In the multicentre study on factors determining the differential spread of HIV in four African cities, we investigated whether high-risk sexual behaviour and/or factors that enhance the transmission of HIV during sexual intercourse are more common in two cities with a high prevalence of HIV (Kisumu, Kenya and Ndola, Zambia) than in two cities with lower prevalence of HIV (Cotonou, Benin and Yaoundé, Cameroon) . We found that reported high-risk sexual behaviour was not more common in Kisumu and Ndola than in Cotonou and Yaoundé . In the high HIV prevalence cities, more men were uncircumcised and the prevalence of herpes simplex virus type 2 (HSV-2) was higher than in the low HIV prevalence cities [3,4]. This led us to conclude that differences in sexual behaviour were out-weighed by differences in factors that enhance the transmission of HIV in explaining the observed differences in prevalence of HIV.
The present paper examines three major, potential sources of bias that could invalidate the conclusions of the study. First, the multicentre study took place in 1997, more than 15 years after the start of the HIV epidemics in the four cities. We were concerned that, in the high HIV prevalence cities, present sexual behaviour would be different (i.e., on average, safer than at the start of the HIV epidemics). A shift towards safer sex could be the result of high awareness about HIV/AIDS and/or increased mortality among those with more risky sexual behaviour. In Kisumu, sexual behaviour change as a result of the HIV/AIDS epidemic was explored with qualitative research methods . The majority of young men and women who were interviewed reported that they had changed their behaviour in the sense that they were now faithful to one partner, had reduced their numbers of sex partners and had started using condoms. However, high rates of partner change were still apparent: 65% of young married men and 33% of young married women reported that they ever had extra-marital relationships. Regarding commercial sexual activity, there was contradictory information coming from different sources. According to some informants, numbers of sex workers were on the increase in Kisumu but, according to others, commercial sexual activity had decreased  (Maina Kahindo, personal communication, 1998). In this paper, changes in sexual behaviour are examined and compared across the four cities, using data from the population-based survey.
A second potential source of bias is the low participation rate of men. In Yaoundé, Kisumu and Ndola, we were not able to interview 18-25% of eligible men; in Cotonou, this percentage was 5% . If the non-participants in Yaoundé all had lower risk behaviour and the non-participants in Kisumu and Ndola all had higher risk behaviour, the comparison of the sexual behaviour data would be severely biased.
Finally, the validity of self-reported sexual behaviour is a recurring problem in sex surveys . In our study, the problem was compounded by the fact that our main objective was to compare sexual behaviour patterns. If in each of the four cities, men and women under-reported their sexual behaviour to the same extent, the comparison between the four cities would still be valid. However, if misreporting was a more serious problem in some cities than in others, the comparison would be biased.
Bias due to changes in sexual behaviour over time
To assess whether there have been any changes in sexual behaviour over time, the following parameters were compared between different age groups: age at first sexual intercourse, age at first marriage, interval between first sexual intercourse and first marriage, and number of sex partners before first marriage. Reported numbers of partners before first marriage were also compared between the four cities, in each age group and sex. In men, the marriage-related parameters were compared between the age groups 30-39 and 40-49 years; in women, the comparison also included the age group 20-29 years. Median age at first sexual intercourse, median age at first marriage and median interval between first sexual intercourse and first marriage were computed, with their 95% confidence intervals, by Kaplan-Meyer survival analysis. Numbers of partners before first marriage were categorized and the statistical significance of the differences was assessed with the χ2 test.
The possible impact of mortality from AIDS on the measured sexual behaviour of the age group 40-49 years was assessed with a simulation exercise. It was assumed that all HIV-infected men aged 30-39 years in Kisumu (51 out of 154) and Ndola (67 out of 169) would have died before age 40. Files were created in which 51 and 67 men were added to the men aged 40-49 years in Kisumu and Ndola, respectively, with the same sexual behaviour as HIV-infected men aged 35-39 years. The latter age group was chosen for the sexual behaviour characteristics because the contrast between HIV-positive and HIV-negative men was bigger than in the age group 40-49 years, yet the overall reported lifetime number of partners was not very different from the age group 40-49 years. Lifetime numbers of partners reported by men aged 40-49 years were then compared across the four cities.
Demographic and Health Surveys (DHS) and other sources of data were consulted to look for trends in condom use in each of the four countries in which the survey took place [7-9].
A simulation exercise was carried out to estimate the maximum likely extent of bias due to non-participation of men in Yaoundé, Kisumu and Ndola. It was assumed that non-participants in Yaoundé were less sexually active than those interviewed, whereas non-participants in Kisumu and Ndola were more sexually active. For each of the three cities, a new file was created. For each age group (15-19, 20-29, 30-39 and 40-49 years), the records of non-participants in Yaoundé were replaced by duplicates of the records of study participants who had the least reported lifetime numbers of sex partners. For instance, in the age group 15-19 years, all 48 non-participants were replaced by records of study participants who denied that they had ever had sex. For the age group 20-29 years, the records of non-participants were replaced by records of men who reported no more than three lifetime sex partners. For Ndola and Kisumu, the records of non-participants were replaced by duplicates of records of study participants who reported more than nine lifetime sex partners. The lifetime number of partners and number of non-spousal partners in the past 12 months were then compared across the three cities.
Validity of sexual behaviour data
Four methods were employed to assess the validity of the sexual behaviour data. First, reports of sexual activity were compared with biological data. More specifically, the prevalence of sexually transmitted infections, including HIV infection, HSV-2 infection, syphilis, gonorrhoea and chlamydial infection, was assessed in men and women 15-24 years old who denied that they had ever had sexual intercourse.
A second method consisted of comparing the reports of spouses. The parameters that were compared included type of marriage (monogamous or polygamous), number of sex acts in the past week and age of the spouse. The latter two parameters were only compared in monogamous couples. For the dichotomous variable 'type of marriage', agreement was tested using the kappa statistic; for the other variables, agreement was tested using the Kendal tau-b test.
The third method consisted of comparing the numbers of non-spousal partners in the past 12 months as reported by men and women. Discrepancies between men's reports and women's reports have been noted in several sex surveys, where men invariably reported more sex partners than women. Several possible explanations other than misreporting have been put forward for these discrepancies: men and women may have sex with partners outside the age range of the survey population or outside the geographical area of the survey; or men have sex with sex workers (or other high-activity women) who may be missed in a survey because they constitute only a small percentage of the female population and/or under-report their sexual activity [10,11]. We computed the ratio of the total number of non-spousal partners reported by all men in the study to the total number of non-spousal partners reported by all women. This crude ratio was adjusted for variations in the response rates. For instance, in Yaoundé, the number of partnerships reported by men was divided by 0.76 (the proportion of eligible men interviewed), and the number reported by women was divided by 0.86 (the proportion of eligible women interviewed). This adjustment assumed that, on average, the numbers of non-spousal partners in the past 12 months was the same for men and women who were interviewed as for men and women who could not be contacted. The following partnerships were excluded from the comparison: partnerships with partners younger than 15 years or older than 49 years (age groups not covered by the survey); partnerships that were initiated outside the city; and partnerships of men with women whom they thought had more than nine other partners in the past 12 months. The latter partnerships were excluded because they were with high-activity women (possibly sex workers) who would be missed in the female population. The ratios in partnerships for men and women were analysed by age group of the female partner.
Finally, selected parameters of sexual behaviour were compared between our survey and DHS . For Cotonou, data were used from the urban stratum of the Atlantic Region from the Benin DHS of 1996. For Yaoundé, the urban data were used of Central/Southern Province from the Cameroon DHS of 1998. Our data from Kisumu were compared with the urban data of Nyanza Province from the Kenya DHS of 1998. The data from Ndola were compared with the urban data of the Copperbelt Province from the Zambia DHS of 1996. The following parameters were compared: proportion of men and women who were never married, proportion of men and women who had their sexual debut before age 15, and number of non-spousal partners in the past 12 months reported by men and women.
All computations were carried out with SPSS 8.0 for Windows (SPSS Inc. 1997, Chicago, Illinois, USA).
Change in sexual behaviour
Among men in Yaoundé, but not in the other cities, there was a statistically significant shift towards younger age at first sexual intercourse in the younger age groups (see Table 1). In all cities except Kisumu, age at first marriage among men was delayed and the interval between first sexual intercourse and first marriage was longer in the younger age groups (data not shown). In Cotonou, men aged 40-49 reported more partners before first marriage than younger men; in the other cities, there was no significant difference in reported numbers of premarital partners. However, numbers of partners before first marriage reported by married men aged 30-39 years are an underestimate because not all men in this age group are married, especially in Cotonou and Yaoundé where 18 and 33%, respectively, of men aged 30-39 years were never married.
Among women, there was a tendency towards later age at first sexual intercourse in the younger age groups in all cities except Cotonou. In all four cities, younger women had married at a later age than older women and the interval between first sex and first marriage was longer in the younger age groups (data not shown). Women younger than 40 years old reported significantly more partners before first marriage than women aged 40-49 years in all cities except Ndola.
When making the comparison across the four cities, in each age group, men in Yaoundé reported the highest numbers of partners before marriage, followed by men in Kisumu (P ≤ 0.001 in all age groups). Women of all age groups in Kisumu reported higher numbers of partners than women in Yaoundé, Cotonou and Ndola (P ≤ 0.001 in all age groups).
The median lifetime number of partners reported by men aged 40-49 years was eight for Cotonou, 20 for Yaoundé, seven for Kisumu and six for Ndola. After adding to the files of Kisumu and Ndola simulated records of HIV-infected men, in order to account for the influence of HIV-related mortality, the median lifetime number of partners was 10 for Kisumu and six for Ndola, and was still less than that reported by men in Yaoundé.
In our survey, condom use reported by men was quite similar in the four cities, but women in the low HIV prevalence cities reported less frequent condom use than women in the high prevalence cities . Comparable data from before our survey are not available, but it is not unreasonable to assume that condom use was very low in all four cities throughout the 1980s. This is suggested by data on the use of condoms as a contraceptive. In 1989, 4% of respondents in Cotonou reported they used condoms for contraception . DHS from the early 1990s found that 1.4% of urban women in Cameroon (1991 DHS), 1.3% of urban women in Kenya (1993 DHS) and 2.6% of urban women in Zambia (1992 DHS) used condoms for contraception . Condom use reported by sex workers showed larger variations between the four cities . The highest rates of condom use were reported by sex workers in Cotonou. In this city, consistent condom use with all clients has increased from 23% in 1993 to 40% in 1998-1999 as a result of interventions targeted at sex workers that were started in 1992 . The high rates of condom use are likely to have contributed to the avoidance of widespread dissemination of HIV from sex workers to the general population . In conclusion, in each of the four cities, condom use has increased substantially, probably starting from the early 1990s, but this increase came too late (and the extent of use is still too low) to have had a major impact on the course of the HIV epidemics, with the possible exception of Cotonou.
The median lifetime number of sex partners reported by men in Yaoundé was 10, that in Kisumu was five, and that in Ndola was four. After replacing non-participants by simulated records, the median lifetime number of sex partners was five in Yaoundé and six in Kisumu and Ndola, but the difference was not statistically significant. Numbers of non-spousal partners in the past 12 months remained higher in Yaoundé (35% had more than one non-spousal partner) than in Kisumu and Ndola (22 and 21%, respectively, had more than one non-spousal partner).
Validity of the sexual behaviour data
The prevalence of any sexually transmitted infection (i.e., HIV, HSV-2, syphilis, gonorrhoea or chlamydial infection, alone or in combination) among those who said that they had never had sex ranged from 0.9% among men in Cotonou to 18% among women in Ndola (see Table 2). The prevalence of HIV infection among men who denied that they had ever had sex was 0% in all cities except Ndola, where it was 4.8%. Among women who denied that they had ever had sex, the prevalence of HIV infection was 0% in Cotonou, 1.1% in Yaoundé, 10.8% in Kisumu and 7.6% in Ndola. At least 50% of these HIV infections occurred in association with another sexually transmitted infection, suggesting that most of these infections were acquired through sexual intercourse.
Information on both members of married couples was available for 305 couples in Cotonou, 209 in Yaoundé, 304 in Kisumu and 299 in Ndola. Married men were asked whether they had more than one spouse, while married women were asked whether their husband had any other spouse beside themselves. Agreement between responses of men and of women was high except for Ndola. The kappa statistic was 0.84 in Cotonou, 0.78 in Yaoundé and 0.86 in Kisumu, but only 0.28 in Ndola. Table 3 presents the comparison of the number of sex acts in the last week as reported by monogamous men and as reported by their wives. Agreement between partners' reports was poor in all cities. It was similar in Cotonou and Yaoundé, but there was less agreement in Kisumu and in Ndola. Agreement on reporting of spouse's age was high in all four cities. The Kendal taub value ranged between 0.84 and 0.92.
After excluding partners aged less than 15 years and over 49 years as well as partners from outside the city and high-activity women, and after adjusting for differences in response rates, the ratio of the number of non-spousal partners in the past 12 months reported by men to the number reported by women was 3.0 for Cotonou, 1.9 for Yaoundé, 2.7 for Kisumu and 3.0 for Ndola (Table 4). When comparing the ratios by age group of the female partners of male respondents or the age group of the female respondents, the ratio was well over 3 in the age group 15-19 years in all four cities (Table 4). This means that the number of female partners aged 15-19 years reported by men aged 15-49 years was more than three times higher than the number of male partners reported by female respondents aged 15-19 years. This ratio decreased in the older age groups. In Yaoundé, Kisumu and Ndola, the ratios for the age groups 20-29 and 30-49 years were quite similar, but they were higher in Cotonou. There were, however, large differences in the percentage of partnerships for which the male respondents could not estimate the age of the partner. This was 1% in Cotonou and Yaoundé, 6% in Kisumu and 21% in Ndola.
Table 5 presents the comparison of selected parameters of sexual behaviour in the DHS and in our study. In the DHS, the proportion of never-married men and women in Yaoundé was smaller than in our survey, while in Ndola it was higher than in our survey. The proportion of men and women who reported that they had their sexual debut before age 15 was higher in the DHS than in our survey in Cotonou and Yaoundé, lower in Kisumu and higher in men in Ndola. In contrast to our study, according to the DHS data, women in the high HIV prevalence cities did not have their sexual debut at an earlier age than women in the low HIV prevalence cities. Unmarried men and women reported more non-spousal partners in the past 12 months in the DHS than in our survey, in all four cities. For the married respondents, the agreement between the DHS data and our data was very good except for Cotonou, where married men reported more non-spousal partners in the DHS than in our survey (data not shown). Although there were important differences between the findings of the DHS and our survey, the DHS confirmed our conclusions regarding differences in rate of partner change between the four cities (i.e., that rates of partner change were not higher in Kisumu and Ndola than in Cotonou and Yaoundé).
We examined three potential sources of bias that could distort the results of the multicentre study on factors determining the differential spread of HIV in four African cities and that could invalidate the main conclusions of the study  (i.e., that differences in sexual behaviour alone cannot explain the observed differences in HIV prevalence between the four cities).
First of all, we were concerned that changes in sexual behaviour over time could have masked any differences in behaviour that existed at the start of the HIV epidemics between the high HIV prevalence cities and the low HIV prevalence cities. Comparison of selected parameters of sexual behaviour between different age groups suggests that there is a tendency for young women to have higher rates of partner change than older women in all cities except Ndola, one of the high HIV prevalence cities. The most striking change in behaviour of men has taken place in Yaoundé, where younger men start sexual activity earlier than older men. The comparison between the four cities of the numbers of pre-marital partners reported by the older age groups suggests that, over the past 20-30 years, rates of partner change for men have always been highest in Yaoundé and for women in Kisumu. The results of the simulation exercise suggest that the differences in rate of partner change in the older age groups cannot be explained away by differential mortality of individuals with higher risk behaviour. Still, one has to be cautious when interpreting these data as they are subject to recall bias. On the contrary, there is no reason why men and women in Kisumu and Ndola should be more 'forgetful' than men and women in Cotonou and Yaoundé. In conclusion, we did not find evidence for a shift towards safer sexual behaviour in the high HIV prevalence cities, as one would have expected, and we believe that, although changes in sexual behaviour have taken place over time, these cannot explain the fact that we did not record more high-risk sexual behaviour in the two high HIV prevalence cities than in the two low HIV prevalence cities.
A second major concern was the low participation rate of men in Yaoundé, Kisumu and Ndola. Among the men who did not participate in the study, very few frankly refused to do so. The majority of non-participants were never found at home despite repeated visits by the study team . There may be differences in sexual behaviour between participants and non-participants, but the results and the conclusions of the multicentre study would be seriously biased if the non-participants in Yaoundé all had low-risk sexual behaviour and the non-participants in Kisumu and Ndola all had high-risk behaviour. We simulated this situation and compared the rates of partner change in the three cities. In the simulation exercise, lifetime number of partners reported by men in Yaoundé was slightly lower than the numbers reported by men in Kisumu and Ndola, but men in Yaoundé still had more non-spousal partners in the past 12 months than men in the other cities. So, even under extreme assumptions of sexual behaviour of non-participants, rates of partner change were similar in the two high HIV prevalence cities (Kisumu and Ndola) and in one of the low HIV prevalence cities (Yaoundé).
Third, we were concerned about the validity of the sexual behaviour data. Several studies have been conducted to test the reliability and validity of sexual behaviour data in a number of African countries [11, 14]. These studies confirmed that there are problems with the validity of survey data on sexual behaviour, but they also found that there are large variations in the validity between different parameters and between different populations . A comparable bias in the data in the four cities would not invalidate the comparison of the sexual behaviour data between the four cities, but differences in the magnitude and direction of the bias could seriously distort the results and the conclusions of the study. We used several methods to assess the validity of the data on sexual behaviour in the four cities.
By testing for various sexually transmitted infections among men and women aged 15-24 years who reported that they had never had sexual intercourse, we could establish that, in all four cities, at least 1-9% of men and 6-18% of women had misreported their sexual activity. It is, however, impossible to get more exact estimates of the proportions of men and women who gave wrong reports. Individuals who misreported their sexual activity may have a different sexual behaviour than those who are more forthcoming about their sexual activity, and their risk of acquiring a sexually transmitted infection may thus be different.
The comparison of the reports given by spouses suggests that misreporting of sexual behaviour may be more of a problem in Kisumu and Ndola than in Cotonou and Yaoundé. However, the validity of spouses' reports on number of sex acts and on the spouses' age may be different from the validity of individuals' reports on non-spousal partnerships and lifetime number of sex partners. In fact, the analyses of the data on contacts with sex workers suggest that men in Cotonou and Ndola under-reported their contacts with sex workers more than in the other two cities .
Numerous studies, from industrialized countries as well as from developing countries, have found that men report more sex partners than women [10,15-17]. In our study, men reported two to three times more partners in the past 12 months than women, even after excluding partners outside the age range of the study population and outside the geographical area of the study, as well as high-activity partners of men. We found a decrease in the ratio of the numbers reported by men to the numbers reported by women in the older age groups. A similar observation has been made in the French sex survey . The most likely explanation for the discrepancy between male and female reports is believed to be under-reporting of partners by women, possibly because they only report the partners that mattered to them. When comparing the ratios between the four study sites, it seems that under-reporting by women was more of a problem in Cotonou than in the other cities. However, in Ndola, the age of the partner was not known for 21% of partnerships reported by men, so the ratios for Ndola may be biased.
Finally, we compared selected parameters from our survey with an external source, Demographic and Health Surveys that were conducted in the four countries around the same time as our study. There were important differences between the findings of the DHS and our survey. The most consistent differences were found in the numbers of non-spousal partners in the past 12 months reported by never-married men and women. In all four cities, respondents reported more non-spousal partners in the DHS than in our survey. The DHS data did, however, confirm our main conclusions regarding the comparison of the sexual behaviour data (i.e., that rates of partner change were not higher in the two high HIV prevalence cities than in the two low HIV prevalence cities).
In conclusion, we found evidence in all four cities that men as well as women misreported their sexual behaviour. Several analyses also suggest that misreporting may have been more of a problem in some cities than in others, but there is only one parameter that was misreported more in the two high HIV prevalence cities than in the two low HIV prevalence cities, i.e., number of sex acts in the past week in married couples. Men under-reported their contacts with sex workers more in Cotonou, one of the low HIV prevalence cities, and Ndola, one of the high HIV prevalence cities, than in Kisumu and Yaoundé. The data on the numbers of non-spousal partners reported by men and women suggest that women in Cotonou and Ndola under-reported their non-spousal partners more than women in the other two cities. Overall, it seems that under-reporting of sexual activity was not more common or more serious in the two high HIV prevalence cities than in the two low HIV prevalence cities. We believe that the main conclusions of the multicentre study still hold and that differences in sexual behaviour by themselves cannot explain the differences in HIV prevalence between Cotonou, Yaoundé, Kisumu and Ndola.
1. Buvé A, Caraël M, Hayes RJ, et al.Multicentre study on factors determining differences in rate of spread of HIV in sub-Saharan Africa: methods and prevalence of HIV infection.AIDS
2001, 15 (suppl 4):
2. Ferry B, Caraël M, Buvé A, et al.Comparison of key parameters of sexual behaviour in four African urban populations with different levels of HIV infection.AIDS
2001, 15 (suppl 4):
3. Auvert B, Buvé A, Lagarde E, et al.Male circumcision and HIV infection in four cities in sub-Saharan Africa.AIDS
2001, 15 (suppl 4):
4. Weiss HA, Buvé A, Robinson NJ, et al.The epidemiology of HSV-2 infection and its association with HIV infection in four urban African populations.AIDS
2001, 15 (suppl 4):
5. Njue CW. Study on youth sexual behaviour.Report on Additional Qualitative Studies in Kisumu.
6. Catania JA, Binson D, Van Der Straten A, Stone V. Methodological research on sexual behaviour in the AIDS era.Ann Rev Sex Res
7. Demographic and Health Surveys. Macro International MEASURE Gateway.
8. Donadje F. Stratégies Masculines de Reproduction et de Planification Familiale au Bénin.
Geneva: Rapport pour I'OMS; 1990.
9. Alary M, Mukenge-Tshibaka L, Anagonou S, et al.Impact of condom promotion and STD control on HIV and STD prevalence in female prostitutes in Cotonou, Benin: observations from the field in the context of a changing prostitution milieu [abstract WePpC1315].
Presented at the XIIIth International AIDS Conference.
Durban, South Africa
, 9-14 July 2000.
10. Lagrange H. Le nombre de partenaires sexuels: les hommes en ont-ils plus que les femmes?Population
11. Dare OO, Cleland JG. Reliability and validity of survey data on sexual behaviour.Health Trans Rev
1994, 4 (suppl):
12. Lagarde E, Auvert B, Chege J, et al.Condom use and its association with HIV/sexually transmitted diseases in four urban communities of sub-Saharan Africa.AIDS
2001, 15 (suppl 4):
13. Morison L, Weiss HA, Buvé A, et al.Commercial sex and the spread of HIV in four cities in sub-Saharan Africa.AIDS
2001, 15 (suppl 4):
14. Konings E, Bantebya G, Caraël M, Bagenda D, Mertens T. Validating population surveys for the measurement of HIV/STD prevention indicators.AIDS
15. Morris M. Telling tails explain the discrepancy in sexual partner reports.Nature
16. Le groupe ACSF. L'activité sexuelle de la population.
In Les Comportements Sexuels en France.
Paris: La documentation Française; 1993.
17. Caraël M. Sexual behaviour.
In Sexual Behaviour and AIDS in the Developing World.
Edited by Cleland J, Ferry B. London: Taylor and Francis, on behalf of the World Health Organisation; 1995.
This publication is sponsored by the Joint United Nations Programme on HIV/AIDS (UNAIDS) and the Agence Nationale de Recherches sur le SIDA (ANRS) Paris, France. The Editors of this supplement wish to acknowledge the referees who provided peer review of the manuscripts.
The study was supported by the following organizations: UNAIDS, Geneva, Switzerland; European Commission, Directorate General XII, Brussels, Belgium; Agence Nationale de Recherches sur le SIDA/Ministère français de la coopération, Paris, France; DFID, London UK; The Rockefeller Foundation, New York, USA; SIDACTION, Paris, France; Fonds voor Wetenschappelijk Onderzoek, Brussels, Belgium; Glaxo Wellcome, London, UK; and BADC, Belgium Development Cooperation, Nairobi, Kenya.