HIV-infected Rwandan women have a high frequency of long-term survival

Objectives: To evaluate rates of long-term survival in a prospective, longitudinal, closed HIV cohort in Africa between 1986 and 2006.

Methods: A total of 548 HIV-infected Rwandan women were recruited from prenatal clinics in Kigali and followed at 3-6 month intervals to February 2006. Overall, 401 women (73%) were HIV positive at initial cross-sectional testing in 1986 (seroprevalent cohort) and 147 women (27%) were initially HIV negative but seroconverted during follow-up from 1986 to 1993 (seroincident cohort). Kaplan-Meier survival methods were used to calculate survival times censored in mid-2003.

Results: In February 2006, 109 women (20%) remained alive in the cohort. Time to 50% non-genocide mortality was 11.9 years among seroincident women and 8.9 years among seroprevalent women. Smoothed mortality rates increased with duration of follow-up to a peak of 0.12 deaths per person-year at 9.5 years of follow-up but subsequently declined. After 15 years of follow-up (pre-HAART introduction), the survival probability was 36% for seroincident women and 26% for seroprevalent women. Most survivors had virological and immunological evidence of disease progression. The median CD4 cell count of survivors declined from 447 cells/μl in 1998 to 268 cells/μl in 2003. Among survivors, 57 women (52%) met treatment criteria and initiated antiretroviral treatment by 2006.

Conclusion: Although median survival times in this cohort were similar to those observed in high-income countries, the rates of long-term survival after 15 years of follow-up were higher than expected. A levelling off of mortality rates during the late stages of follow-up may explain this high rate of long-term survival.

Introduction

Despite the devastating impact of HIV in sub-Saharan Africa ^[1], few studies have evaluated survival with HIV-1 infection in African populations. Large seroconverter cohort studies in Europe, north America, and Australia have proved essential to understand the natural history of HIV-1 disease in high-income countries ^[2,3]. An international collaboration of seroconverter cohorts in high-income countries estimated that the median survival from seroconversion in the pre-HAART era was 10.9 years [95% confidence interval (CI) 10.6-11.3 years] for individuals aged 25-34 years ^[2]. These cohorts demonstrated remarkable variability in rates of disease progression. At the extremes in an age-adjusted analysis, 10% of individuals died within 5 years of seroconversion and 38% of individuals were still alive 13 years after seroconversion ^[2]. In addition, a small subset of HIV-infected individuals have demonstrated minimal evidence of disease progression after 10-15 years of HIV-1 infection ^[4]. In high-income countries, these survival probabilities have been described by a Weibull distribution (a flexible continuous probability distribution) with a mortality rate that monotonically increases over time ^[5].

Relatively small seroconverter studies in resource-limited settings have observed different rates of survival. In a cohort of 42 patients from Haiti, the estimated median survival from seroconversion was 7.4 years ^[6]. A cohort of 235 young Thai men also observed a relatively rapid progression to AIDS and death ^[7]. In Africa, a cohort study of 163 female commercial sex workers with known dates of seroconversion in Nairobi, Kenya, estimated a median duration of only 4.4 years from seroconversion to AIDS (Centers for Disease Control and Prevention stage IV C disease) using a Weibull survival model ^[8]. More recent evidence, however, has suggested that survival with HIV-1 infection in Africa may be longer than previously thought. A cohort of 168 seroincident cases from rural Uganda had a median survival of 9.8 years ^[9]. In addition, a cohort of 1950 South African miners had a median survival of 10.5 years, which is similar to the rates observed in high-income countries ^[10].

An accurate understanding of the natural history of HIV-1 infection in Africa is essential to plan and evaluate healthcare inventions, such as the expansion of HAART. We therefore evaluated long-term survival in a cohort of HIV-infected women in Rwanda from 1986 to 2006.

Methods

Study population

This study took place at Projet San Francisco, a research clinic in Kigali, Rwanda, founded in association with the Rwandan Ministry of Health. Projet San Francisco has cared for HIV-infected patients since 1986 and has been well described previously ^[11-14]. Although HIV testing was performed in 1986, the baseline date for seroprevalent participants was left-censored to their study enrollment date in 1988 for all analyses. In 1986, 3702 consecutive women, from 18 to 35 years of age, were recruited from prenatal and pediatric clinics at the Centre Hospitalier de Kigali and tested for HIV-1. Testing for HIV-1 was performed using an enzyme-linked immunosorbent assay (Wellcome Diagnostics, Research Triangle Park, North Carolina, USA), with confirmation by immunofluorescence (Virion, Lucerne, Switzerland) or Western blot (DuPont, Wilmington, Delaware, USA). Women received their HIV test results confidentially with pre- and post-test counselling.

In 1988, a random sample of 401 HIV-infected and 1057 HIV-negative women was selected from these tested women for a prospective cohort study ^[11-14]. Participants were followed at 3-6 month intervals. HIV-negative women had repeat testing for HIV at each visit and 147 seroconversions were identified from 1986 to 1993. The median interval from the last negative HIV test to the first positive HIV test was 274 days [interquartile range (IQR) 46-522 days]. The estimated date of seroconversion was left-censored to the date of the first positive HIV test for all analyses. Projet San Francisco was temporarily forced to close in 1994 during the Rwandan genocide, but reopened when the new government took control. After 1994 only HIV-infected women were followed in the study. Beginning in 2003, generic antiretroviral drugs were provided free of charge to patients who met World Health Organization (WHO) and national guidelines as part of an antiretroviral treatment programme initiated with private funding and expanded through a grant from the Global Fund to fight AIDS, Tuberculosis, and Malaria in mid-2004.

Clinical and laboratory follow-up procedures

In 1988 all women in the prospective cohort study completed an evaluation of their demographics, socioeconomic factors, and past medical history. Interviews and examinations were conducted by Rwandan nurses and physicians in their native language, Kinyarwanda. Participants were seen at scheduled 3-6 month intervals to complete a medical history and weight measurement. At 12 month intervals, participants had a physical examination, complete blood count, erythrocyte sedimentation rate (ESR), and malaria smear. Ongoing outpatient clinical care and treatment was provided free of charge at the project site between scheduled study visits, and the study pharmacy maintained medications recommended by the WHO essential drug list. Inpatient and specialty care was provided at the Centre Hospitalier de Kigali (the only major community hospital in Kigali). If an individual died, a certified nurse would administer a verbal autopsy to the surviving spouse or relatives to ascertain the cause of death. Data relevant to the cause of death were also abstracted from clinic charts, hospital records, and death certificates. If a woman did not return for her 6-monthly study visit, a community worker was sent to her address to determine the reason.

We used the modified Kigali combined (MKC) staging system to identify HIV-infected women with advanced clinical disease at baseline. MKC staging was developed and validated as a strong predictor of mortality in this cohort of HIV-infected Rwandan women ^[15]. MKC staging combines clinical staging with hematocrit and ESR results and has been described previously in detail ^[15]. Laboratory results are stratified as either normal or abnormal. Normal laboratory results are defined as an ESR of 65 mm/h or less and a hematocrit of 0.38 or greater ^[15]. Abnormal laboratory results are defined as an ESR greater than 65 mm/h or a hematocrit less than 0.38 ^[15]. MKC stages 1 and 2 disease correspond to asymptomatic and mild HIV disease, respectively, whereas MKC stages 3 and 4 disease correspond to moderate and advanced HIV disease.

In 1998, 96 surviving women had viral load (Amplicor 1.0 assay; Roche Diagnostic Systems, Inc., Branchburg, New Jersey, USA) and CD4 cell count testing (FACS; Becton Dickinson, Franklin Lakes, New Jersey, USA) performed. Since 2004, CD4 cell count testing has become routinely available.

Statistical analysis

The primary endpoint was non-genocide mortality. Confirmed genocide-related traumatic deaths were censored in the main analysis. Kaplan-Meier survival methods and log-rank tests were used to evaluate median survival. For Kaplan-Meier analysis, survival time was censored on 31 August 2003 for all surviving participants, because HAART became available on that date. Survival times were stratified by seroconversion status. Cox proportional hazards regression was employed (SAS version 9.1; SAS Institute, Cary, North Carolina, USA) to identify variables associated with mortality. We described the hazard function of mortality over time by plotting smoothed hazard curves (according to the Kernel function) with bandwidths of 2 years ^[16]. We also developed a prognostic survival model using maximum likelihood methods to fit a Weibull model to the observed mortality data over the first 9 years of follow-up. We evaluated whether the Weibull model could accurately forecast the observed survival rates to mid-2003 (before HAART was introduced).

Ethics

Study procedures were approved by the Rwanda National Ethics Committee in Kigali, Rwanda, and the Institutional Review Board of Emory University. Human research guidelines were followed, and informed consent was obtained from all participants.

Results

The median baseline age for the entire cohort was 27 years (IQR 24-30 years). HIV-seroincident women were older (P < 0.0001) than seroprevalent women but otherwise had similar demographics (Table 1). Seroprevalent women were more likely to have been previously hospitalized and to have a baseline body mass index less than 19 kg/m² (Table 1). At the time of enrollment, 71% of seroprevalent women (n = 283) had MKC stage 1 or 2 disease, 24% (n = 98) had MKC stage 3 disease, and 5% (n = 21)had MKC stage 4 disease.

Table 1

Over a maximum of 20 years of total follow-up, there were 329 (60% of total cohort) confirmed non-genocide deaths (Table 2). Remarkably, 109 women (20%) have survived and remain active in the cohort at February 2006. Of these 109 women, 57 have initiated HAART. The remaining 52 women have still not met the clinical criteria to initiate HAART and are being monitored closely for clinical or CD4 cell count evidence of disease progression. Excluding the genocide, 46 study participants (8%) were lost to follow-up. During the genocide, 31 women (6%) were confirmed to have been killed and an additional 33 women (6%) were lost. Seroincident women were more likely to have been killed in the genocide (P = 0.0003).

Table 2

Natural history data (censored in mid-2003 before HAART introduction)

Among all participants (n = 548), the median time from HIV diagnosis to death was 9.2 years (95% CI 8.7-9.6 years) when genocide deaths were censored. Among women 25-29 years old, the median time from HIV diagnosis to death was 9.3 years (95% CI 8.3-10.6 years). When the cohort was stratified by seroconversion status, seroincident women had a median time from seroconversion to death of 11.9 years (95% CI 8.9-13.5 years) compared with 8.9 years (95% CI 8.3-9.4 years) for seroprevalent women with unknown dates of seroconversion (P = 0.005, log rank test; Fig. 1). After 15 years of follow-up (before the introduction of HAART), the survival probability was still 36% (95% CI 26-46%) for seroincident women and 26% (95% CI 21-31%) for seroprevalent women. We have previously reported associations between baseline clinical and laboratory factors such as a body mass index of 19 or less, hematocrit less than 38%, and ESR greater than 65 mm/h, with mortality in this cohort ^[15]. Although there was a trend towards more rapid progression to death with older age, this association did not reach statistical significance. Women greater than 35 years of age had a hazard ratio of 1.46 (95% CI 0.8-2.6) and women aged 30-34 years had a hazard ratio of 1.14 (95% CI 0.9-1.5) for death compared with women 18-25 years of age. When known genocide deaths were not censored, the median time from HIV diagnosis to death for seroincident participants decreased to 9.0 years (95% CI 7.6-11.9 years) and for seroprevalent participants it decreased to 8.6 years (95% CI 7.3-9.2 years; P = 0.08, log rank test).

Fig. 1

Symbol. No caption a...

Symbol. No caption a...

Over a total of 4647 HIV-infected person-years of follow-up, mortality rates (non-genocide deaths per HIV-infected person-year of follow-up) varied over time (Fig. 2). In general, the smoothed hazard function increased reflecting an accelerating mortality rate with increased duration of HIV infection. The overall smoothed mortality rate peaked at 0.12 deaths per HIV-infected person-year at 9.5 years of follow-up. Seroincident participants peaked at a smoothed mortality rate of 0.11 deaths per person-year and seroprevalent participants peaked at a smoothed mortality rate of 0.14 deaths per person-year (Fig. 2). The hazard function, however, levelled off and began to decrease at later durations of follow-up for both seroincident and seroprevalent participants. A Weibull model generated using only the first 9 years of survival data (when the mortality rate was steadily increasing) started to diverge from the observed survival rates at later timepoints (> 14 years). Although this Weibull model accurately predicted a time to 60% survival of 11.7 years (95% CI 10.6-13.0) compared with an observed survival time of 11.5 years, the Weibull model's prediction of a time to 70% survival of 14.0 years (95% CI 12.5-15.7) was an underestimation of the observed 70% survival time of 15.1 years.

Fig. 2

Symbol. No caption a...

Symbol. No caption a...

Data including HAART era (censored in February 2006)

Mortality rates stratified by significant historical events revealed a similar pattern. In the 6 years before the genocide in 1994, there were 0.05 deaths per HIV-infected person-year. During the genocide the mortality rate spiked to 0.31 deaths per person-year (these deaths are censored in Fig. 1 and Fig. 2). In the 5 years immediately after the genocide, when access to medical care was still compromised, the mortality rate was 0.10 per person-year. From 2000 to mid-2003, when access to medical care had improved, the mortality rate averaged 0.08 per person-year. Since the introduction of HAART in mid-2003 the mortality rate has decreased to 0.02 per person-year, which was a statistically significant decline (hazard ratio 0.21; 95% CI 0.07-0.62) compared with the 2000-mid-2003 time period, P = 0.0039 by log-rank test.

In the subset of women (n = 96) who had viral load and CD4 cell count testing in 1998, both viral load and CD4 cell count results were associated with the subsequent clinical outcome (Table 3). Women who died in follow-up had the highest median viral loads and lowest median CD4 cell counts in 1998. Women who have survived to 2006 but have required HAART initiation had intermediate viral load and CD4 cell count results. Women who still have not required HAART by 2006 had the lowest median viral loads and the highest median CD4 cell counts in 1998.

Table 3

Although a high rate of long-term survival was observed in this cohort, most women still had immunological evidence of disease progression. Of the 109 surviving women, 69 women (63%) had CD4 cell count data from 1998 and 2003. Their median decline in CD4 cell count was 31 cells per year (IQR 16-57 cells/year). Only 17 women (16% of all survivors and 3% of the original cohort) continually have CD4 cell counts persistently of 500 cells/μl or greater without antiretroviral therapy. Viral load testing has been performed on 11 of these women, of whom eight (73%) had a viral load of 2000 copies/ml or less (range < 400-7527 copies/ml) and only two women had undetectable viral loads.

Discussion

The median survival time in this cohort of HIV-infected Rwandan women was similar to survival times observed in high-income countries. Interestingly, we observed higher rates of long-term survival than would be predicted by current natural history models of HIV-1 infection ^[5]. A levelling off of mortality rates during the late stages of follow-up may explain this high rate of long-term survival. The majority of these long-term survivors still had evidence of disease progression, however, as demonstrated by CD4 cell count declines.

The most striking observation from this prospective cohort was the high rate of long-term survivors. Although HIV-1 mortality rates generally increase with the duration of infection, we actually observed a flattening and decrease of mortality rates during the late stages of follow-up (before HAART had been introduced). Cohorts from high-income countries have also observed mortality rates that have levelled off over time, suggesting the presence of a subgroup that is relatively resistant to HIV-1 disease progression ^[17]. Our Weibull model, which assumed a monotonically increasing mortality rate ^[5], confirmed the significance of this late decrease in the mortality rate on overall survival times.

We can only speculate as to the factors that resulted in the high rate of long-term survivors in this Rwandan cohort. An interesting possibility is the effect of the viral subtype on disease progression. Although still controversial, there is now evidence from Senegal ^[18], Uganda ^[19], Tanzania ^[20], and Kenya ^[21] that individuals infected with subtype A viruses may have slower HIV-1 progression compared with other subtypes. The biological basis for these differences in disease progression is not understood. Other viral factors can also contribute to disease progression. For example, an unusual Vpr deletion in the N-terminal domain was detected in the HIV-1 virus of one Rwandan long-term non-progressor and needs to be investigated further ^[22]. Several human gene alleles such as HLA B*57 ^[23] and CCR5 Δ32 have been associated with slower HIV-1 progression. A higher rate of HLA B*57 has been detected in long-term survivors in this cohort compared with the entire cohort ^[24]. CCR5 Δ32 has not been found in Rwanda ^[25], but a recent study identified a novel 24 base pair deletion in the coding region of CCR5 (hCCR5Δ24) in four Rwandan individuals, although it is unknown whether this mutation modulates HIV-1 disease progression ^[26]. Immunological factors should also be considered but have not been evaluated in this cohort. The frequency of long-term survival in our analysis suggests that these viral, genetic, and immunological factors that favor slower HIV-1 disease progression may be more common than anticipated.

True long-term non-progressors are strictly defined as being asymptomatic for more than 15 years with a CD4 cell count greater than 500 cells/μl and an undetectable viral load without antiretroviral therapy ^[27]. The proportion of individuals with true long-term non-progression in this cohort was similar to rates described in other populations (0.8%) ^[27], and was not large enough to explain the observed high rate of long-term survival. Most long-term survivors had evidence of disease progression, as demonstrated by significant CD4 cell count declines over time, and should technically be classified as slow or delayed progressors ^[28]. The biological basis for slow or delayed progression, however, may be mediated by the same factors that favor long-term non-progression.

We observed similar median survival times in HIV-infected women in Rwanda (a median of 11.9 years from seroconversion to death) as reported in high-income countries ^[2], in rural Uganda ^[9], and among South African miners ^[10]. These survival data suggest that HIV-1 does not progress more rapidly in Africa than in high-income countries. Shorter median survival times from seroconversion have been estimated in Kenya ^[8], but the vulnerability of that cohort of female sex workers to other sexually transmitted infections and HIV-1 super-infections could explain the difference. Unexpectedly, the median survival of seroprevalent women was only 3 years less than the median survival of seroincident women, suggesting that the seroprevalent women had been identified at an early stage in their infection. Women who are still healthy enough to get pregnant are probably at an early stage of their HIV-1 infection. Although HIV-1 had spread extensively in Kigali by 1986 ^[29], it is also possible that the epidemic was still in an expansion phase among participants recruited to this study.

The age difference between seroincident and seroprevalent women was surprising, but may be attributable to our closed study design. In a community-based sample seroincident women would be expected to be younger than seroprevalent women because by definition they are at an earlier stage of their HIV-1 infection. At the time of our study enrollment, however, the median age of HIV-infected women was lower than that of HIV-negative women ^[11]. This age difference was caused by a greater seroprevalence among younger women (aged 19-24 years) in Rwanda at that time ^[11]. Since our cohort was closed, the population of women who could seroconvert was, therefore, already older than the seroprevalent women. In addition, HIV-infection may decrease fertility ^[30,31] especially at older ages ^[32]. Therefore older HIV-infected women may be less likely to get pregnant compared to the general population which could result in a different balance of ages (fewer older HIV-infected women) in the antenatal clinic than the general population.

Our prospective cohort study had several limitations. Overall, our cohort contained a relatively homogenous population with respect to age, sex, residence, and ethnicity. In a homogeneous population it is possible that a genetic or viral factor favoring long-term survival could be overrepresented, which would limit the generalizability of our findings. Censoring known traumatic genocide deaths was a necessary limitation, but could bias our results if genocide victims were more likely to have advanced HIV-1 disease. Seroincident women, who by definition are earlier in the course of their disease than seroprevalent women, were more likely to be genocide victims, however, suggesting that women with advanced HIV-1 disease were not necessarily more likely to be killed in the genocide. Another limitation is that CD4 cell count testing only became routinely available in 2004 (around the time of HAART introduction) in our cohort. Although limited CD4 cell count data were available from 1998 in a subset of participants, more intensive CD4 cell count monitoring would have provided more insight into the rate of disease progression. Finally, women in our cohort had more access to medical care and treatment than the general population in Africa, which could impact observed survival times.

In conclusion, the median survival time in this cohort of HIV-infected Rwandan women was similar to survival times observed in high-income countries, and we observed greater rates of long-term survival than would be predicted by current natural history models of HIV-1 infection in Africa. A levelling off of mortality rates during the late stages of follow-up may explain this high rate of long-term survival.

Acknowledgements

The investigators would like to thank all of the women in Rwanda who participated in this study, all the staff at Projet San Francisco in Kigali who made this study possible, and grant funding from NIAID, NICHD, NIMH, and the International AIDS Vaccine Initiative. The investigators would also like to thank Mary Ann Hall for editorial assistance with this paper.

Sponsorship: This study was funded by grants NIAID R01 23980, R01 40951, R01 51231, NICHD R01 40125, NIMH R01 66767, FIC D43 TW001042, and the International AIDS Vaccine Initiative.

Conflicts of interest: None.