Discordant Immunologic and Virologic Responses to Antiretroviral Therapy and Associated Mortality in a Large Treatment Program in Rwanda

To the Editors: We read with interest the study by Tan et al1 showing that discordant immunologic and virologic responses at 3-9 months after starting highly active antiretroviral treatment play important roles in predicting long-term clinical outcomes (incidence of opportunistic disease and death) in treatment-naive patients. Although the prevalence and risk factors for discordant responses in low-income countries are thought to be similar to those in high-income countries,2 the prognostic significance of discordant responses has not been assessed in the former, where there is clearly a different burden of opportunistic disease. The patient population too is different in low-income settings where it is predominantly female, individuals often present with advanced disease, and patients starting treatment are mostly treatment naive and generally start antiretroviral therapy (ART) with a first-line ART regimen containing a nonnucleoside reverse transcriptase inhibitor. For this reason, we evaluated the association between discordant immunologic and virologic responses and short-term mortality in a relatively large ART cohort in Rwanda between October 2003 and July 2007. Médecins Sans Frontières has been supporting an ART program in 2 urban government health centers in Kigali, Rwanda, since 2003, with currently over 3000 patients started on treatment and with high patient retention (94% by 1 year).3,4 Patients were started on ART according to World Health Organization eligibility criteria. Approximately 90% of this cohort was started on generic fixed-dose combination regimens containing stavudine, lamuvudine, and nevirapine. Viral load (VL) measurements were performed routinely after 1 year of treatment. Virologic success (VL+) was defined as a VL 50 cells per microliter from baseline. Virologic and immunologic responses were categorized as concordant or discordant and subgrouped as follows: concordant responses were either concordant positive responders (VL+/CD4+) or concordant nonresponders (VL−/CD4−); discordant responders were differentiated into patients displaying a virologic-only (VL+/CD4-) or an immunologic-only (VL−/CD4+) response on ART. The all-cause mortality rate, starting from the time of VL measurement, was calculated per response category. Kaplan-Meier estimates were used to compare survival between groups. The association of response category with risk of mortality was assessed using a Cox proportional hazards model. A total of 1251 ART-naive adults were started on treatment during the study period, of whom, 289 had incomplete CD4 data and were excluded. Of the 962 patients who were included in the analysis, VL testing was done at a median period of 1.3 years after starting ART. The median baseline CD4 count was 149 cells per microliter, and most patients were in World Health Organization clinical stage II or III at baseline (Table 1). In total, 627 (65.2%) were concordant responders, and 46 (4.8%) were concordant nonresponders. Discordant responses were seen in 289 patients (30.0%), with 172 (17.8%) virologic-only responses (VL+/CD4−) and 117 (12.2%) immunologic-only responses (VL−/CD4+). The median follow-up period after VL determination was 13 months (interquartile range 6-15 months). Over more than 903 patient-years of follow-up, only 11 deaths were observed, reflecting (besides the relatively short follow-up period) the fact that these patients had been on ART for more than a year and hence had passed the more pronounced early mortality phase during the first few months of ART. Still, a difference in mortality rate between response categories was observed, ranging from 5.0 of 1000 patient-years for concordant responders to 54.2 of 1000 patient-years for concordant nonresponses. Discordant responses had a mortality rate of 28.8 of 1000 patient-years for virological-only responses and 11.4 of 1000 patient-years for immunological-only responders. Kaplan-Meier estimates summarizing the time to death by response category also showed significant differences between the categories (log-rank test: P = 0.004), with an estimated 1-year mortality (after response category determination) ranging from 0.4% for concordant responders, 3.4% for discordant responders, and 5.8% for concordant nonresponders. Relative to the concordant positive responders, a significant difference in prognosis was seen for concordant nonresponders (P < 0.001) and for discordant responders (P = 0.005). No significant difference between both discordant response categories was found (P = 0.373). In multivariate analysis, both a discordant (hazard ratio 4.7; P = 0.03) and a concordant nonresponse (hazard ratio 8.5; P = 0.02) remained associated with increased mortality (Table 1). In a separate analysis, a threshold of 1000 copies per milliliter was used, to exclude viral “blips,” which essentially provided similar results (data not shown).TABLE 1: Cox Proportional Hazard Analysis of All-Cause Mortality by Immunologic and Virologic Response Categories*Given the constrained resources, for health facilities in resource-limited settings that are faced with high patient burdens and shortages of staff (both in terms of numbers and capacity), it would be useful to focus more specialized medical attention toward those most at risk. Our data are in line with those of Tan et al1 and suggest that patients with discordant responses represent a population at risk. The fact that a significant difference with concordant responders was observed within a relatively short period suggests that the risk is likely to be substantial. There are a number of unanswered questions. Is there a relation between the level of drug adherence over time and the discordant response category? Do subclinical or undiagnosed opportunistic infections particularly tuberculosis play a role in the incidence or outcomes of discordant responses? In this regard, the observed association between the occurrence of tuberculosis while on ART and virological-only responses might be of interest (data not shown). In these resource-limited settings, does nutrition, helminth infection,5 or other inflammatory conditions interfere with immune recovery on ART or does a poor immune response reflect irreversible damage due to advanced HIV disease? Would patients with immunologic-only responses benefit from earlier switches to protease inhibitor therapy?6 Finally, would isoniazid preventive therapy or other adjunctive treatments reduce the incidence and/or associated mortality of discordant responses? The answers to these questions merit further research. There are a number of limitations to this analysis. First, it is an observational study performed retrospectively on existing data, and as such residual confounding cannot be excluded. More detailed data on adherence, clinical events, laboratory results, and baseline and repeat VL measurements might have been informative but were not available due to resource limitations. The strengths of this study are that the data come from a routine setting and are thus likely to reflect the operational reality on the ground; patients lost to follow-up were traced by community support groups, and deaths were reliably ascertained; and finally, in contrast to the study of Tan et al,1 where the response category was determined at 3-9 months after ART initiation, this was done after at least 1 year on treatment in our cohort. The latter probably better reflects what is achievable in most resource-constrained settings, with limited access to VL tests. Anyhow, our data provide another strong argument for wider availability of VL tests in these settings. In conclusion, discordant responders and concordant nonresponders are a population at risk in low-income countries meriting targeted medical attention. Better understanding of the mechanisms underlying discordant responses would allow the design of management strategies to reduce associated mortality in low-income countries. With currently more than 3 million patients on highly active antiretroviral treatment in low-income countries, of whom, approximately 1 in 3 (30%) might be discordant responders who are at relatively high risk of death even while on ART, this issue is of high relevance to clinical management and public health. ACKNOWLEDGMENTS We are grateful to all the staff of the Kimironko and Kininya health centers for their work on HIV/AIDS. This HIV/AIDS program was supported by Médecins Sans Frontieres and ran in collaboration with the Rwandese Ministry of Health. We would like to thank the Rwandese Ministry of Health for the excellent collaboration. Johan van Griensven, MD, PhD* Rony Zachariah, MD, PhD† Freya Rasschaert, MD† Tony Reid, MD† *Médecins Sans Frontières Kigali, Rwanda †Medical Department Médecins Sans Frontieres Operational Center Brussels Brussels, Belgium