Participants provided separate informed oral consent for interviews, blood draws and blood storage. Survey protocols were approved by the Ethics Review Committee of the Kenya Medical Research Institute (KEMRI) and the Institutional Review Board of the U.S. Centers for Disease Control and Prevention (CDC).

The Kenya AIDS Indicator Survey (KAIS) was a nationally representative, population-based household survey involving persons aged 15–64 years. The design for KAIS was a stratified, two-stage cluster sample that was for comparable with the design of the 2003 Kenya Demographic and Health Survey (DHS). The first stage involved selecting clusters from the same sampling frame that was used for the 2003 DHS, based on the 1999 national census, and the second stage involved the selection of households per cluster with equal probability of selection in the rural-urban strata within each district. A total of 29 field teams, each consisting of six data collectors (four interviewers and two laboratory technicians), one supervisor and one driver, conducted fieldwork from August to December 2007. Teams administered questionnaires in local languages where necessary, in addition to questionnaires in Kiswahili and English, to accommodate respondents that were not conversant in vernacular languages. All questionnaires were back translated into English. Survey personnel participated in intensive two week training in KAIS procedures. All questionnaire responses were double-entered using Census and Survey Processing System (CSPro) version 3.3 (U.S. Census Bureau. Washington DC, USA). A series of internal consistency and range checks helped to identify any illogical responses and to verify that responses adhered to skip patterns in the questionnaire. Demographic and HIV/AIDS-specific indicators included HIV testing history, self-reported HIV status, and partner-specific behaviors and disclosure. Blood specimens were obtained in households for testing at the national reference laboratory in Nairobi for HIV, HSV-2 and syphilis. Individual test results were linked to individual and household questionnaires through unique identification numbers.

Vironostika HIV Uni-Form II antigen/antibody (BioMérieux Bv, Boseind, Netherlands) and Murex HIV antigen/antibody (Abbott/Murex-Biotech Ltd, Kent, UK) tests were used for HIV screening and confirmation, respectively, in a serial testing algorithm. Discordant samples were retested with the two assays and polymerase Chain Reaction (PCR) (Roche HIV DNA v 1.5) tests were conducted on all samples with two sets of discordant results. For quality control, all positive specimens and a random sample of 5% of negative specimens was retested in a different laboratory using the same testing algorithm. The Kalon HSV Type 2- specific IgG EIA was used for HSV-2 testing. This was a recombinant type 2 antigen (gG2) modified to eliminate reactivity arising from HSV type 1 infection and at the same time retaining the natural antigenic characteristics of HSV-2. For syphilis testing, the Treponema pallidum particle agglutination (TPPA) assay was used as a screening test and rapid plasma reagin (RPR) for confirmation at a dilution of 1∶1.

Couples were defined as being in a married or cohabitating partnership between the male head of household and his wife. In case of a polygamous partnership the husband and only one partner were included (n = 298 couples). A couple was considered HIV-discordant if one partner was HIV-infected and the other partner HIV-uninfected in HIV testing done during the study. A couple was considered to be concordantly uninfected if neither partner was HIV-infected and considered concordantly HIV-infected if both partners were HIV-infected. Couples were considered HIV-affected when they were either HIV-discordant or -concordant. We developed 2 different models to assess factors associated with couple status – one for HIV discordance as the outcome and a second one with HIV concordance as the binary outcome. In the first model, concordant HIV-infected couples were excluded from the denominator and in the second model concordant uninfected couples were excluded from the denominator. Predictor variables included socio-demographic characteristics, age difference between partners, length of relationship in years, male circumcision, syphilis, HSV-2, CD4 cell count, perception of HIV risk, outside partners, consistent condom use with outside partner in the last 12 months, number of lifetime sexual partners, and desire to have a child within next two years among women. All predictor variables were self-reported except for CD4 cell count of infected partners, HSV-2 and syphilis status. Male circumcision was based on self-report as physical examinations were not conducted. Correct knowledge of HIV status was defined as self-reported HIV status validated by laboratory testing during the study. Low perception of HIV risk was defined as small or no self-reported risk for HIV.

We performed all analyses in SAS version 9.1 (SAS Institute Inc., Cary, North Carolina, USA). Using multivariable logistic regression (SAS Survey Logistic), we calculated adjusted odds ratios (AOR) and 95% confidence intervals (CI) to identify variables associated independently with each primary outcome. Variables that were statistically significant at a 0.05 p-value level in bivariate analyses were selected for the final multivariable model by use of backwards elimination if they did not remain significant. Two way interactions between variables were considered. Confidence intervals for percentages were computed. Logistic regression was used to assess whether the probability of having a relationship of 20 years or more was related to knowledge of status and disclosure. Analyses accounted for the stratified cluster design of the survey. Each response was weighted to account for its sampling probability and to adjust for non-response rates. National estimates of numbers of concordant positive and discordant positive couples were derived by multiplying the estimated national population aged 15–64 years (19,319,000) by the estimated proportion of this group that was married/cohabitating based on survey results (0.604), then dividing by 2 to get the number of couples and finally multiplying that result by the estimated proportion of couples that were discordant (0.058) or concordant positive (0.038). Correct knowledge and disclosure of HIV status were not included in the models because of uncertainty of the cause-effect relationship in a cross-sectional study and because of our inability to construct a variable by HIV status to measure the prevention effect in the infected discordant partner. We also decided not to include condom use between married or cohabitating partners in the models because we were not able to adjust for correct knowledge and disclosure of HIV infection status. Widow-hood was not included in models because this information was not available for men and numbers were too small for women.

A total of 9,049 women and 6,804 men aged 15–64 years living in 9,691 households from 402 clusters participated in both interviews and blood sample collection. This included 2,748 married or cohabitating couples. The response rate for couples was 84.4% (Figure 1). Overall, 9.6% (95% CI: 8.3–10.9) of couples were affected by HIV; 3.8% (95% CI: 3.0–4.5) were concordantly infected and 5.8% discordantly infected (95% CI: 4.7–6.8). In 48.5% (95% CI: 40.3–56.5) of discordant couples the woman was the HIV-infected partner. Extrapolating to population estimates, this corresponds to an estimated 219,000 concordant infected couples and 338,000 discordant couples in Kenya. Of all couples, 30.4% (95% CI: 28.3–32.5) were concordantly positive for HSV-2 and 20.7% (95% CI: 18.9–22.5) were HSV-2 discordant.

Among all married or cohabitating HIV-infected Kenyans, 42.8% (95% CI: 36.5–49.0) had a previous HIV test and 16.4% (95% CI: 11.5–20.8) knew their HIV status correctly. Among the 83.6% HIV-infected Kenyans living in a couple who did not know their status correctly, 24.8% (95% CI: 20.1–29.6) self-reported that they were HIV-uninfected based on their most recent test results, and 57.2% (95% CI: 51.0–63.5) never had an HIV test or received their test result. Only 14.9% (95% CI: 10.4–19.4%) of HIV-infected Kenyans living in a couple correctly knew their status and also reported that they disclosed to their partner; however, this meant that most disclosed when they knew their status. Among discordant couples, in 5.5% (95% CI: 1.7–9.3) both partners correctly knew their status and disclosed their status to the partner. In 31.8% of discordant couples the infected partner reported being HIV-uninfected based on a prior test. In infected concordant couples, 16.2% (95% CI: 10.0–22.4) correctly knew their status and disclosed their status to the partner. In 34.7% (95% CI: 26.5–42.9) of concordant couples at least one partner had a prior HIV-negative test and was unaware of his or her current HIV infection. Partnerships of 20 years or more duration were reported by 38.3% (95% CI: 29.3–47.3) of discordant and 28.4% (95% CI: 18.6–38.2) of concordant couples. Among those, rates of correct knowledge of status and disclosure were not significantly different than among couples with relationships of less than 20 years. In 3.2% uninfected (95% CI: 2.4–5.0), 6.6% discordant (95% CI: 2.5–10.6) and 16.3% concordant (95% CI: 9.2–23.5) infected couples one or both partners reported consistent condom use in the last 12 months.

Table 1 shows characteristics of married or cohabitating couples for predictor variables included in bivariate analyses. Proportions of HIV-affected couples were highest in the 15–24 years age categories for both female (4.1%, 95% CI: 1.9%–6.4%) and male (4.3%, 95% CI: 1.8%–6.9%) HIV-infected partners. Furthermore, the proportions of HIV-affected couples were highest in Nyanza (21.8%, 95% CI: 17.3%–26.4%) and lowest in North Eastern (2.8%, 95% CI: 0%–5.8%) (Table 1), the provinces with the highest and lowest HIV prevalence rates (14.9% and 0.8%, respectively [19]).

Table 2 shows results of bivariate and multivariable analyses for variables associated with HIV discordance compared to HIV uninfected concordance. In final multivariable models, predictors for HIV-discordance included younger age in women (AOR 1.5, 95% CI: 1.2–1.8 per 10 years; p<0.0001), increasing number of lifetime sexual partners in women (AOR 1.5, 95% CI: 1.3–1.8; p<0.0001), HSV-2 infection in either (man infected: AOR 3.1, 95% CI: 1.6–6.3; p = 0.001; woman infected: AOR 3.5, 95% CI: 1.9–6.4; p<0.0001) or both partners (AOR 4.1, 95% CI: 2.3–7.2; p<0.0001), and lack of male circumcision (AOR 1.6, 95% CI: 1.0–2.5; p = 0.032) (Table 2).

Table 3 shows results of bivariate and multivariable analyses for variables associated with HIV concordance compared to HIV discordance. In final multivariable models, independent factors for HIV-concordance included HSV-2 infection in both partners (AOR 6.5, 95% CI: 2.3–18.7; p = 0.001) and lack of male circumcision (AOR 1.8, 95% CI: 1.0–3.3; p = 0.043) (Table 3).

No significant interactions were found in the models. Also, in both models there were too few people in partnerships with CD4 cell counts to use it as a proxy for length of infection or on ART to compute the effect of ART with any degree of confidence (e.g. there were 153 discordant couples with one partner on ART and available CD4 counts; of those, 24 had a CD4 count less than 250 cells/µl).