African patients. Baseline absolute CD4 cell counts as well as clinical and socio-demographic characteristics were not predictive of virological failure. Analyses of longitudinal data from those who developed virological failure revealed that absolute CD4 cell counts and CD4 cell count changes (ΔCD4 cell counts and CD4 cell count slopes) were significantly correlated with viral load measurements at a group level. However, subsequent analysis showed that none of these methods of analysing CD4 cell counts could be used to identify individual patients at the time they developed virological failure. Since the distributions of CD4 cell counts and CD4 cell count changes among those with virological failure did not differ significantly from those of patients who maintained virological suppression, these could not be used to provide a clinically useful means for individual patient assessment for virological failure.
A unique feature of our study is that we used a novel modelling approach that accounted for all CD4 cell count and VL values measured during follow-up from the first date that VL suppression was achieved. Some previous studies have modelled the difference between the CD4 cell counts measured at initiation of treatment and at a single arbitrary point during ART defined a priori
. Other studies assessing factors associated with virological failure did not account for all CD4 cell count measurements performed during follow-up [15
]. Neither of these approaches fully evaluates CD4 cell count dynamics during ART, increasing the potential for misclassification bias. Our analytic approach also differs in that we modelled virological failure as the end-point of interest rather than virological treatment success as reported elsewhere. We used a VL > 1,000 copies/ml to define treatment failure consistent with local protocols. However, the same outcomes were obtained using thresholds of > 400 and > 10,000 copies/ml, which is consistent with previous data from this setting [25
Baseline CD4 cell count was not predictive of virological failure in this ART-naïve population. However, in patients who developed virological failure, absolute CD4 cell count measurements, ΔCD4 cell counts and CD4 cell count slopes during ART each correlated significantly with VL measurements taken at the same time-points. Of these three parameters, the CD4 cell count slope was the most strongly correlated. This indicates that the rate of increase or decrease of CD4 cell count at a given time-point was the parameter that was most strongly associated with current VL. However, the distributions of ΔCD4 cell count and CD4 cell count slope values were very broad even among patients who maintained virological suppression. This suggests that considerable fluctuations in CD4 cell counts occur among patients despite sustained virological control. When these distributions were compared with the distributions of data from patients who had current virological failure, they almost completely overlapped. This demonstrated that absolute CD4 cell counts and CD4 cell count changes could not be used to identify patients who have developed virological failure. These findings were further corroborated by the observation that the distributions of CD4 cell count changes in the 30 patients who never achieved virological suppression were also broadly overlapping with the distributions of data from those who maintained virological suppression.
To investigate these associations further, we focussed on the use of CD4 cell count slopes since this was the parameter most strongly associated with VL at a group level. However, ROC curve analysis confirmed that use of CD4 slopes provided very poor test characteristics for predicting virological failure. The specificity and sensitivity of a negative CD4 cell count slope was low, showing that this parameter was not of practical utility in this clinical setting. Furthermore, the data show that a negative CD4 cell count slope could not even be used as a screen to identify those at high risk of virological failure as a means of rationing scarce viral load monitoring resources.
A strength of this study is that patients were closely followed in a multicentre clinical trials unit with strict protocols for regular clinical and laboratory monitoring every 2–3 months, leading to reliable identification of virological failure. As soon as a VL > 1,000 copies/ml was first detected, confirmatory viral load testing was done. The cohort characteristics were diverse and so the data are not only relevant to those with advanced immunodeficiency. Despite differing cohort characteristics, follow-up procedures and analytic approaches, our data are consistent with and extend previous studies that have found a poor association between CD4 cell counts and the development of virological failure [8
We acknowledge the limitations of this study. An important potential limitation is that all patients studied were ART-naïve. Therefore these findings may not be generalisable to treatment-experienced patients. Our patients participated in international multicentre clinical trials. Their experience may differ from that of patients accessing treatment in a community-based setting. We do not have good assessments of treatment compliance although the mechanism underlying virological failure is unlikely to affect the relationship between CD4 cell counts and viral load. Despite a limited cohort size, follow-up in this study was prolonged, a substantial proportion developed virological failure and the number of paired CD4 cell counts and VL measurements was large.