The results of analyses of the CIPRA-SA trial data show a clearly increased risk of death or virologic failure associated with initiating ART at lower CD4 counts. We found that those who started ART at CD4 counts ≤200 had roughly twice the risk of death or virologic failure as those initiated at CD4 counts >200 (HR: 1.94; 95% CI: 1.14–3.30) and twice the risk of developing incident tuberculosis (HR:1.90; 95% CI 0.89–4.04). These findings are in line with numerous observational studies from resource-limited settings showing low baseline CD4 count is a major predictor of death and loss to follow-up[23
]. Recently an interim analysis of the CIPRA-HT001 trial in Haiti showed a nearly four-fold increased risk of death among those started with a CD4 count ≤200 vs. 200–350,[19
] very similar to our five-fold increased risk (RR: 5.4; 95% CI: 1.3–23.0)[19
]. Thus a body of evidence is beginning to emerge showing the benefits of earlier treatment initiation in resource-limited settings. This, along with a recent analysis by Lawn and colleagues[28
] showing that the longer a patient maintains a CD4 count <100, the higher the risk of death suggest that starting patients at higher CD4 counts may allow them to maintain their CD4 count above the point at which they are at increased risk of death.
Our findings are also consistent with data from resource rich environments. Observational data has shown that higher CD4 counts are associated with lower risk of death[7
]. More recently, Kitahata and colleagues have shown[13
] that patients initiating ART at CD4 counts >500 had substantially reduced risk of mortality vs. those below. While our data cannot be used to comment on CD4 counts >350, our finding of decreased mortality and virologic failure risk associated with having a starting CD4 count >300 is suggestive of a dose-response with increasing baseline CD4 count associated with better outcomes.
While we found a substantial benefit to earlier ART initiation, we also found a slightly increased risk of being lost to follow-up among those with higher baseline CD4 counts which could potentially offset some of the benefits of initiating treatment earlier. However, we urge caution in interpreting these results. Under ideal conditions, assessing the effectiveness of initiating ART at higher CD4 counts would come from randomizing patients to either: 1) immediate initiation of ART when the CD4 count falls <350; and 2) follow patients and delay ART until the CD4 count falls <200[14
]. In both arms patients would be followed from the time of their first CD4 count <350. In our study patients were initiated onto ART at enrollment as long as their CD4 count was <350 so we do not have any follow-up time to approximate what happens to patients in the time their CD4 count is between 200–350; however, we anticipate some death and loss to follow-up occurs in this time. While methods exist to adjust for this lead time bias,[30
] they require pre-ART data which we did not have.
The current analysis has several strengths. The data were from a large prospective randomized trial with excellent follow-up data at standardized intervals which allowed the assessment of the impact of starting treatment at higher CD4 counts. Although the data was from a randomized trial of another intervention, because the trial showed no differences between randomization groups (i.e. nurse- vs. doctor-monitored care) and because adjustment for randomization group had no impact on our current results, there is little evidence that the primary intervention had any impact on our findings.
Still the current analysis should be considered in light of several limitations. First, as noted above, we did not have the ideal comparison group to assess death and virologic failure (i.e. a group followed from a CD4 count of 350 until 200 and then initiated on ART). Thus, any deaths occurring between 350-200 would not be included in our analysis. Since we are missing deaths in the high CD4 count group, this analysis likely underestimates the treatment benefits of starting at higher CD4 counts. Thus our estimates cannot be considered the true causal effect of starting treatment at higher CD4 counts. Second, as the data came from a randomized trial with conservative definitions of toxicity, many subjects who were treatment failures for toxicity might have otherwise continued on treatment under usual practice conditions. This could have biased our toxicity results towards the null and prevented us from observing a true difference between the groups if one existed. Third, in our analysis of loss to follow-up, we were not able to determine the final outcomes of patients lost and therefore we cannot say if patients left care because they were feeling well nor could we determine how many of them have since died. Finally, as the data were from a trial, the study population may have been healthier than the general clinic population.
In conclusion, we found that patients initiated on standard first-line South African ART regimens were at increased risk of death and virologic failure if initiated at CD4 counts <200 compared to those initiated >200. This is consistent with findings from developed areas which have shown that the benefits of starting at earlier CD4 counts outweigh the risks of toxicity and long-term adherence. While the cost implications are unknown, national and international guidelines on the topic of when to initiate ART should consider our findings when deciding on whether to increase initiating CD4 count thresholds. If thresholds are increased, substantial efforts will need to be made to move patients into care earlier in their disease progression in order to obtain the maximum benefit from ART.