The authors have declared that no competing interests exist.
Conceived and designed the experiments: SB HAW. Performed the experiments: SB. Analyzed the data: SB HAW. Contributed reagents/materials/analysis tools: SB HAW. Wrote the paper: SB ER AK JW HG HAW. Acquisition of data, or analysed and interpreted data: SB ER AK JW HG HAW. Drafted the article or revised it critically for important intellectual content: SB ER AK JW HG HAW. Approved the version of the article to be published: SB ER AK JW HG HAW.
Early initiation of antiretroviral therapy reduces risk of transmission to the uninfected partner in HIV discordant couples, but there are relatively little observational data on HIV transmission within couples from non-trial settings. The aims of this paper are to estimate HIV incidence among HIV discordant couples using longstanding observational data from a rural Ugandan population and to identify factors associated with HIV transmission within couples, including the role of HSV-2 infection.
Using existing data collected at population-wide annual serological and behavioural surveys in a rural district in southwest Uganda between 1989 and 2007, HIV discordant partners were identified. Stored serum samples were tested for HSV-2 serostatus using the Kalon ELISA test. HIV seroconversion rates and factors association with HIV seroconversion were analysed using Poisson regression.
HIV status of both partners was known in 2465 couples and of these 259 (10.5%) were HIV serodiscordant. At enrolment, HSV-2 prevalence was 87.3% in HIV positive partners and 71.5% in HIV negative partners. Of the 259 discordant couples, 62 converted to HIV (seroconversion rate 7.11/100 PYAR, 95%CI; 5.54, 9.11) with the rate decreasing from 10.89 in 1990–1994 to 4.32 in 2005–2007. Factors independently associated with HIV seroconversion were female sex, non-Muslim religion, greater age difference (man older than woman by more than 15 years), higher viral load in the positive partner and earlier calendar period. HSV-2 was not independently associated with HIV acquisition (HR 1.62, 95%CI; 0.57, 4.55) or transmission (HR 0.61, 95%CI; 0.24, 1.57). No transmissions occurred in the 29 couples where the index partner was on ART during follow up (872 person-years on ART).
HIV negative partners in serodiscordant couples have a high incidence of HIV if the index partner is not on antiretroviral therapy and should be provided with interventions such as couple counselling, condoms and antiretroviral treatment.
In generalised HIV epidemics in sub-Saharan Africa, a substantial proportion of new HIV infections occur in cohabiting couples
About 50% of married or cohabiting HIV positive individuals in stable partnerships in East and Southern Africa are in an HIV-serodiscordant relationship
HIV serodiscordant couples are suitable for studying potential interventions for HIV prevention and have been studied in clinical trials evaluating HSV-2 suppressive treatment
The study population comprises approximately 20,000 residents of 25 neighbouring villages in southwestern Uganda. The community is stable and homogeneous, with most people from the Baganda tribe (73%), and 15% of Rwandese origin. Religious affiliation is mostly Christian, with a significant Muslim minority (28%). HIV prevalence is high (7.7% in 2005)
Details of the cohort and annual HIV serosurvey have been published previously
The present analyses use data collected at annual surveys between 1989 and 2007. At each survey, household members were assigned a code indicating their relationship to others in the household. In addition, each married participant was asked to name their spouse. These data allowed for retrospective linkage of partners who participated in the surveys as individuals. The study included couples where both spouses were aged 18 to 59 years.
HSV-2 status was determined using the Kalon ELISA assay
Statistical analyses were performed using Stata version 11.0 (StataCorp, Texas USA). Couples identified in the database were defined as concordant negative if both spouses were HIV negative, discordant if one spouse was HIV positive and the other was HIV negative, concordant positive if both were HIV positive, incomplete/unknown if the HIV status of one or both was not known. When an individual did not have an HIV test result, the HIV status was imputed as follows: a participant testing positive at one survey but missing a result on a subsequent survey at which they were present was imputed as positive, while one testing negative at a survey but missing a result on an earlier survey at which they were present was imputed as negative.
Missing HSV-2 data were imputed using the same method. In addition, to increase precision of the estimated HSV-2 prevalence and to reduce bias in analyses for the effect of HSV-2 on HIV seroconversion in discordant couples, missing HSV-2 status was further imputed using multiple imputation
HIV incidence rates and 95% confidence intervals (CI) were estimated using Poisson regression. Follow-up time started at the visit when couples were first seen as serodiscordant and ended either at estimated seroconversion date, or the date last seen (for couples who did not seroconvert). If the couple was not seen on the same date, the date for HIV test results from the HIV negative partner was used. The date of HIV seroconversion was estimated as the midpoint of the last negative and first positive test.
Factors associated with HIV incidence were estimated using Poisson regression with random-effects to allow for within-person clustering of men included with more than one partner. There was a-priori interest in potential interactions between age and sex of the susceptible partner and of the index partner, and these were examined using the Wald test. However, as there was no evidence of interaction, primary analyses were not stratified by sex or age. Separate models were used to analyse characteristics for the negative partner and the positive partner. A hierarchical conceptual framework approach was used
The study obtained ethics approval from the scientific and ethics review boards from the London School of Hygiene and Tropical Medicine, Uganda Virus Research Institute and the Uganda National Council for Science and Technology. The survey population was actively encouraged to test for HIV infection, using the freely available testing and counselling services, including couple counselling, and condom provision. The present study used existing data and therefore did not offer couple counselling. Beginning with 2004 when HAART became available in this population, eligible study participants were started on treatment according to the Uganda national guidelines.
From 1989 to 2007, a total of 4,480 couples were seen at least once, with a total of 22,782 visits at which both partners were seen. Of these 3,358 couples (74.9%) had known HIV status at one or more visits and 2,465 of these (73.5%) were seen twice or more (
Subsequent analyses focus on the 259 HIV serodiscordant couples (221 couples who were serodiscordant at their first visit and 38 couples who were initially concordant negative but became serodiscordant during follow up;
The male was the HIV positive partner in 140 (54.1%) of couples. The median age was 33 years (IQR: 27, 39) for males and 26 years (IQR: 21, 32) for females, and was similar by HIV status (p = 0.70). Approximately half (49%) of female HIV positive participants were in-migrants, compared with just 19% of male HIV positive participants (p<0.001). Male HIV negative participants were most likely to be Muslim (33%), compared with 15% of male HIV positive participants. Female HIV positive participants had the lowest levels of education (18% with no education), and male HIV positive participants had the highest levels (6% with no education).
Of the 259 HIV positive partners, 189 (73.0%) had known HSV-2 status at baseline. Of these, 156 were HSV-2 positive (HSV-2 prevalence = 83.7%; 95%CI: 81.7–85.7%). Among the 259 HIV negative partners, 209 (80.7%) had known HSV-2 status, and 148 were HSV-2 positive (HSV-2 prevalence = 71.5%; 95%CI: 69.0–73.9%). Among the 162 couples where both partners had known HSV-2 status at baseline, 13 (8.0%) were HSV-2 concordant negative, 97 (59.9%) were HSV-2 concordant positive, and 52 (32.1%) were HSV-2 discordant at baseline. Among the 52 who were HSV-2 discordant, the HIV index partner was HSV-2 positive in 65.4%. HSV-2 prevalence was higher in females than males among HIV negative (OR = 2.15, 95% CI: 1.52–3.04) and HIV positive participants (OR = 1.32, 95% CI: 0.82, 2.12).
HIV seroconversion occurred in 62/259 (23.9%) partners (incidence rate = 7.11/100 PYAR; 95%CI: 5.54–9.11). HIV incidence was two times higher in females than in males (HR = 2.02, 95% CI: 1.06–3.83;
HIV negative partner | HIV positive partner | |||||||
Cases/PYAR | HR |
p-value | Cases/PYAR | HR |
p-value | |||
|
||||||||
|
p = 0.03 | p = 0.03 | ||||||
|
19/382 | 4.97 [4.14, 5.98] | 1 | 43/490 | 8.77 [7.76, 9.90] | 1 | ||
|
43/490 | 8.77 [7.76, 9.90] | 2.02 [1.06, 3.83] | 19/382 | 4.97 [4.14, 5.98] | 0.49 [0.26, 0.94] | ||
|
p = 0.5 | p = 0.6 | ||||||
|
16/145 | 11.06 [9.06, 13.51] | 1 | 5/109 | 4.60 [3.22, 6.58] | 1 | ||
|
13/168 | 7.74 [6.19, 9.66] | 0.79 [0.35, 1.76] | 11/154 | 7.15 [5.62, 9.09] | 1.89 [0.60, 5.99 ] | ||
|
20/356 | 5.61 [4.69, 6.71] | 0.57 [0.27, 1.22] | 31/389 | 7.97 [6.91, 9.21] | 2.05 [0.73, 5.81] | ||
|
13/204 | 6.38 [5.11, 7.97] | 0.59 [0.26, 1.38] | 15/221 | 6.78 [5.51, 8.34] | 1.83 [0.59, 5.76] | ||
|
p = 0.19 | p = 0.4 | ||||||
|
57/757 | 7.53 [6.77, 8.37] | 1 | 56/765 | 7.32 [6.57, 8.14] | 1 | ||
|
5/115 | 4.33 [3.03, 6.21] | 0.52 [0.20, 1.37] | 6/107 | 5.60 [4.04, 7.77] | 0.69 [0.28, 1.68] | ||
|
p = 0.02 | p = 0.08 | ||||||
|
56/679 | 8.25 [7.41, 9.18] | 1 | 54/679 | 7.95 [7.13, 8.87] | 1 | ||
|
6/193 | 3.10 [2.24, 4.30] | 0.33 [0.13, 0.84] | 8/194 | 4.13 [3.11, 5.49] | 0.46 [0.19, 1.18] | ||
|
p = 0.6 | p = 0.9 | ||||||
|
8/80 | 9.94 [7.49, 13.19] | 1 | 6/77 | 7.75 [5.59, 10.75] | 1 | ||
|
44/652 | 6.75 [5.98, 7.61] | 0.69 [0.28, 1.71] | 43/585 | 7.35 [6.51, 8.31] | 1.07 [0.40, 2.83] | ||
|
10/140 | 7.15 [5.55, 9.20] | 0.59 [0.19, 1.83] | 13/210 | 6.18 [4.95, 7.72] | 0.89 [0.29, 2.75] | ||
|
p = 0.3 | p = 0.14 | ||||||
|
45/695 | 6.47 [5.75, 7.29] | 1 | 44/698 | 6.31 [5.59, 7.12] | 1 | ||
|
17/177 | 9.58 [7.89, 11.64] | 1.38 [0.70, 2.70] | 18/175 | 10.29 [8.52, 12.42] | 1.64 [0.84, 3.18] | ||
|
p = 0.6 | p = 0.3 | ||||||
|
58/794 | 7.30 [6.57, 8.11] | 1 | 56/735 | 7.62 [6.85, 8.48] | 1 | ||
|
4/78 | 5.12 [3.43, 7.64] | 0.73 [0.24, 2.19] | 6/137 | 4.37 [3.15, 6.05] | 0.55 [0.21, 1.44] | ||
|
p<0.001 | |||||||
|
62/843 | 7.35 [6.64, 8.14] | 1 | |||||
|
0/29 | 0 | 2.3e-13 [0,.] | |||||
|
p = 0.9 | |||||||
|
1/22 | 4.50 [2.022, 10.02] | 1 | |||||
|
3/41 | 7.36 [4.64, 11.69] | 1.78 [0.16, 19.75] | |||||
|
2/37 | 5.41 [3.08, 9.53] | 1.49 [0.11, 19.87] | |||||
|
5/72 | 6.96 [4.87, 9.96] | 1.56 [0.15, 15.73] | |||||
|
p = 0.11 | |||||||
|
1/59 | 1.71 [0.77, 3.80] | 1 | |||||
|
2/49 | 4.12 [2.34, 7.26] | 2.99 [0.23, 39.21] | |||||
|
5/39 | 12.74 [6.81, 18.22] | 9.79 [0.97, 98.34] | |||||
|
p = 0.13 | p = 0.6 | ||||||
|
4/107 | 4.08 [2.88, 5.76] | 1 | 5/43 | 9.55 [6.75, 13.51] | 1 | ||
|
42/612 | 7.53 [6.76, 8.40] | 2.25 [0.78, 6.44] | 41/668 | 6.84 [6.13, 7.62] | 0.74 [0.23, 2.39] | ||
|
||||||||
|
p = 0.6 | |||||||
|
52/709 | 7.43 [6.65, 8.30] | 1 | |||||
|
10/173 | 5.79 [4.49, 7.46] | 0.82 [0.37, 1.82] | |||||
|
p = 0.009 | |||||||
|
50/780 | 6.41 [5.72, 7.17] | 1 | |||||
|
12/92 | 13.00[10.32,16.39] | 3.32 [1.36, 8.15] | |||||
|
p = 0.15 | |||||||
|
25/229 | 10.89 [9.28, 12.78] | 1 | |||||
|
9/146 | 6.18 [4.73, 8.07] | 0.57 [0.26, 1.29] | |||||
|
19/289 | 6.57 [5.47, 7.89] | 0.66 [0.34, 1.29] | |||||
|
9/208 | 4.32 [3.31, 5.64] | 0.40 [0.18, 0.92] |
Based on 57 couples where positive partner had at least one CD4 count result.
Based on 52 couples where positive partner had at least one Viral load result.
Account for clustering for polygamy.
Seroconversion rates calculated from imputed data, but actual number of cases and PYAR presented.
HIV incidence was lowest when the HIV negative partner was Muslim rather than Christian (RR = 0.33, 95%CI 0.13–0.84), and was higher in couples where the man was more than 15 years older than the woman (RR = 3.32, 95%CI 1.36–8.15;
HIV viral load was associated with increased risk of transmission (RR = 9.79, 95%CI 0.97–98.34 for those with viral load >50,000 copies/mL versus <10,000 copies/mL;
HSV-2 infection status of the negative partner increased the risk of HIV seroconversion two-fold, although this was not statistically significant (RR = 2.25, 95%CI: 0.78–6.44;
In multivariable analyses of risk factors for HIV acquisition in the HIV negative partner, HIV incidence was independently associated with female sex and non-Muslim religion (
HIV negative partner |
HIV positive partner |
|||
|
p = 0.03 | p = 0.03 | ||
|
1 | 1 | ||
|
1.83 [1.06, 3.18] | 0.55 [0.32, 0.95] | ||
|
p = 0.001 | p = 0.006 | ||
|
1 | 1 | ||
|
0.27 [0.11, 0.68] | 0.36 [0.16, 0.80] | ||
|
p = 0.25 | |||
|
1 | |||
|
2.18 [0.19, 24.55] | |||
|
5.85 [0.68, 50.60] | |||
|
p = 0.3 | p = 0.4 | ||
|
1 | 1 | ||
|
1.62 [0.57, 4.55] | 0.61 [0.24, 1.57] | ||
|
p<0.001 | p = 0.04 | ||
|
1 | 1 | ||
|
3.68 [1.83, 7.4] | 3.13 [1.55, 6.31] | ||
|
p = 0.09 | p = 0.06 | ||
|
1 | 1 | ||
|
0.56 [0.26, 1.23] | 0.57 [0.26, 1.25] | ||
|
0.62 [0.34, 1.15] | 0.62 [0.33, 1.15] | ||
|
0.39 [0.18, 0.86] | 0.37 [0.17, 0.80] |
Adjusted by all other factors in the model.
There are relatively few long-term observational studies of HIV discordant couples in the pre-ART era. Strengths of this study include the availability of data since 1990, before many interventions became widely available to the population, and the observational study design rather than a trial population, which is likely to be generalisable. This study found a high HIV incidence rate in HIV serodiscordant couples (7.11/100 PYAR), and incidence was twice as high in females as in males.
HIV negative partners in steady HIV serodiscordant partnerships are at high risk for HIV acquisition if the HIV positive partner is not on ART. The HIV rate in this study is comparable to those reported from HIV serodiscordant couples elsewhere in sub-Saharan Africa (range 4–10/100 PYAR)
The main factors associated with HIV transmission within a couple were a male index partner, non-Muslim couple, high viral load in the index partner, and a greater age difference between spouses. The median age at HIV seroconversion was substantially higher in men (40 years) than in women (28 years), and this is likely to partly reflect the fact that in this population men tend to be older than their female partners. However, this difference was also observed in the general population from which the cohort came, in which median age at seroconversion was higher in men
HSV-2 infection in the HIV negative partner was associated with a doubling of the rate of seroconversion (although the confidence intervals were wide). This magnitude of association is consistent with a meta-analysis of 25 cohort studies in which prevalent HSV-2 increased the risk of HIV acquisition three-fold (adjusted RR 2.8 (95% CI 2.1–3.7) in men and 3.4 (95% CI 2.4–4.8) in women)
HSV-2 infection in the positive partner was associated with a slightly lower rate of HIV transmission to the HIV negative partner (adjusted HR 0.61 [95% CI: 0.24, 1.57], p = 0.4). The result was unexpected as HSV-2 is thought to increase the infectiousness of HIV in co-infected persons. Previous observational studies of the association of HSV-2 with HIV incidence in discordant couples have also been inconclusive, with one study
Over half the index partners in this study were male (54%). This is in contrast to the Partners in Prevention study from 7 eastern and southern African countries, in which 33% of index partners were male
The rate of HIV transmission was higher when the man was older by more than 15 years, especially if the HIV negative partner was female. Similar results were seen in a longitudinal study in Zimbabwe that reported increased vulnerability to HIV in young women who have sexual relationships with older, and usually high risk men
Muslims were at a significantly lower risk of HIV acquisition in our study (aHR = 0.27, 95%CI 0.11–0.68) presumably because of the almost universal practice of male circumcision among Muslims in this population. There is little evidence that male circumcision directly reduces risk of male to female HIV transmission
Rates of HIV seroconversion reduced over time. During the early period (1990–1994), there was low awareness of one’s own, or partners’, HIV status. Counselling advice e.g. for the use of condoms in the context of serodiscordant partnerships was also not widely available
This study had a number of limitations. Firstly it was not ascertained whether HIV seroconversions occurred as a result of transmission from within the partnership or from an external partner. Genetic sequencing of couples’ virus has found up to 30% non-matching virus indicating infection acquired outside of the partnership
HIV negative partners in serodiscordant couples have a high incidence of HIV infection if the index partner is not on anti-retroviral therapy. Before these become available, there should be continued emphasis on couples counselling and testing (for example within the programmes of increased voluntary medical male circumcision scale-up), and HIV serodiscordant couples should be strongly advised to use the existing interventions to minimise risk of HIV transmission.