Glynn, Judith R.a; Dube, Albertb; Kayuni, Ndoliweb; Floyd, Siana; Molesworth, Annaa; Parrott, Fionaa; French, Neila; Crampin, Amelia C.a
HIV prevalence varies considerably between populations in ways that cannot be easily explained in terms of partnership numbers or known co-factors for transmission . It has been suggested that certain types of partnership patterns and sexual networks put individuals or their partners and community at greater risk for HIV and other sexually transmitted infections.
From a population angle, there is particular interest in concurrent partnerships; that is partnerships in which sex with one partner occurs between two acts of sexual intercourse with another partner. In theory, this could increase spread of HIV more than the same number of partners in serial monogamous relationships because of the increasing connectedness of sexual networks, and because of the variation in transmissibility of HIV, with high transmission shortly after initial infection [2–5]. Levels of reported concurrency vary widely from 2 to 55% in African studies, but the importance of concurrency has been challenged [6,7].
Concurrency is expected to increase the risk for the partners and community rather than for the individual themselves. This makes its effect difficult to study directly, other than through mathematical modelling. A recent study in an area of very high HIV incidence in South Africa found that male lifetime number of partners but not concurrency levels in the local community were associated with subsequent HIV incidence in women . This relied on a single measure of concurrency: being in more than one sexual relationship at the time of interview. If this is a poor measure of concurrency this would tend to underestimate any associations.
UNAIDS (Joint United Nations Programme on HIV/AIDS) guidelines recommend measuring concurrency as ‘point prevalence’ 6 months before the interview . This was chosen to avoid the problem of not knowing whether a recent relationship will continue or not. Point prevalence undercounts short-term partnerships compared to those with long-term overlaps: long-term overlaps are more important in the (theoretical) impact of concurrency . In practice, since start and end dates (unless very recent) are likely to be reported as x months ago, the ‘point’ is actually the 1-month period 6 months ago. So, for example, one-off sex recorded as having occurred 6 months ago might be counted, but one-off sex at any other time would not. Furthermore, since a relationship that starts in the same time period as another one finishes is discounted, this one-off sex will only contribute to concurrency if another relationship clearly spans this period, rather than starting or ending then. Given the difficulty of recall, and the tendency to choose round numbers (such as 6 months) when timing is uncertain, this may underestimate concurrency. UNAIDS guidelines also suggest that information from the three most recent partners will be sufficient.
Using prevalence 6 months ago as the measure relies on recording dates to calculate overlaps. An alternative is to ask directly about concurrency, either currently, as in the South African study , or for each relationship [e.g. during your relationship (ideally, between your first and last sex) with x did you have any other sexual partners?]. This can be refined to consider a certain period (e.g. during the past year) but might be difficult to restrict to a point 6 months ago. This direct approach is sometimes avoided by researchers because it is felt to be too intrusive, but recall about concurrency may be better than about dates, and missing and inaccurate data are less likely. Among young adults attending sexually transmitted disease clinics in the USA and reporting two or more partners, concurrency with the last partner showed poor agreement between direct reports and a dates-based measure (although the overall proportion was similar) . In a population-based study in Cape Town, direct questioning about concurrency identified partners not mentioned at all otherwise, who would therefore be missed by any dates-based measure .
Additional issues arise in polygynous societies, in which concurrency may be common, but in small discrete networks. This would be expected to be a relatively low risk form of concurrency . However, marriages may start as extra-marital partnerships, with a pregnancy increasing the likelihood of a partnership becoming a marriage, and this can occur more easily when polygyny is common.
Much has been written on different models of concurrency; less on the problems of actually measuring concurrency in practice [11,12,14]. As well as the usual problems of getting valid data on sexual partnerships [15,16], the different methods of measurement may give different answers.
In this study we use data from a population-based survey in rural Malawi to explore different methods of measuring concurrency and the explicit and implicit assumptions behind the UNAIDS recommendations.
The study was carried out as part of the Karonga Prevention Study in northern Malawi. Demographic surveillance was set up in 2002 in an area of the district covering a population of about 33 000, with a biannual census, and then annual censuses from 2007 . A sexual behaviour survey was added in 2007 for all adults aged 15–59. This was conducted a few weeks after the census update in each area, in conjunction with counselling and testing for HIV. The sexual behaviour questions were asked before the counselling and testing. Written consent was requested, and individuals could consent for the interview and refuse HIV testing or vice versa. The survey was conducted with individuals, in private. Further details are given elsewhere [18,19]. The study was approved by the ethics committee of The London School of Hygiene and Tropical Medicine, UK, and the National Health Sciences Research Committee of Malawi.
Information on spouses included current spouses (up to four for men), and up to two spouses with marriages ending in the past year. Information was collected on up to eight nonmarital partners in the past year. For spouses, date of marriage was recorded, but not date of first sex. For nonmarital partners date of first sex was recorded as a number of days, weeks, or months ago. Last sex was recorded for all partners as number of days, weeks or months ago. In addition participants were asked for each partner if the relationship was on-going. For each partner participants were asked: Over the past year, during your relationship with him/her, did YOU have other sexual partners (including husband/wife). They were also asked whether they thought their partner had other partners during the same period, and whether their partners were married. The total number of sex acts was asked for nonmarital partnerships (categorized as once, 2–4, 5–10, >10); and time since last sex for spouses.
Analyses were conducted separately for men and women. Because of the small number of women reporting multiple partnerships, most comparisons of different methods of measurement were conducted only for men. Several different measures of concurrency were calculated as summarized in Table 1. It was (arbitrarily) assumed that date of first sex with the spouse was 6 months before the marriage date (since it was not asked directly).
We tested the following assumptions inherent in the UNAIDS recommendations:
1. Use of dates to calculate overlaps will give a reasonable estimate of concurrency
a. By comparing self-reported and calculated concurrency
b. By measuring the effect of excluding possible overlaps with ties on recorded start and end dates
2. Concurrency 6 months ago is representative
a. By comparing this measure with other measures including current concurrency by characteristics of the participants
b. By evaluating possible biases of recall in partnerships and start dates
3. Use of the three most recent partners is enough to get a good estimate
a. By comparing results restricted to three most recent partners to results using all partners
4. Short duration (especially one-off) partnerships are relatively unimportant
a. By comparing concurrency measures when one-off partnerships are excluded
b. By assessing the correlation between duration of partnership and number of sex acts and between duration and concurrency
In addition, the association between number of sex acts and concurrency was investigated using the information on the total number of sex acts reported for each partnership, and the reported concurrency for that partnership. For marital partnerships recency of last sex was used as a proxy for frequency.
Of 8232 women and 7338 men aged 15–59 resident in the surveillance area, 7245 women and 5725 men were found and seen; 6825 women and 5283 men agreed to be interviewed about their sexual behaviour, and 6796 women and 5253 men were interviewed. Compared to those eligible, the response rate was 83% for women and 72% for men, and varied little by age.
Marital and nonmarital partners
The demographic characteristics of the population are shown in Table 2. Men started sex and married at an older age than women . By the older age groups almost all men and women had been married at least once. Among those who were currently married, 15% of men and 26% of women were in polygynous marriages. Remarriage was more common for men than for women, but in the age group 35–59, 43% of women had had at least two spouses, and 10% had had at least three.
The total lifetime number of partners was higher for men than for women, both marital and nonmarital partners, with only 2% of women but 28% of men reporting more than four partners. Similarly, only 1% of women but 19% of men reported sex with more than one partner (including spouses) in the past year.
Table 3 shows the different types of concurrent partnership reported by ever sexually active men and women, both their own concurrency, and that of their partners. Very few women reported own concurrency: 1.7% overall. Concurrency by at least one of their female partners was reported by 5.8% of men, with another 30% unknown. In contrast, 24% of men reported own concurrency, of whom nearly half were in active polygynous relationships. 24% of women reported that at least one of their partners had other partners, with another 27% unknown. Combining the reports of concurrency about themselves and their partners, as an estimate of sex within a connected network, 25% of both men and women who had ever been sexually active were reported to be in such networks, with a similar number unknown, whether reported by men or by women.
Comparison of measures of concurrency
Since women had few concurrent partnerships, further exploration of alternative measures of concurrency were limited to men. In line with UNAIDS recommendations , the total number of men in the survey was used as the denominator (not just sexually active men; Table 4).
All measures based on overlapping dates gave lower estimates of concurrency than self-reports. For example, even the broadest estimate of concurrency based on dates was 16.7% overall, compared to 19.2% as self-reported. Reclassifying possible ties in dates as nonoverlapping reduced the estimate very slightly. Using information only on the last three partners made no difference to the results. Excluding one-off partnerships reduced the estimate further to 15.1%. Using the UNAIDS definition for point prevalence 6 months ago gave an estimate of 12.0%, similar to the proportion with current ongoing partnerships (11.5%).
Differences between the different concurrency measures were much more marked in younger age groups and in unmarried individuals (Table 4). For example, concurrency based on dates in unmarried men aged 15–24 was only 61% of that self-reported (6.0 vs. 9.9%), whereas in married men aged 35–59 it was 97% of that self-reported (23.8 vs. 24.6%). Concurrency using the UNAIDS definition was even more sensitive to marriage. This point prevalence measure is expected to be lower than the 12-month measure, but was only 21% of the maximum 12-month estimate in unmarried individuals (2.3 vs. 11.1%), compared to 76% in married men (18.7 vs. 24.7%).
Almost all of the men with two or more partners in the past 12 months self-reported concurrency during this period (Fig. 1a). Using dates, a much smaller proportion had overlapping partnerships, except for married men and unmarried men with four or more partners. The two point prevalence measures (UNAIDS measure and current concurrency) gave lower estimates, but refer to different time periods than the number of partners.
Although most of the discrepancy between self-reported concurrency and that based on dates was due to failure to identify concurrent partnerships using dates, 13 individuals had overlapping partnerships identified by dates but denied concurrency. In contrast, 139 men self-reported concurrency when asked directly but had no overlapping dates identified even using the broad definition. Of these, 33 only reported one sex partner in the past 12 months.
Although the UNAIDS estimate and the current concurrency estimate gave similar proportions overall, different individuals reported concurrency in the different periods: 106 were currently concurrent but not 6 months ago, and 133 were concurrent 6 months ago but not currently.
Partnership duration, start dates and sex acts
Since date of first sex (distinct from date of marriage) was not recorded for marital partners, examination of the association of partnership duration and start dates was restricted to nonmarital partners. 94% of these partners were described as ‘girlfriends’. Using the UNAIDS definition, partnerships would only be included if they started 6 months or more before, and one-off partnerships only if they took place exactly 6 months before. If reporting is unbiased the start dates of partnerships should be evenly distributed over the year (excluding those that started 12 months or more ago). Figure 1b shows the distribution of start dates for all (on-going and finished) nonmarital partnerships that started less than 12 months ago. There is a decline in the number reported, month by month going back in time, with some evidence of heaping at 6 months. The figure also shows the trend in start dates for noncurrent partnerships with one-off sex. Again, this shows a decline, but only accounts for a small proportion of partnerships. Comparing the number of partnerships which started 6 months ago with the number starting within the past month, the ratio was 77% for all partnerships (96/125) and 57% for one-off partnerships (12/21).
The number of sexual acts increased with duration of partnership, but many long-duration partnerships had few sex acts: 45% of those lasting 6 months and 26% of those lasting 1 year or more had had only 2–4 sex acts (Fig. 1c). Overall 55% of nonmarital partnerships were reported to be concurrent, and this varied little by duration of partnership (not shown).
Nonmarital partnerships that were concurrent had a higher number of sex acts than those that were not concurrent (P < 0.001; Fig. 1d). Since this could be due to longer-duration partnerships having more sex acts and giving more opportunity for concurrency, the analysis was repeated restricted to partnerships that had lasted less than 1 month. The results were similar (P < 0.001). The same pattern was seen when restricted to unmarried men (not shown). For marital partnerships there was a slight trend to most recent sex being longer ago in partnerships with concurrency (Fig. 1d; P = 0.02).
In this population, period prevalence measures relying on recall of dates gave lower estimates of concurrency than those that asked directly about concurrent partnerships for each partnership. This was partly due to failure to include partners at all (as reported in South Africa)  and probably also due to failure to recall dates accurately. As shown in Fig. 1b, nonmarital partnerships were much more likely to be mentioned if they started recently, and this was not restricted to partnerships with one-off sex. It is also possible that the direct questioning included some concurrent partnerships that actually happened more than a year ago. Whereas it is possible that men and women who were missed by this study had different levels of concurrency from those included, the comparisons of the different measures of concurrency were made on the same individuals so remain valid.
The high marriage and remarriage levels and polygyny increased the stability of the measures: there was considerably more underestimation when using the date-based measures among unmarried men. This reflects the tendency of the date-based measures to include long-term relationships more reliably than short-term ones, especially when limited to a ‘point’ prevalence. However, exclusion of one-off relationships only reduced the 12-month estimate of concurrency based on dates by 9% (Table 4). A rationale for the greater weight given to long-term than short-term partnerships is that there is more opportunity for transmission . However, as shown in Fig. 1c, a surprisingly high proportion of long-term nonmarital partnerships had few sexual acts. It is also possible that those short-term partnerships that are missed by some measures – including those partnerships that are not mentioned at all in the partner listing – are those that are particularly likely to be with an HIV-infected partner with multiple other partners. In this study 94% of nonmarital partners in the past 12 months were described as girlfriends.
In this analysis the rules suggested in the UNAIDS recommendation to exclude ties, and to collect information only on the last three partners made little difference to the results. This would not necessarily hold if there were higher rates of partner change, though concurrency increases with number of partners (Fig. 1a).
Current concurrency is potentially biased by the interviewees’ assessment of whether relationships are ongoing (which is why the 6 months UNAIDS measure was proposed) and in this study it was not asked directly. Given the evidence that more distant partnerships are not well recalled (Fig. 1b) it is a potentially useful measure. A study elsewhere in Malawi also found better concordance of reporting in more recent relationships . Overall the prevalence of current concurrency in our study was similar to the 6 months UNAIDS estimate. Since a current concurrency measure need not rely on dates, it is easier to collect, but if asked directly, different people may interpret ‘currently’ being in more than one sexual relationship differently.
In this study we also asked individuals whether they thought that their partner had other partners. Men reported that a higher proportion of their partners were in concurrent relationships than was reported by women about themselves (Table 3). This does not necessarily mean that the women under-reported concurrency: the women the men were referring to may have been those not included in the survey. For example, we have previously shown that sex workers seem not to have been included – either because they were nonresident, so not eligible, or because they were not found or refused . Women's estimates of their partners’ concurrency were similar to those reported by the men about themselves, but the women also responded ‘Don’t know’ frequently. It is possible that direct reporting also underestimated concurrency. Combining the reports about individuals and their partners, at least 25% of individuals who had ever been sexually active were linked into wider sexual networks within the past 12 months.
One of the arguments against the importance of concurrency in the spread of HIV is that concurrent partnerships would tend to have fewer sex acts, which has been termed ‘coital dilution’ . This can be nonprimary partnerships having fewer sex acts, and/or concurrency reducing the number of sex acts in the primary relationship. As shown in Fig. 1(c, d), many nonmarital partnerships had few episodes of sex, and we have also found this in in-depth interviews (unpublished results). However, in this setting sexual frequency was not lower with concurrency: sex was just as or more frequent in nonmarital relationships in which the man also reported a concurrent partner. This may be because those with a higher sexual drive tend both to have more sex within each relationship and more concurrent partners. Sex in marriage was slightly less likely to be recent if the relationship was concurrent.
In common with other studies, we have shown that men (as well as women) may under-report partnerships [12,14,19], and that this is more likely for partnerships that were longer ago (Fig. 1b). The high marriage and polygyny rates in this population gave some stability to measurement of concurrency, but among the unmarried, undercounting was common. Estimates are likely to be more biased in populations with lower marriage rates and without polygyny. Recall of dates is difficult, time-consuming, and inaccurate. The theoretical advantages of measuring concurrency 6 months ago are outweighed by the practical advantages of measuring current concurrency, and self-reported concurrency for each partnership. Current concurrency should be asked about, and a standard definition of current concurrency is needed.
The study was designed by J.G., A.C., N.F. and conducted by A.D., N.K., A.M., A.C., and F.P. It was analysed by J.G. with contributions from S.F. The first draft was written by J.G. All authors contributed to the final draft.
We thank the Government of the Republic of Malawi for their interest in this Project and the National Health Sciences Research Committee of Malawi for permission to publish the paper.
We thank Michel Caraël and Basia Zaba for comments on an earlier draft.
Source of funding: The Wellcome Trust.
Conflicts of interest
There are no conflicts of interest.
1. Buve A, Carael M, Hayes RJ, Auvert B, Ferry B, Robinson NJ, et al. The multicentre study on factors determining the differential spread of HIV in four African cities: summary and conclusions. AIDS 2001; 15 (Suppl 4):S127–S131.
2. Watts CH, May RM. The influence of concurrent partnerships on the dynamics of HIV/AIDS. Math Biosci 1992; 108:89–104.
3. Hudson CP. Concurrent partnerships could cause AIDS epidemics. Int J STD AIDS 1993; 4:249–253.
4. Morris M, Kretzschmar M. Concurrent partnerships and the spread of HIV. AIDS 1997; 11:641–648.
5. Eaton JW, Hallett TB, Garnett GP. Concurrent sexual partnerships and primary HIV infection: a critical interaction. AIDS Behav 2011; 15:687–692.
6. Lurie MN, Rosenthal S. Concurrent partnerships as a driver of the HIV Epidemic in sub-Saharan Africa? The evidence is limited. AIDS Behav 2010; 14:17–24.
7. Sawers L, Stillwaggon E. Concurrent sexual partnerships do not explain the HIV epidemics in Africa: a systematic review of the evidence. J Int AIDS Soc 2010; 13:34.
8. Tanser F, Barnighausen T, Hund L, Garnett GP, McGrath N, Newell ML. Effect of concurrent sexual partnerships on rate of new HIV infections in a high-prevalence, rural South African population: a cohort study. Lancet 2011; 378:247–255.
9. UNAIDS Reference working group on estimates, modelling and projections; working group on measuring concurrent sexual partnerships. HIV: consensus indicators are needed for concurrency.Lancet 2010; 375:621–622.
10. Morris M, Epstein H, Wawer M. Timing is everything: international variations in historical sexual partnership concurrency and HIV prevalence. PLoS ONE 2010; 5:e14092.
11. Nelson SJ, Manhart LE, Gorbach PM, Martin DH, Stoner BP, Aral SO, et al. Measuring sex partner concurrency: it's what's missing that counts. Sex Transm Dis 2007; 34:801–807.
12. Maughan-Brown B, Venkataramani AS. Measuring concurrent partnerships: potential for underestimation in UNAIDS recommended method. AIDS 2011; 25:1549–1551.
13. Reniers G, Watkins S. Polygyny and the spread of HIV in sub-Saharan Africa: a case of benign concurrency. AIDS 2010; 24:299–307.
14. Helleringer S, Kohler HP, Kalilani-Phiri L, Mkandawire J, Armbruster B. The reliability of sexual partnership histories: implications for the measurement of partnership concurrency during surveys. AIDS 2011; 25:503–511.
15. Buve A, Lagarde E, Carael M, Rutenberg N, Ferry B, Glynn JR, et al. Interpreting sexual behaviour data: validity issues in the multicentre study on factors determining the differential spread of HIV in four African cities. AIDS 2001; 15 (Suppl 4):S117–S126.
16. Langhaug LF, Sherr L, Cowan FM. How to improve the validity of sexual behaviour reporting: systematic review of questionnaire delivery modes in developing countries. Trop Med Int Health 2010; 15:362–381.
17. Jahn A, Crampin AC, Glynn JR, Mwinuka VM, Mwaiyeghele E, Mwafilaso J, et al. Evaluation of a village-informant driven demographic surveillance system. Demographic Res 2007; 16:219–248.
18. Glynn JR, Kayuni N, Floyd S, Banda E, Francis-Chizororo M, Tanton C, et al.Age at menarche, schooling, and sexual debut in northern Malawi.PLoS ONE 2010; 5:e15334.
19. Glynn JR, Kayuni N, Banda E, Parrott F, Floyd S, Francis-Chizororo M, et al. Assessing the validity of sexual behaviour reports in a whole population survey in rural Malawi. PLoS ONE 2011; 6:e22840.
20. Sawers L, Isaac AG, Stillwaggon E. HIV and concurrent sexual partnerships: modelling the role of coital dilution.J Int AIDS Soc 2011; 14:44.
© 2012 Lippincott Williams & Wilkins, Inc.