Risk Factors for Recent HIV Infections among Adults in 14 Countries in Africa Identified by Population-Based HIV Impact Assessment Surveys, 2015–2019

Study Design

PHIAs are nationally representative, cross-sectional, population-based surveys of households across each country (16,17). A 2-stage, stratified cluster sample design was used in each survey: enumeration areas were selected within strata (subnational units, such as regions) by using a probability proportional to size method, and households within enumeration areas were randomly selected in the second stage. Weights were calculated to account for unequal probability of household selection, nonresponse, and noncoverage. Within selected households, the household head completed a household survey, and eligible household members completed individual interviews and had blood collected after providing consent for each survey component.

Study Population

We used data from PHIA surveys completed in 14 countries during 2015–2019: Cameroon, Cote d’Ivoire, Eswatini, Ethiopia, Kenya, Lesotho, Malawi, Namibia, Nigeria, Rwanda, Tanzania, Uganda, Zambia, and Zimbabwe. We pooled data across PHIAs in a multicountry analysis because of small sample sizes for recent infections within each country. We included adults 15–49 years of age who reported engaging in sexual activity during the year before their interview. We only included persons who consented to a blood draw and had a valid final RITA classification.

Variable Definitions

We explored demographic and behavioral variables collected during participant interviews to identify potential risk factors. Demographic factors were country, sex, age, marital status, education, and household wealth. Behavioral factors were number of recent sexual partners (during the previous 12 months), age of sexual debut, HIV status of partner(s), age of partner(s), condom usage, and voluntary medical male circumcision status. Age of sexual debut was divided into <18 or >18 years categories; 18 years of age was the median. The numbers of partners in the previous year were grouped into categories (0, 1, or >2 partners), consistent with previous literature and an examination of the data (12,13). Age groups were 15–24, 25–34, and 35–49 years, according to published precedent.

We calculated community viremia levels within each stratum in each country. We defined participants with long-term HIV infections and detectable viral load of >1,000 copies/mL as viremic and all HIV-negative or HIV-positive participants with an undetectable viral load as nonviremic. We excluded persons with recent HIV infection (as defined in the next paragraph) from community viremia calculations, which were calculated as the weighted number of viremic persons divided by the weighted number of viremic plus nonviremic persons within each stratum. We then categorized each stratum into quantiles representing the percentages of persons with nonsuppressed HIV infection within the stratum.

The 3 primary outcome categories for each participant were recent HIV infection, long-term HIV infection, or HIV negative. All participants were tested for HIV in the household according to each country’s national testing algorithm. Confirmatory HIV testing was completed in all countries except Uganda by using the Geenius HIV-1/2 rapid test (Bio-Rad Laboratories, https://www.bio-rad.com). We classified HIV according to confirmatory testing and excluded a small number of participants (n<25) who tested positive for HIV-2 but not HIV-1 from recency testing because the LAg-Avidity EIA is meant for HIV-1 recency classification only. Among HIV-positive participants, RITA was used to distinguish recent from long-term infections. The first step of RITA used the LAg-Avidity EIA (Sedia Biosciences Corp., https://www.sediabio.com, for plasma specimens or Maxim Biomedical, https://www.maximbio.com, for dried blood spot specimens), which assesses development of antibody avidity. We classified participants with a median normalized optical density <1.5 for plasma samples (or <1.0 for dried blood spot samples where venous blood could not be collected [1,000 copies/mL, and an absence of antiretroviral drug metabolites in their blood by using RITA (14,15). All HIV-1–positive participants who did not meet the criteria for recent infection were categorized as having long-term HIV infection.

Statistical Analysis

Initial data analysis strategies were bivariate comparisons between HIV-negative participants and those who had a recent HIV infection or long-term HIV infections. We calculated an overall χ² test statistic and unadjusted odds ratios for categorical variables. We used Taylor series weights and variables representing strata and units for all analyses. We performed analyses by using SAS version 9.4 (SAS Institute, https://www.sas.com) and considered p values <0.05 statistically significant.

Our model-building strategy incorporated candidate exposure variables with bivariate p values <0.20. In cases where variables were colinear, we included only 1 variable or set of variables in the final model. Only variables collected consistently across countries were eligible for the multivariable model. We used multinomial logistic regression to calculate adjusted weighted odds ratios accounting for the complex survey sample design.

Ethics Statement

PHIA surveys were funded by the US President’s Emergency Plan for AIDS Relief (PEPFAR) (18); technical assistance was provided by the US Centers for Disease Control and Prevention. The surveys were conducted through cooperative agreements with grantees/federal entities, including country Ministries of Health, ICAP at Columbia University (New York, New York, USA), and the University of Maryland (Baltimore, MD, USA). Each survey was approved by human subject institutional review boards specific for each country, cooperative agreement grantees/federal entities conducting the survey, or the US Centers for Disease Control and Prevention.

Population Description

Across the 14 countries included in this analysis, we identified 16,831 HIV-positive and 241,909 HIV-negative PHIA participants (Figure). Of the 16,831 HIV-positive participants, 264 (1.6% of all HIV-positive participants) had recent infections and 16,567 had long-term infections. Sample sizes of participants who met inclusion criteria ranged from 5,874 in Eswatini to 95,463 in Nigeria (Table 1). Biomarker (blood draw) response rates ranged from 86.7% in Malawi to 99.0% in Uganda for female participants and 85.3% in Namibia to 98.5% in Uganda for male participants.

Bivariate Analysis Findings

We performed weighted bivariate comparisons between HIV-negative and recently infected participants (Table 2). Region was significantly associated with recent HIV infection; persons living in western Africa represented only 27.1% of persons with recent HIV infections but represented 51.5% of those who were HIV negative. In addition, female sex, age 25–34 years, and divorced or separated marital status were associated with recent HIV infection. Among young adults 15–24 years of age, female participants accounted for 72.1% of recent infections. Demographic variables not associated with recent infection were urban and rural locations, household wealth, and working during the previous 12 months.

Behavioral factors were also associated with the prevalence of recent infections. For example, compared with HIV-negative participants, those who had recent HIV infections were more likely to have had a sexual debut at 1 partner in the previous 12 months, had sex with a partner with whom they were not living, not used condoms during their last sexual intercourse with a nonregular partner, and had sex within the previous 12 months with a partner who had unknown HIV status or an HIV-positive status (Table 2).

We also performed weighted bivariate comparisons of recently infected participants and those with long-term infections (Table 2). Compared with participants 35–49 years of age, adolescents and young adults (15–34 years of age) had a greater percentage of recent infections than long-term infections. Participants who were never married or were divorced/separated had a greater percentage of recent than long-term infections, whereas those who were widowed were more likely to have long-term infections. Sex, employment, education, household wealth, and urbanicity were not associated with recent versus long-term infections. Behavioral factors more common in participants who had recent HIV infections (compared with those with long-term HIV infections) were sexual debut at 1 sexual partner in the previous 12 months, having partners other than a spouse or live-in partner in the previous 12 months, not using a condom during their last sexual intercourse, and having a partner in the previous 12 months who had unknown HIV status (Table 2).

We performed bivariate exploratory comparisons of variables that were not collected consistently across all PHIAs or were only relevant to population subgroups (Table 3). Participants who had a sexually transmitted disease diagnosis or sexually transmitted infection symptoms and male participants who did not have a medical circumcision were more likely to be recently infected. In addition, participants who had long-term infections were more likely to have previously been tested for HIV >1 year ago and were more likely to be uncircumcised than those who had a recent HIV infection (Table 3).

Multivariate Analysis Findings

In the adjusted model, the southern, southeastern, and eastern Africa regions and community-level viremia remained significantly associated with recent infections (Table 4). Participants living in countries within eastern, southeastern, and southern Africa had higher odds of recent HIV infection compared with those in western Africa, even after adjusting for community-level viremia. In addition, participants grouped in the third and fourth highest quartiles of community-level viremia were more likely to have a recent HIV infection than those in the lowest quartile; the odds of recent infection increased with each viremia quartile.

Individual demographic characteristics, including sex and marital status, remained significantly associated with HIV acquisition; female participants had 1.82 times greater odds of recent HIV infection than for male participants. In addition, participants who were divorced, separated, or widowed had 3.58 times greater odds of recent HIV infection than for those who were never married. Age group was not significantly associated with HIV acquisition risk in adjusted models (Table 4). Sexual behavior characteristics that remained significantly associated with recent HIV acquisition were having >1 partner in the previous 12 months (adjusted odds ratio [aOR] 1.92) and having >1 partner who thought or was told they were HIV-positive (aOR 7.25) or had unknown HIV status (aOR 2.05).

In the multivariable model, those 15–24 years of age (aOR 4.79) and 25–34 years of age (aOR 2.23) were more likely to have a recently acquired infection than were participants who were 35–49 years of age (Table 5). Participants who used a condom the last time they had sex (aOR 0.46) and had >1 partner in the previous 12 months that the participant believed or knew to be HIV-positive (aOR 0.17) were less likely to have a recent infection than a long-term infection. Region, sex, marital status, age of sexual debut, partner age differences, and community-level viremia did not significantly differ between those with recent and long-term infections.