HIV acquisition in pregnancy: implications for mother‐to‐child transmission at the population level in sub‐Saharan Africa

Abstract Introduction A recent sero‐discordant couple study showed an elevated risk of HIV‐acquisition during the pregnancy/postpartum period per‐condomless‐coital‐act. This, along with previous studies, has led to concern over possible increased risk of mother‐to‐child (vertical) transmission, due to the initial high viral load in the first months after seroconversion, in a time when the woman and health services may be unaware of her status. This study looks at whether behavioural differences during the pregnant/postpartum period could reduce the impact of elevated risk of HIV acquisition per‐condomless‐coital‐act at the population level. Methods We used data from 60 demographic and health surveys from 32 sub‐Saharan African countries. Using the HIV status of couples, we estimated differences in serodiscordancy between HIV‐negative women who were pregnant/postpartum compared to those who were not pregnant/postpartum. We compare the risk of sexual activity over the pregnant/postpartum period to those not pregnant/postpartum. Using these risks of serodiscordancy and sexual activity along with estimates of increased HIV risk in the pregnancy/postpartum period per‐condomless‐coital‐act, we estimated a population‐level risk of HIV acquisition and acute infection, during pregnancy/postpartum compared to those not pregnant/postpartum. Results Sexual activity during pregnancy/postpartum varies considerably. In general, sexual activity is high in the first trimester of pregnancy, then declines to levels lower than among women not pregnant/postpartum, and is at its lowest in the first months postpartum. Adjusted for age and survey, pooled results show HIV‐negative pregnant women are less likely to have an HIV‐positive partner compared to those not pregnant/postpartum (risk ratio (RR) = 0.78, 95% CI = 0.68–0.89) and comparing the postpartum period (RR = 0.85, 95% CI = 0.73–0.99). Estimated population‐level risk for HIV acquisition and acute infection in pregnancy/postpartum was lower than would be inferred directly from per‐condomless‐coital‐act estimates in most countries, over the time of most risk of mother‐to‐child transmission, though there was variation by country and month of pregnancy/postpartum. Conclusions Estimates of population‐level HIV acquisition risk in sub‐Saharan Africa should not be taken directly from per‐condomless‐coital‐act studies to estimate vertical transmission. Changes in sexual behaviour and differences in HIV‐serodiscordancy during pregnancy/postpartum reduce the impact of increased risk of HIV acquisition per‐condomless‐coital‐act, this will vary by region.


Additional information for sections of the methods Data
We used data from 60 Demographic and Health Surveys (DHS) and AIDS indicator surveys (AIS) conducted in 32 sub-Saharan African countries between 2000 and 2017 in which HIV testing outcomes were available. DHS and AIS are nationally representative household surveys. All analyses account for the two-stage cluster sampling survey design and use the HIV weights or male sample weights, as appropriate, provided by DHS. In pooled analysis, surveys are re-weighted so that each survey contributes equally toward the analysis. From these surveys, 53 surveys from 29 countries were available with both data linking couples and HIV status data. The couple data are linked data from the individual men and women files where they have declared to be married (living together) at the time of the survey. A respondent may have a regular partner, or a spouse who resides de jure in the same household, but if, at the time of the survey that person is not a de facto cohabiting partner then they are not included as there is little information about the partner. Women included in the analysis were aged 15-49.

Analysis
Definitions of pregnancy and the postpartum period The risk of mother to child transmission occurs in the period of pregnancy and during breastfeeding in the postpartum period. Women were classified as pregnant if they answered yes to "are you currently pregnant", duration of the current pregnancy was also asked in months. Postpartum women were identified by looking at the date of birth of the last child and date of interview. We randomly impute the day of birth of the child across the month they were born in. If the month they were born is the same as the month of interview the day of birth was randomly imputed to be between the first of the month and the day of interview. Months postpartum were calculated from this (assuming a month is 30 days), as long as the woman was not currently pregnant. Using this information, we created a maternity status variable "months since conception" with months 1-9 being pregnancy and months 10-15 the six months postpartum. We further refined the postpartum period to include only women who reported they were currently breastfeeding. For some analyses this variable was grouped into three-month periods, the three trimesters of pregnancy and early and late postpartum.

Analysis of differences in couple serodiscordancy by months since conception
We repeated the analysis restricting to HIV negative women who reported having sex in the last week and no condom use in order to look at any variation in the levels of discordancy in HIV negative women by sexual activity.  (Table S2). This is true also for many countries for the early and late postpartum period, however the confidence intervals are very wide. Zimbabwe, 2015 2510 5.9 (4.9-7.1) 6.5 (5.3-7.8) 3.4 (1.7-6.4) 1.9 (0.6-6.0) 4.9 (2.4-9.9)    Figure S1: Risk of reporting having condomless sex in the last week by months since conception compared to those not pregnant/postpartum for HIV negative women by country adjusted for five-year age group and survey year. Figure S1 continued: Risk of reporting having condomless sex in the last week by months since conception compared to those not pregnant/postpartum for HIV negative women by country adjusted for five-year age group and survey year. Figure S2: Population level risk ratios adjusted for five-year age group, comparing stages of pregnancy and the postpartum period (shaded in grey) to not pregnant or postpartum women, for selected countries. Thick lines represent point estimates, lighter lines the 95% confidence intervals. Green lines represent estimates using risk ratios per coital act as constant over early pregnancy, late pregnancy and postpartum; red lines represent coital frequency using gradients of risk ratios per coital act. Grey line represents the risk ratio per coital act from Thomson et al. (Note: y-axis on the log scale) Figure S2 continued: Population level risk ratios for HIV acquisition, comparing stages of pregnancy and the postpartum period (shaded in grey) to not pregnant not postpartum women, for selected countries. Thick lines represent point estimates lighter lines the 95% confidence intervals. Green lines, using risk ratios per coital act as constant over early pregnancy, late pregnancy and postpartum; red lines represent coital frequency using gradients of risk ratios per coital act. Grey line represents the risk ratio per coital act from Thomson et al. (Note: y-axis on the log scale) Figure S2 continued: Population level risk ratios for HIV acquisition, comparing stages of pregnancy and the postpartum period (shaded in grey) to not pregnant not postpartum women, for selected countries. Thick lines represent point estimates lighter lines the 95% confidence intervals. Green using risk ratios per coital act as constant over early pregnancy, late pregnancy and postpartum; red lines represent coital frequency using gradients of risk ratios per coital act. Grey line represents the risk ratio per coital act from Thomson et al. (Note: y-axis on the log scale) Figure S3: Estimated population level acute infection ratios, comparing the last month of pregnancy and the breastfeeding postpartum period to not pregnant not postpartum women Assuming an acute infection window of three months (blue line) and two months (red line). The grey line is the estimated acute infection ratio using direct per-condomless-coital-act HIV acquisition risk ratios with an assumption of a three month infection window. Figure S3 continued: Estimated population level acute infection ratios, comparing the last month of pregnancy and the breastfeeding postpartum period to not pregnant not postpartum women Assuming an acute infection window of three months (blue line) and two months (red line). The grey line is the estimated acute infection ratio using direct percondomless-coital-act HIV acquisition risk ratios with an assumption of a three month infection window. Figure S3 continued: Estimated population level acute infection ratios, comparing the last month of pregnancy and the breastfeeding postpartum period to not pregnant not postpartum women Assuming an acute infection window of three months (blue line) and two months (red line). The grey line is the estimated acute infection ratio using direct percondomless-coital-act HIV acquisition risk ratios with an assumption of a three month infection window. Figure S4: Estimated ratio of acute infections in the fourth to sixth month of pregnancy, compared to not pregnant/postpartum(pp) women, assuming no ART initiation. Estimates use per-condomless-coital-act estimates taking into account differences in sexual activity in the last week and HIV-serodiscordancy with a partner, assuming an acute infection window of three months. Shown for each country given their sexual activity patterns (blue). Confidence intervals represent uncertainty around the per-condomless-coital-act risk ratio estimates. Also plotted is the point estimate of the per-condomless-coital-act risk ratios from Thomson et al 2018 (red). Sexual activity patterns Figure S5: Estimated ratio of acute infections in the seventh to ninth month of pregnancy, compared to not pregnant/postpartum(pp) women, assuming no ART initiation. Estimates use per-condomless-coital-act estimates taking into account differences in sexual activity in the last week and HIV-serodiscordancy with a partner, assuming an acute infection window of three months. Shown for each country given their sexual activity patterns (blue). Confidence intervals represent uncertainty around the per-condomless-coital-act risk ratio estimates. Also plotted is the point estimate of the per-condomless-coital-act risk ratios from Thomson et al 2018 (red). Figure S6: Estimated ratio of acute infections in the first to fourth month postpartum (pp), compared to not pregnant/postpartum women, assuming no ART initiation. Estimates use per-condomless-coital-act estimates taking into account differences in sexual activity in the last week and HIV-serodiscordancy with a partner, assuming an acute infection window of three months. Shown for each country given their sexual activity patterns (blue). Confidence intervals represent uncertainty around the per-condomless-coital-act risk ratio estimates. Also plotted is the point estimate of the per-condomless-coital-act risk ratios from Thomson et al 2018 (red).

Further Analyses
Exploring reasons for differences in serodiscordancy -Methods We hypothesised that the reason for differences in discordancy in HIV negative pregnant and postpartum women compared to those not pregnant/postpartum is that they are likely to have been exposed to their partner for longer or at higher frequency than those not pregnant/postpartum due to a) sexual activity being necessary to become pregnant and b) It is common to have children with the same partner, therefore pregnancies beyond the first will likely be with a partner to whom the women has been exposed.
We first looked at the relationship between duration of partnership and having an HIV infected partner for HIV negative women using the couple's dataset. We then looked at how the relationship between maternity statuses differs by duration of partnership. The DHS asks age at first marriage and calculates years since first marriage, therefore for those who have only had one union (identified in the DHS), we have the duration of the partnership, however for those currently in a union that is not their first we do not know the length of the partnership. Even in the partnership loop questions, where it is asked how long since your partnership started, if the partner is a spouse, the data is unavailable. Therefore, we restricted our analysis to HIV negative women who are still in their first union. We used marital duration first grouped into less than one year, one to two years, three to five years and five years and over. We then grouped it into a binary variable of less than one year or greater than one year. We use log binomial regression to calculate the risk ratios of being serodiscordant comparing pregnancy and postpartum women to those not pregnant/postpartum by duration of partnership, further adjusting for five-year age group, country and calendar year. We also calculated risk ratios comparing the risk of having an HIV positive partner with a marital duration of over one year compared to less than one year.

Results
The couple dataset used for the serodiscordancy analysis represents a varying proportion of all women by survey and by maternity status. For women not pregnant or postpartum the couple dataset represents between 24% to 72% of all women, for pregnant women it is much higher ranging between 39% to 93% and the postpartum period similar to pregnancy at between 29%-93%.
Using the data from cohabiting partners, HIV negative women who had only been married or cohabited once, those in their first year of the relationship were 1.37 (95%CI 1.07-1.74) times more likely to have an HIV positive infected partner compared to those in a relationship duration longer than a year when adjusting for five-year age group, place of residence, country and survey year. There was no evidence of a further decrease as duration of marriage increased beyond one year. There was no evidence that this effect varied by five-year age group (Wald test F=045, p=0.771).
The risk of an HIV negative women having an HIV positive partner comparing pregnant and postpartum women with those not pregnant or postpartum was consistently lower if the women are in their first year of marriage compared to being married for more than one year (Table S5).
Marital duration as defined as less than one year and more than one year was similar in the couple dataset when compared to the all women dataset, although there was a small decreased risk of being in a relationship for more than one year of 0.983 (95%CI 0.981-0.986) when adjusting for fiveyear age group, country and calendar year. There was no evidence for this varying over months since conception or five-year age group.

Discussion
In this analysis we found that HIV negative women in their first marriage were more likely to have an HIV positive partner in the first year of marriage compared to later years. This is probably due to the fact that at later duration of marriage women have on average had the same exposure whether pregnant or not, as most will have been pregnant before, so those pregnant and postpartum are more similar to those who are not. However, in early marriage, pregnancy would indicate higher frequency of condomless sex therefore a higher chance of becoming concordant positive with a partner sooner than those not pregnant, causing greater differentials between the two groups. From this analysis it appears that among married couples, although longer duration of partnership means that an HIV negative woman is less likely to have an HIV positive partner, what differentiates levels of discordancy in HIV negative pregnant/postpartum women from those not pregnant/postpartum is higher levels of sexual activity prior to or during pregnancy.