Widespread changes in sexual behaviour in eastern and southern Africa: Challenges to achieving global HIV targets? Longitudinal analyses of nationally representative surveys

Abstract Introduction Sexual behaviour change contributed to reductions in HIV incidence in eastern and southern Africa between 1990 and 2010. More recently, there are indications that non‐regular partnerships have increased. However, the effect of these increases on population‐level risks for HIV and other sexually transmitted infections could have been reduced by simultaneous increases in condom use. We describe recent trends in sexual behaviour and condom use within the region and assess their combined effects on population levels of sexual risk. Methods Nationally representative Demographic and Health Survey data on sexually active males and females (15 to 49 years) were used for 11 eastern and southern African countries (≥3 surveys for each country; 1999 to 2016) to describe trends in sexual behaviour (multiple, non‐regular, and casual sexual partnerships; condom use; age at first sex). Logistic regressions tested for statistical significance of changes. Analyses were stratified by sex. Results Recent increases in multiple, non‐regular, and/or casual partnerships can be found for males in 10 countries and, for females, in nine countries; five countries exhibited recent decreases in age of sexual debut. Reduction in sex without condoms with non‐regular partners was observed in six countries for males and eight for females. Changes in the proportion of the overall population reporting condomless sex with non‐regular partners varied between countries, with declines in six countries and increases in three. Conclusions Extensive change in sexual behaviour occurred across eastern and southern Africa during the period of scale‐up of antiretroviral therapy programmes. This includes increasing multiple and non‐regular partnerships, but their potential effects on population‐level sexual risks were often offset by parallel increases in condom use. Strengthening condom programmes and reintegrating communication about behavioural dimensions into combination prevention programmes could help countries to meet international targets for reductions in HIV incidence.


Details on data, measures, and methods
Table S1 provides information on the included surveys and the data available in each. Below details on the measures of sexual behaviour as well as the methods used for the analyses are provided. Further details on data and measures are also provided online (https://dhsprogram.com/).

Recent sexual activity:
The denominator for most analyses in this study was everyone who reported sexual intercourse in the past 12 months. This was based on the question "when was the last time you had sexual intercourse?" that was posed to everyone who reported having ever had sexual intercourse before.

Multiple sexual partnerships:
Multiple partnerships were based on the question "in total, with how many different people have you had sexual intercourse in the last 12 months?" that was asked from DHS phase 4 surveys onwards.
Non-regular and casual sexual partnerships: From DHS phase 4 onwards, data on the type of sexual partners in the past 12 months were gathered with the question "what is your relationship to the man/woman with whom you last had sex?" Non-regular sexual partners were all types of non-spousal, non-cohabiting partners. Casual partners were those referred to as 'casual acquaintance' or 'commercial sex worker/client'. Casual partners are one type of non-regular partners. In DHS phase 4 surveys, these data were collected for the last two sexual partners in the past 12 months; from DHS phase 5 surveys, these data were collected for the last three sexual partners in the past 12 months. Some phase 5 or later surveys did only collect these data on the last two sexual partners. However, since very few people report three previous sexual partners, these differences have negligible effects.

Age at first sex:
The question on age at first sex was asked everyone reporting ever having had sexual intercourse: "How old were you when you first had sexual intercourse?" Analyses on age at first sex were restricted to everyone who had sex (irrespective of sexual activity in the past 12 months) and those aged 20-29 years.

Condom use:
A question on condom use was asked from DHS phase 3 surveys onwards. In phase 3 surveys, those who reported having ever had sexual intercourse were asked: "the last time you had sex, was a condom used?" From phase 4 onwards, this question was asked for each sexual partner in the past 12 months. In phase 4, the wording was "the last time you had sexual intercourse, was a condom used?" and, referring to another sexual partner in the past 12 months, "have you had sex with any other man in the last 12 months?" From phase 5 onwards, the wording was "the last time you had sexual intercourse (with this second/third person), was a condom used?" Condom use with non-regular or casual partners was defined as condom use the person had last sex with a non-regular or casual partner, which may not be the last sexual partner in the past 12 months.

Details on methods:
To create nationally representative samples, surveys tended to over-sample certain region, which required the application of survey sampling weights to the data so that results are nationally representative. These sampling weights were supplied with each DHS data set. Moreover, individuals were sampled in clusters in each survey. The size and numbers of these clusters differed markedly between countries and within countries over several surveys. Details are provided on the DHS website (https://dhsprogram.com/). Responses of individuals within the same clusters tend to be correlated (nonindependence), so this sample design clustering needs to be accounted for. All analyses were conducted in STATA Release 14 (StataCorp. 2015. College Station, TX: StataCorp LP); the svy command was used to apply survey sampling weights and account for sample clustering (see the survey data reference manual for details [1]).  Table S2 provides the years when ART was introduced in the eastern and southern African countries included in this analysis and when 30% of adults living with HIV were on treatment. In figures in which these periods are indicated, the first day of each of these years were used as the date of introduction. indicates the year when antiretroviral therapy (ART) was introduced in the public healthcare sector (ART may have been available earlier through private healthcare providers). These years refer to the year when ART was made widely available for free or highly subsidised as opposed to small-scale pilot projects (see notes below). The year of 30% ART coverage refers to the year when 30% of adults (15+ years) who were HIV-positive received ART (eligibility criteria for ART were disregarded). a Pilot provision of ART started in 2001 in 5 sites and expanded to 100 sites at the end of 2004. b ART was first made available at the end of 2004 in one hospital and expanded to 22 sites by the end of 2005. c For calculating ART coverage, the number of adults (15+ years) on treatment was taken from the referenced report and divided by the total estimated number of people living with HIV in the year provided by UNAIDS [3].

Additional data and results on primary sexual behaviour measures
Tables S3-6 provide levels of multiple partnerships, non-regular partnerships, condomless sex with non-regular partners among those with non-regular partners, and condomless sex with non-regular partners among everyone who had sex with 95% CIs by sex for each country and survey, and p-values of logistic regressions comparing consecutive surveys, adjusted for age. Figures 1-4 in the main article present these trends over time. 9044 Sample sizes (n/N) refer to unadjusted numbers of people reporting multiple sexual partnerships (n) among everyone who sexual intercourse in the past 12 months with data on this variable (N). Proportions (%) and 95% confidence intervals (95% CI) are adjusted for survey design and sampling weights. The p-values refer to results from logistic regressions with odds ratios of multiple sexual partnerships calculated for one survey compared with the preceding one, adjusted for age. These are also adjusted for survey design and sampling weights.  (12.97-15.40) 0.0012 Sample sizes (n/N) refer to unadjusted numbers of people reporting non-regular sexual partnerships (n) among everyone who sexual intercourse in the past 12 months with data on this variable (N). Proportions (%) and 95% confidence intervals (95% CI) are adjusted for survey design and sampling weights. The p-values refer to results from logistic regressions with odds ratios of non-regular sexual partnerships calculated for one survey compared with the preceding one, adjusted for age. These are also adjusted for survey design and sampling weights. 0.0102 Sample sizes (n/N) refer to unadjusted numbers of people reporting not using condoms during the non-regular casual sexual intercourse (n) among everyone who reported non-regular sexual intercourse in the past 12 months with data on this variable (N). Proportions (%) and 95% confidence intervals (95% CI) are adjusted for survey design and sampling weights. The p-values refer to results from logistic regressions with odds ratios of not using condoms during last non-regular sexual intercourse calculated for one survey compared with the preceding one, adjusted for age. These are also adjusted for survey design and sampling weights. * The Lesotho 2004 DHS only collected condom use information on females.  (4.40-5.66) 0.6293 Sample sizes (n/N) refer to unadjusted numbers of people reporting not using condoms during the non-regular casual sexual intercourse (n) among everyone who reported any sexual partner in the past 12 months with (N). Proportions (%) and 95% confidence intervals (95% CI) are adjusted for survey design and sampling weights. The p-values refer to results from logistic regressions with odds ratios of not using condoms during last nonregular sexual intercourse calculated for one survey compared with the preceding one, adjusted for age. These are also adjusted for survey design and sampling weights. * The Lesotho 2004 DHS only collected condom use information on females.

Data and results on secondary sexual behaviour measures
Tables S7-9 provide levels of casual partnerships, condomless sex casual partners, and age at first sex below 18 years with 95% CIs by sex for each country and survey, and p-values of logistic regressions comparing consecutive surveys, adjusted for age. Figures S1-3 present these trends over time. 3503 Sample sizes (n/N) refer to unadjusted numbers of people reporting casual sexual partnerships (n) among everyone who sexual intercourse in the past 12 months with data on this variable (N). Proportions (%) and 95% confidence intervals (95% CI) are adjusted for survey design and sampling weights. The p-values refer to results from logistic regressions with odds ratios of casual sexual partnerships calculated for one survey compared with the preceding one, adjusted for age. These are also adjusted for survey design and sampling weights.   1878 Sample sizes (n/N) refer to unadjusted numbers of people reporting condom use during the last casual sexual intercourse (n) among everyone who reported casual sexual intercourse in the past 12 months with data on this variable (N). Proportions (%) and 95% confidence intervals (95% CI) are adjusted for survey design and sampling weights. The p-values refer to the results from Pearson's chi-squared (χ 2 ) tests comparing the values for one survey with the previous one. These are also adjusted for survey design and sampling weights.    0.0009 Sample sizes (n/N) refer to unadjusted numbers of people reporting an age at first sex before the age of 18 years (n) among everyone who reported sexual intercourse before with data on this variable and who were aged 20-29 years (N). Proportions (%) and 95% confidence intervals (95% CI) are adjusted for survey design and sampling weights. The p-values refer to results from logistic regressions with odds ratios of age at first sex before 18 years calculated for one survey compared with the preceding one, adjusted for age. These are also adjusted for survey design and sampling weights. Figure S3a. Trends in age at first sex before the age of 18 years among males (20-29 years), eastern and southern Africa. The sample included males aged 20-29 years who reported ever having had sexual intercourse. Dates refer to the mid-points of the survey data collection period. Data from AIS are indicated in red. Data from different survey types are linked with dashed lines. Shaded areas indicate the years in which ART was introduced into the public healthcare sector in each country and from when 30% of adult PLWH (15+ years) were in treatment (disregarding treatment eligibility criteria; see Supporting Information, Section 2). Figure S3b. Trends in age at first sex before the age of 18 years among females (20-29 years), eastern and southern Africa. The sample included females aged 20-29 years who reported ever having had sexual intercourse. Dates refer to the mid-points of the survey data collection period. Data from AIS are indicated in red. Data from different survey types are linked with dashed lines. Shaded areas indicate the years in which ART was introduced into the public healthcare sector in each country and from when 30% of adult PLWH (15+ years) were in treatment (disregarding treatment eligibility criteria; see Supporting Information, Section 2).

Analysis by HIV status
Five countries had at least three surveys with HIV status data available (Lesotho, Malawi, Rwanda, Tanzania, and Zimbabwe) (Table S1), which allows for comparing trends in risky sexual behaviour over time between those HIVnegative and HIV-positive. This HIV status was determined at the time of the survey and does not necessarily mean that participants are aware of their status. Particularly in earlier surveys, the proportion of participants who have ever done an HIV test is likely to have been low.
Only multiple sexual partnerships, non-regular partnerships, and condom use during last non-regular partnership are considered here. Figures S4-6 show trends in these behaviours over time by sex and HIV status. To formally test whether there are differences in trends by HIV status, logistic regression models were estimated with the sexual behaviour as the outcome and age, survey, HIV status, and an interaction term between the survey and HIV status as independent variables. Regressions included a survey variable covering all surveys for a country (not just consecutive surveys as in previous regressions). Regressions are estimated separately by sex. Results are presented in Table S10.
As can be seen from the figures and the interaction terms, trends for those HIV-positive and HIV-negative tend to be similar. Levels of multiple partnerships tended to be higher among those HIV-positive ( Figure S4a-b). Only In Tanzania there is a markedly different trend between those HIV-positive and HIV-negative, although this interaction is not significant (Table S10). For non-regular partnerships, levels among those HIV-negative tend to be higher among males but lower among females ( Figure S5a-b), while trends are generally similar (with the exceptions of Zimbabwe among males and Lesotho among females). For condom use with the last non-regular partner, significant interactions are indicated for Lesotho (males) and Zimbabwe (females) (Table S10). However, the direction of the trend is the same; only the magnitude of increase varies by HIV status. 0214 For each country and by sex, a logistic regression was estimated with one sexual behaviour variable as the outcome and age, the survey, HIV status, and an interaction term between survey and HIV status as independent variables. The p-values refer to the individual interactions of the surveys and the interaction as a whole. Samples included everyone who reported sexual activity in the past 12 months. Regressions adjusted for survey design and sampling weights.

Figure S4a. Trends in multiple sexual partnerships among males (15-49 years) by HIV status, eastern and southern
Africa. Multiple sexual partnerships were defined as reporting more than one sexual partner in the past 12 months. Dates refer to the mid-points of the survey data collection period. Figure S4b. Trends in multiple sexual partnerships among females (15-49 years) by HIV status, eastern and southern Africa. Multiple sexual partnerships were defined as reporting more than one sexual partner in the past 12 months. Dates refer to the mid-points of the survey data collection period.

Figure S5a. Trends in non-regular sexual partnerships among males (15-49 years) by HIV status, eastern and southern Africa.
Non-regular sexual partnerships were defined as reporting at least one sexual partner in the past 12 months who the participant was not married to and did not live with. Dates refer to the mid-points of the survey data collection period. Figure S5b. Trends in non-regular sexual partnerships among females (15-49 years) by HIV status, eastern and southern Africa. Non-regular sexual partnerships were defined as reporting at least one sexual partner in the past 12 months who the participant was not married to and did not live with. Dates refer to the mid-points of the survey data collection period.

Figure S6a. Trends in condom use with the last non-regular sexual partners among males (15-49 years) by HIV status, eastern and southern Africa.
The condom use refers to the last sexual intercourse in the past 12 months with a partner who the participant was not married to and did not live with among everyone who had such a non-regular partner in the past 12 months. Dates refer to the mid-points of the survey data collection period. Figure S6b. Trends in condom use with the last non-regular sexual partners among females (15-49 years) by HIV status, eastern and southern Africa. The condom use refers to the last sexual intercourse in the past 12 months with a partner who the participant was not married to and did not live with among everyone who had such a non-regular partner in the past 12 months. Dates refer to the mid-points of the survey data collection period.

Sensitivity analysis: Including non-sexually active individuals
The analyses presented in the main article and other analyses in this supplementary material were restricted to those who ever had sex (and further those reporting sexual activity in the past 12 months). Conclusions drawn from these analyses regarding trends in the sexual behaviour may be biased if there are changes in levels of sexual activity in the population. For example, if the proportion of those who ever had sex declines over time then there may be declines in population levels of multiple sexual partnerships in the population, even if there are increases in multiple sexual partnerships among those who are sexually active.
Table S11 and Figure S7a-b presents trends in reporting of having had sex before. The trends in sexual activity reflect, to a large degree, the trends in age of sexual debut presented in Figures S4a-b. Particularly in earlier surveys among males ( Figure S7a), there were declines in the proportion of those reporting sexual debut, often correlating with decreasing proportions reporting an age of first sex before the age of 18 ( Figure S3a), while trends among females were less pronounced ( Figure S7b). Generally, changes in tended to be small in absolute terms. More importantly, decreases in the proportion reporting sexual debut tended not to occur between later surveys (exceptions are Malawi, and Namibia for males). This means it is unlikely that the conclusions drawn from the main analyses of recent increases in risky sexual behaviour are invalid due to overall decreases in sexual activity in recent surveys.
To test formally whether conclusions are impacted by the selected sample, a sensitivity analysis was conducted to estimate the proportions and statistical differences over time for the three primary indicators of the main analysis (multiple, nonregular, and casual partnerships) including those who have not had sex before and who have not had sex in the past 12 months (who were recorded as not reporting multiple, non-regular, or casual partnerships). Moreover, proportions of those aged 20-29 reporting an age at first sex before the age of 18 were estimated in this sample including those who not had their sexual debut as estimates in this sample may be particularly sensitive to changes in the proportion of those sexually active. The results are presented in Table S12 (multiple partnerships), S13 (non-regular partnerships), S14 (casual partnerships), and S15 (age at first sex). Generally, there is limited indication that the conclusions drawn in the main analysis were biased by restricting the sample to those sexually active. The same trends appear in the sample including those not sexually active and similar statistical differences between surveys within countries tended to be found in this sample. There were few instances in which a difference between surveys was statistically significant in the sample that excludes those who have not had sex compared to the sample that included these (and vice versa). We therefore concluded that no bias was introduced when analysing the sample excluding those who have not had sex before, which is better suited to represent trends in sexual behaviour. 6724 Sample sizes (n/N) refer to unadjusted numbers of people reporting to have had sex (n) among everyone with data on this variable (N). Proportions (%) and 95% confidence intervals (95% CI) are adjusted for survey design and sampling weights. The p-values refer to results from logistic regressions with odds ratios of having had sex calculated for one survey compared with the preceding one, adjusted for age. These are also adjusted for survey design and sampling weights.     (13.28-15.16) <0.0001 127/9900 1.10 (0.88-1.38) 0.9754 Sample sizes (n/N) refer to unadjusted numbers of people reporting multiple sexual partnerships (n) among everyone (including those who have not had sex before) (N). Proportions (%) and 95% confidence intervals (95% CI) are adjusted for survey design and sampling weights. The p-values refer to results from logistic regressions with odds ratios of multiple sexual partnerships calculated for one survey compared with the preceding one, adjusted for age. These are also adjusted for survey design and sampling weights. S13: Non-regular sexual partnerships (including those who have never had sex before), eastern and southern Africa.