Rates of sexual partner acquisition from nationally representative surveys: variation between countries and by age, sex, wealth, partner and HIV status

Introduction

There is a substantial burden of sexually transmitted infections (STIs), which is largely concentrated in low- and middle-income countries. Worldwide, this includes 376 million new cases of curable STIs (chlamydia, gonorrhoea, trichomoniasis, and syphilis) annually,¹ 417 million prevalent infections of herpes simplex virus, type 2 (HSV-2),² and, in 2022, 1.3 million new HIV infections and 630 000 HIV-related deaths.³ There have been substantial reductions in HIV infections in recent years, attributable to roll out of interventions including antiretroviral treatment (ART), voluntary medical male circumcision and pre-exposure prophylaxis.^4–6 Despite this, the target of the UN General Assembly’s 2016 Political Declaration on Ending AIDS to reduce new HIV infections to fewer than 500 000 by 2020 was not achieved and countries are not on track to reach zero infections by 2030. Progress in reducing new infections of other STIs has stagnated in much of the world where there is continued reliance on syndromic management guidelines developed in the 1980s.^1,7

Mathematical modelling has shown that stratifying prevention interventions by behaviour offers gains in both efficiency and infections prevented,⁸ but requires a metric that captures a group of individuals who are high risk. Structural and individual-level risk factors have been identified for HIV and other STIs, but it has not proved straightforward to use these to identify groups who would most benefit from interventions in terms of finding a balance between specificity and sensitivity. For example, interventions targeted at sex workers might omit those who do not yet identify as sex workers, while those targeted at all young women may redundantly deliver interventions to many young women who were never at risk. Mathematical models have been widely used to understand drivers of STI transmission and to assess the potential impact of interventions.⁹ These models typically incorporate parameters to describe the rates and patterns of contacts between people who are susceptible to STIs and those who are infected with, or infectious for, STIs.^9–13 However, despite the acknowledged importance of rates of partner acquisition in determining the spread of STIs and HIV through populations, few studies have attempted to systematically quantify levels of partner change and assess how they differ between and within populations.¹⁴

Existing empirical studies have generally relied on relatively simple measures; for example, the number of people who have had a new partner in a given period of time (which disregards multiple acquisitions) or the number of people with more than one partner.^15,16 We hypothesise that new partnerships may have a more important effect on STI transmission than multiple partnerships for behavioural, network and biological reasons. Behaviour is likely to differ between new and established partnerships due to differences in emotional connection, power balance, communication and ability to negotiate. The position in the sexual network is different for people who have a new partner than for those with an existing partner since, on average, they are more closely connected to others in the network at the start of the partnership. In a situation where everyone in the population was serially and mutually monogamous, the start of a partnership is the point in time when the partner has most recently had sex with another person. If this gap is smaller than the period of infectiousness, then there is a risk of STI transmission.¹⁷ Furthermore, for HIV, there may be a biological reason if alloimmunity to sexual partners protects against acquisition of infection. In South African couples, class 2 human leukocyte antigen antibodies (which are triggered by sexual contact) were associated with reduced odds of HIV seroconversion in exposed individuals.¹⁸ Therefore, at the start of a partnership, the levels of these antibodies may be lower.

Rates of new partnership formation have not been systematically estimated and compared between populations and the most recent review of global data on partnerships was in 2006.¹⁹ Measuring the incidence of new partnerships alongside the prevalence of recent partnerships could provide a more complete picture of population-level partnership dynamics. In our study, we used data from Demographic and Health Surveys (DHS) to conduct an updated review of the variation in the number of sexual partnerships reported by men and women in different countries. We calculated partner acquisition rates (PAR) to focus on the newest partnerships. We sought to compare PAR by characteristics that may be used to target STI prevention interventions: sex, age, wealth, current partner status and HIV status. To explore the utility of PAR as a means to identify groups at high risk of STIs, we assessed how they correlate with HIV incidence estimates at the population-level.

Materials and methods

We conducted secondary analyses of cross-sectional, nationally representative household surveys conducted by DHS since 2010 and estimated, for men and women, the rates at which new sexual partners were acquired and characteristics associated with variation in those rates.

Results

Partner acquisition rates

The PAR for men had a median across surveys of 30 per 100 py and varied from 7/100 py in Niger to 64/100 py in Gabon with an interquartile range (IQR) of 21 to 45 (Table 1). For women, the median PAR was 13/100 py and ranged from 4/100 py in Cambodia to 32/100 py in Gabon with an IQR of 6–18. In all surveys, the rate reported was higher for men than women (Fig. 1).

Fig. 1.

Partner Acquisition Rate (PAR) for men and women in the most recent survey in each country.

Fig. 2 shows the age-specific PAR for men and women. The PAR peaked for men in their early 20s and then declines in all settings. Across all settings, the peak was both much lower and slightly earlier in women (late teens).

Fig. 2.

Age-specific partner acquisition rates (per person per year) by country for (a) men and (b) women using data from the most recently conducted survey in each country.

Our synthetic measures of partner acquisition (TPAR) are in Fig. 3. For men, the lowest TPAR was 2.4, reported in Niger, and the highest was 21 reported in Gabon. Women in Cambodia had the lowest TPAR with 1.2 partners by age 50 years. The highest TPAR for women was 9.0, in Gabon. There was little correlation at the national level between the TPAR for men and women.

Fig. 3.

Synthetic measure of the cumulative total numbers of partners acquired by age 50 years (TPAR) for men and women by country (most recent survey).

Variation in partner acquisition rates by wealth

The age-adjusted hazard ratios (HR) for partner acquisition by household wealth quintile for men and women in each survey showed some clear trends with wealth (Fig. 4; estimates in Tables S3 and S4) although in 16 countries there was no pattern of association for either men or women.

Fig. 4.

Age-adjusted hazard ratio (HR) for new partner acquisition by wealth quintile (baseline group = middle) for (a) men and (b) women stratified by survey.

The most common pattern was an upwards trend in PAR with increasing wealth, where the richest had the highest rates of partner acquisition. This was seen for both sexes in 10 countries (Benin, Burkina Faso, Cameroon, Cote d’Ivoire, East Timor, Guatemala, Haiti, Niger, Sierra Leone and Togo), for women only in three countries (Kenya, Namibia and Nigeria) and for men only in Cambodia, Gambia and, with less certainty, Papua New Guinea and Uganda. This was the dominant pattern of association in western African countries.

There was considerable variation across the remaining surveys. In India and Madagascar, the pattern was reversed with the highest rates observed among the poorest quintile. Elsewhere there was not a linear trend but a pattern of association such that the richest and poorest have higher rates compared to the middle quintile (Burundi and Chad), or the inverse with highest rates in the middle quintile (Angola, Armenia, Colombia, Ghana, Zambia). In 10 countries, the patterns of association were seen only for one sex.

Variation in partner acquisition rates by partner status at the time of acquisition

Fig. 5 shows the HR for acquiring a new partner comparing those who are already cohabiting with at least one partner with people who do not have a partner (estimates in Tables S5 and S6). Everywhere, the hazards of acquiring a new partner were lower for people already in a cohabiting partnership. The HRs for men ranged from 0.01 (95% CI 0.006–0.029) in Albania 2017 to 0.60 (95% CI 0.42–0.88) in Niger and for women from 0 in the Philippines to 0.09 (95% CI 0.06–0.15) in Kenya. The HR for women was consistently lower than that for men, but the degree of the difference varied between surveys.

Fig. 5.

Age-adjusted hazard ratio (HR) for partner acquisition among people already cohabiting with a partner compared to single people. Estimates from Zimbabwe are from 2010 for men and 2015 for women. There are no estimates for the Philippines because only single women reported acquiring a new partner.

Variation in partner acquisition rates by HIV status

Fig. 6 shows the age-adjusted HR for partner acquisition by HIV status for men and women in 24 surveys (estimates in Tables S7 and S8). For women in all surveys, except India in 2015, the HR indicated a higher PAR among women living with HIV compared to HIV-negative women. For example, women living with HIV had over twice the rate of partner acquisition compared with women without HIV in Zimbabwe (HR = 2.27; 95% CI = 1.96–2.62). In Chad, Ethiopia, DR Congo, Gabon, Gambia, Guinea, India, Senegal and Sierra Leone the CIs included one but in the other 14 surveys they do not. There was not such a clear pattern for men. In only three surveys (Rwanda, Senegal and Zimbabwe), the CIs did not include one and men living with HIV had a higher PAR than men without HIV.

Fig. 6.

Age-adjusted hazard ratio for partner acquisition among people living with HIV compared to people who are HIV negative. CI missing for the India 2019 estimate because only two acquisitions were reported by women living with HIV.

There were no UNAIDS estimates available on HIV incidence for Comoros and Nigeria, but for the other 31 countries with data on PAR and on HIV incidence in Africa, the PAR was positively correlated with HIV incidence among both men and women (Fig. 7; Table S9). In western and central Africa, the coefficients were 0.09 (95% CI 0.01–0.17) for men and 0.27 (95% CI 0.11–0.47) for women (i.e. in western and central Africa there was a 0.09% increase in the HIV incidence with each one unit increase in the PAR for men). In eastern and southern Africa, the coefficients were 0.36 (95% CI 0.26–0.46) for men and 1.17 (95% CI 0.95–1.40) for women. A similar association was seen with the percentage of the population who had a new partner, and for the percentage sexually active. For multiple partners, an association was only evident in eastern and southern Africa (Table S9).

Fig. 7.

Scatterplot of partner acquisition rates from DHS and national level HIV incidence estimates from UNAIDS, by region and sex and fitted regression lines (note different scales).

Discussion

We present updated estimates of partnerships among men and women and the first cross-national estimates of rates of partner acquisition. We show a wide range of estimates for men and women and interesting trends across countries. We see substantial variations between countries in the PAR, with no clear geographical patterns. We consistently see higher PAR among men and accordingly the TPAR are consistently lower for women than men. PAR were highest for young people and people who were single or who had only a non-cohabiting partner(s). Women living with HIV had higher PAR, identifying a group who are both vulnerable to STIs and potentially important for onwards transmission, but the patterns by HIV status were more variable among men. The patterns of PAR by age, sex and HIV are generally consistent with differences observed in other reported sexual behaviour by these characteristics.^26–28

Some of the differences between and within countries in the levels and correlates of PAR are likely to reflect differences in the broader social and demographic context. Acquisition of sexual partners is a social activity and therefore influenced by social norms, gender roles and opportunities to encounter new partners. For example, in contexts where gendered social norms associate masculinity with greater numbers of sexual partners,²⁹ PAR among men, and the gap between men’s and women’s PAR, would be expected to be higher. The demographic context can influence the national PAR through compositional differences between populations, even where the PAR for population subgroups are similar across countries. At national level, the PAR is determined both by the relative size of the subgroups and the PAR for each group. The size of the groups will depend on the age structure and other factors, such as age at first marriage and divorce rates, which vary by country.^28,30

We found widespread, but not universal, associations between wealth and PAR. The variations shown in these data lend further evidence to the contextual factors shaping behaviours and the way these vary by setting. The finding that PAR varies according to socio-demographic characteristics and HIV status demonstrates the importance of accurate parameterisation of partnership dynamics in models of HIV and other STIs to fully capture the differences between sexes and populations and thereby provide the best strategic information to model users.

Risk relating to partnerships, at both the individual and population-level, has usually been described by focusing on multiple partners, rather than new partners (i.e. prevalent rather than incident partners). The two measures are correlated, since multiple new partner acquisitions in a period of time are necessary for someone to have multiple partners. We believe they provide different, complementary information. We have shown a linear correlation between measures of new partner acquisition and national-level HIV incidence in western and central Africa as well as in eastern and southern Africa, whereas the proportion with more than one partner was correlated only in eastern and central Africa. While there could be other explanations for that association, rather than a causal link between numbers of new partnerships and HIV infection, it does suggest that PAR may be a more discriminatory measure than multiple partnerships in identifying populations with high levels of STI risk.

The strengths of the study include the use of nationally representative data available across many low- and middle-income countries. The data come from standardised enquiries conducted by highly experienced survey teams using comparable methods and validated questions and so provide excellent data for cross-survey comparisons.³¹ PAR captures information about people who had multiple new partners which is not accounted for by the simpler measure of the proportion who had a new partner. Complex data manipulation is required to prepare the data for survival analysis and to estimate the PAR from the DHS standard recode. However, this approach permits analyses of partnership acquisition with a time-varying exposure, such as looking at partner status as a determinant of new partner acquisition.

There are limitations to our analysis. First, we have looked only at one aspect of partnerships dynamics, acquisition, because the DHS data do not permit estimation of the rates of partnership dissolution. The longevity of partnerships and the rate at which they dissolve is important, because the ending of a partnership is often a significant life-event that affects wellbeing, sexual and reproductive behaviour, and that has a direct effect on the likelihood of partner acquisition. Second, the data may be subject to some biases. Self-reported data are subject to reporting bias,^26,32 and the partner history can be onerous for respondents and interviewers, potentially affecting data quality. These issues could lead to an underestimation of partner acquisition if respondent or interviewer fatigue, or a social desirability bias, leads to the omission of some partners. Sex differences in the reporting of sexual partnerships arise because of differences in the ages of partners¹⁹ and differences in estimation strategies³³ as well as differences in social desirability biases.²⁶ Women are likely to under-report their numbers of sexual partners^26,33 and this could bias our comparisons of PAR if the extent to which women under-report varies in such a way as to distort the associations with other variables. However, we were principally interested in relative rather than absolute PAR, and a bias that applies to all women would have no effect on this. If under-reporting varies between countries and by, for example, wealth, then differential under-reporting could explain our varied findings with respect to PAR and wealth in different countries.

Third, the partner history asked about a maximum of three partners in the past 12 months. In each survey, a small proportion of people (<1% of female respondents and a median of 1% of male respondents) reported more than three partners in the past year, and therefore, we may not have complete data on all acquisitions. The bias introduced is likely under-enumeration of short-term partners from further back in time and would underestimate PAR but our sensitivity analysis suggests the impact of this is negligible. Fourth, these household based surveys may miss certain groups of people who have very high numbers of partners, such as sex workers, although the effect of this is likely to be minimal for general population estimates because generally these are a small group. Lastly, there were two key limitations to our analysis looking at the individual level association between HIV and PAR: (1) we cannot show that partner acquisition preceded HIV infection; and (2) we rely on HIV status at the time of the survey and assume that this did not change in the 12 months preceding the survey. Given the incidence rates in these populations, we assume that the vast majority of infections would have been acquired more than 12 months before the survey and therefore precede our period of observation.

Simple screening tools to identify individuals in the general population at high risk of STIs, particularly HIV, remain elusive and hard to translate between settings.³⁴ Despite its limitations, PAR provides a metric that can be used to flag groups of people among whom STI transmission might be higher. PAR is unambiguous to define, does not describe a stigmatised behaviour, describes a quantifiable behaviour, and shows considerable variation in the rates (i.e. there is good discriminatory power). Unlike other behavioural factors that have not been found to be associated with HIV at both individual and population level,³⁵ we find PAR to be associated with HIV prevalence at the individual-level and HIV incidence at the population level (and to provide a more consistent association than number of partners). PAR can provide valuable population-level information to forewarn of increasing infection risk and so identify populations for targeted interventions. At individual-level, PAR can be used to refine risk scoring to reflect the additional exposure that results from the behaviour of other people in the population. In populations where there is a lot of risky behaviour, infections can spread faster than in populations with lower levels of risk behaviour.

In summary, this is the first time PAR have been estimated, for men and women in 47 countries using high-quality comparable data and we observe substantial variation in PAR. This variation highlights the need for accurate local data for modelling and planning STI interventions and services since generic estimates of partnership formation are unlikely to be appropriate. The association between HIV and PAR in women at the individual-level and both sexes at the population-level suggests this metric can help identify subgroups in the population that may be linked to STI transmission. This suggests that PAR data, alongside data on dissolution rates, can better describe partnership dynamics in mathematical models of STI transmission and could help explain differences in transmission patterns between population groups. We also propose that new partnerships should be incorporated into risk assessment tools at both the population and individual-levels; at the level of the individual to identify who might need immediate STI prevention interventions, and at the population-level to identify groups where services and interventions need to be strengthened to identify people at high risk of STIs before incidence rates rise.

Supplementary material

Supplementary material is available online.