Cross-sectional trends in HIV prevalence among pregnant women in Botswana: an opportunity for PrEP?

Introduction: Young women in sub-Saharan Africa are at particularly high risk of HIV acquisition. Recent shifts towards “test and treat” strategies have potential to reduce transmission in this age group but have not been widely studied outside of clinical trials. Using data from nationwide surveillance among pregnant women in Botswana, where a “test and treat” program was implemented in 2016, we describe trends in HIV prevalence over time and highlight opportunities for targeted prevention. Methods: The Tsepamo study abstracted data from obstetric records of all women delivering at eight government hospitals in Botswana between 2015 and 2019, accounting for 45% of all births in the country (n = 120,755). We used a stratified analysis to identify prevalence trends and evaluated decreases in HIV prevalence over time using the Cochrane–Armitage test for linear trend. A multivariable logistic regression analysis was also performed to identify factors associated with declines in HIV prevalence. Results: Overall HIV prevalence was 24.1% among 120,755 women who delivered during the study period. Prevalence differed by site of delivery, ranging from 16.1% to 28.2%, and increased markedly with age. Lower educational attainment (adjusted odds ratio [aOR] = 3.28; 95% confidence interval [CI] 3.07–3.50) and being unmarried (aOR = 1.98; 95% CI 1.88–2.08) were associated with HIV infection. HIV prevalence was 10.0% with a first pregnancy, 21.0% with a second and 39.2% with a third or greater (aOR = 2.20; for any prior pregnancy; 95% CI 2.10–2.29). The same age-adjusted trends were seen when data were limited to women aged 15–24, with a two- to three-fold increase in HIV prevalence between a first and third pregnancy. Prevalence decreased linearly during the 5-year study period from 25.8% to 22.7% (p <0.001). Among age-specific strata, the greatest absolute decline occurred in those aged 35–39, with an 8.7% absolute decrease in HIV prevalence from 2015 to 2019. Minimal declines were seen in those 15–24, with a decrease of only 1.5% over the same period. Conclusions: While overall trends in Botswana show HIV prevalence declining among pregnant women, prevalence among the youngest age group has remained stagnant. Preventative interventions utilizing pre-exposure prophylaxis should be prioritized during the high-risk period surrounding a woman's first pregnancy. We conducted a retrospective serial cross‐sectional study using data from the Tsepamo Study [15], which performs birth outcomes surveillance at 18 government hospitals in Botswana. The Tsepamo database is a nationally representative dataset that captures data from the obstetrical records of women in Botswana who deliver live or stillborn infants in participating maternity wards. For this analysis, we utilized data from the eight original Tsepamo sites, which had complete data from January 2015 to December 2019, and accounted for approximately 45% of all births in the country during this period. Ten sites were excluded given that they were added between 2018 and 2019, and did not have complete data covering the study period of interest. The study sites included were comprised of two tertiary referral hospitals, five district hospitals and one primary‐level hospital. Analyses were performed using SAS 9.4 University Edition and Stata (Version 16, StataCorp, College Station, TX). Ethics approval for this study was granted by the Botswana Human Research and Development Division and by the institutional review board of Harvard T. H. Chan School of Public Health. Informed consent was not required because records were deidentified and the study was observational. Our main exposures of interest were maternal age, calendar time and hospital location. Maternal age was documented at delivery and categorized into four strata (<15, 15–24, 25–35 and >35). However, narrower age strata using increments of 5 years were used in the descriptive analysis to allow for more granular trends to be characterized. Calendar time was analysed categorically and stratified by year to allow for the assessment of annual trends. Hospital location served as a proxy for maternal home district and covered both urban and rural areas of Botswana. Additional demographic data, including marital status, education, gravida and occupation, were used as measures of socio‐economic status and included in the regression model as categorical variables to identify factors associated with HIV infection. The primary outcome was HIV status at delivery, which was used to calculate prevalence. Maternal HIV status was obtained directly from the obstetric card and in most cases, was confirmed by additional HIV treatment records and maternal confirmation at the maternity ward. A subset of diagnoses were also verified through direct access to Botswana’s national HIV laboratory system if there were discrepancies. HIV diagnosis in Botswana is generally made using dual enzyme immunoassay testing, and HIV RNA testing by PCR is also performed in the context of treatment. At each site, research assistants abstracted de‐identified data, including the primary outcome and covariates, from maternal obstetrical cards at the time of discharge from the postnatal ward and entered into the Tsepamo database. Maternal deaths during delivery were not included in the dataset. We used descriptive stratified analyses to identify prevalence trends by age, calendar year, gravida and location. Our unadjusted stratified analysis included a quantitative assessment of trends using Cochran–Armitage trend testing to help identify significant decreases in HIV prevalence over time. Subsequently, we used a multivariable logistic regression model, where HIV status at delivery was used as the outcome variable, to estimate the adjusted odds ratios of being HIV positive while controlling for age, calendar year, delivery site, marital status, education, gravida and employment. The model was also used to verify significant trends in HIV prevalence by age, calendar year and location in the descriptive analysis. A complete case analysis was performed and collinearity testing between variables was also conducted to ensure that variables with a variance inflation factor (VIF) greater than five were not included in the regression model. Sensitivity analyses were also performed to determine whether women with a missing or undisclosed HIV serostatus impacted the direction of associations by assuming that women with missing HIV serostatus were either all HIV positive or all HIV negative.