The Effect of Male Circumcision on Risky Sexual Behaviours of Men in Rwanda: A Comparative Analysis of RDHS 2005 and 2014/15

Background Rapid up-take of voluntary medical male circumcision (VMMC) in countries like Rwanda that implemented it as a national HIV prevention strategy since 2008, raised a public health concern of risk compensation. Risk compensation may occur if circumcised men increase their risky sexual behaviours (RSB) because of the awareness of circumcision’s 60% protection from heterosexual HIV/STIs transmission. The aim of this study was to assess the difference in RSBs among adolescent and adult circumcised men before and after the VMMC program implementation in Rwanda Methods Data of 11,037 men aged 15-59 years from Rwanda DHS 2005 and 2014/15 were analyzed using STATA version 13.0. We estimated the prevalence of ever paid sex , extramarital sex , more than 3 total lifetime sex partners, condom use at paid sex, rst sex below age 15 and most recent sex partner being casual/commercial; among circumcised and uncircumcised men and compared between surveys. Association of these behaviours with circumcision status were also determined using bivariate and multivariate logistic regression analysis in separate and in combined datasets. tripled

Rwanda being one of the 14 priority countries earmarked by WHO, included VMMC in its national HIV/AIDS strategic plan of 2008-2012, and began a scale up in 2008 when the prevalence of male circumcision was just 12% (6). The uptake increased and the prevalence of 2014/15 (30%) had almost tripled that of 2005 (7). Although the rate didn't meet up with WHO's 80% target of adolescent and adult population by 2012, it was a considerable effort. The Rwanda MOH trained service providers and VMMC was provided by public and private health facilities. Mass campaigns in rural and urban districts, introduction and scale up of PrePex (bloodless adult male circumcision method) and task-shifting to nurses facilitated access of the services to a large number of people. By the end of 2012, 70% of health facilities had trained staff for circumcision. (8)(9)(10).
Since the implementation of male circumcision for HIV prevention began, there had been public health concerns that following circumcision, adolescent and adult males may increase their risky sexual behaviors from the perception that they had been "fully protected" from HIV and other STIs thus compromising the bene ts of the circumcision. This concept was described as risk compensation or sexual disinhibition. (11)(12)(13)(14)(15)Review of literature (16)(17)(18) showed that the major reason for the increasingly high demand for circumcision in Rwanda was its health bene t of HIV and STI prevention. Also the HIV prevalence in Rwanda had been stable for a decade (2005-2015) (9). Why was it not declining despite all the strategies of HIV prevention? Could it be that the risk was being compensated? Maybe the adolescent and adult men who had undergone circumcision since the implementation of the national VMMC program, including those previously circumcised, had adopted riskier sexual behaviours compared to their uncircumcised counterparts thus compromising the gains of circumcision.
Longitudinal studies on risk compensation were done in Kenya, (11,19,20) Uganda, (14) and Zimbabwe (21). They all found no signi cant differences in risky sexual engagement between circumcised and uncircumcised men following circumcision. However, they suggested a need for the same studies to be done in "real world" settings where participants were not within the con nes of randomized control modalities. They admitted that within the randomized circumcision group, participants were constantly counseled for risk-reduction. This may have mitigated the risk compensation in their study. In the "real world" however, there may not be intense counseling and follow-up but instead promotional messages to increase demand for male circumcision Thus results of a non-experimental operational research design were being awaited. Carrying out such a study in Rwanda was going to be a best-t, being one of the fourteen priority countries that was chosen by WHO/UNAIDS to include VMMC as part of their HIV prevention package since 2008. Therefore, we used "real world" data of men in Rwanda who voluntarily chose to be circumcised in public and/or private health settings. We compared men's data of Rwanda DHS 2005 with that of 2014/15 to assess the differences in RSBs among circumcised and uncircumcised men before and after the VMMC program implementation in Rwanda. The results will add to the evidences that have promoted male circumcision for HIV prevention and provide recommendations for best practices in this eld.

Study Setting
This study was carried out in Rwanda. Geographically located in Central Africa between 1°04' and 2°51' south latitude, and between 28°45' and 31°15' East longitude, Rwanda is a land-locked country, bordered by Burundi in the South; Tanzania in the East; Uganda in the North, and the Democratic Republic of Congo in the West. The borders of Rwanda stretch up to 900 kilometers. The country's administrative division counts for ve provinces: Northern Province, Western Province, Southern Province, Eastern Province and the City of Kigali, which is the fth province. Rwanda is divided into 30 districts (Uturere) which are further subdivided into 416 sectors (Imirenge), 2,148 cells (Utugari) and 14,837 villages (Imidugudu) (NISR, 2014). The village is the smallest politico-administrative entity of the country (MINALOC, 2014).
The population of Rwanda is 10,515,973 residents, of which 52% are women and 48% men based on the 2012 Census. Since the 2002 Census, the population has increased by 2.4 million, which represents an average annual growth rate of 2.6%. The age pyramid of Rwanda has a large base, implying that the majority of the population is young. Around 50% (5.4 million) of the population is under 20. People aged 65 and above account for only 3% of the resident population. The population is essentially young with the mean age being 22.7 years. and mean female age than males (23.5% vs. 21.9%).
The population is largely rural with almost 84 percent of the country's residents living in rural areas. Among the total urban population, 49 percent live in Kigali City, the capital of the country. (7,22)

Study Design and Sampling Procedures
This study was a comparative data analysis of Rwanda Demographic and Health Surveys (RDHS) of 2005 and 2014/15. The RDHS 2005 was conducted before the recommendation, implementation and national scale up of the VMMC program in Rwanda as a strategy of HIV prevention thus serving as a good baseline data. The RDHS 2014/15 was conducted long after the scale-up of the VMMC. The number of years between the surveys allowed for real behaviour changes to be assessed because behaviour changes actually take long to happen.
The prevalence RSB of men aged 15-59 years who were circumcised was estimated from the surveys of 2005 and 2014/15, and then the ratios compared to an unmatched group of those who were uncircumcised during the same time periods.
Both surveys contained nationally representative data on demographic and health characteristics of the total population produced by using a two staged sampling techniques. The rst stage involved creating clusters of the villages (Enumeration Areas; EAs) to represent rural and urban areas. The second stage involved systematic sampling of households, where male interviews were conducted in every second household interviewed. The RDHS 2005 had a total population sample of 16,141 with 4820 completed interviews for men aged 15-59 years and the RDHS 2014/15 had a total representative sample of 12,699 with 6217 completed interviews for men 15-59 years. (7,23). This study was a census of all data of completed interviews in the two surveys; 4820 in 2005 and 6217 in 2014/15, making a total sample size of 11037 men aged 15-59 years.

Data Collection and Variables
Secondary data was used for this study. A registration to the DHS program and an application to use the STATA format of the RDHS male datasets was made. Having been granted the dataset, the variables of interest were extracted in order to form a new dataset that would permit the response of the research questions.
The primary data were collected using validated questionnaires produced by MEASURE DHS. The questionnaires were produced in French and English and translated into Kinyarwanda. Effective trainings were conducted for the interviewers and the questionnaires were pretested before their use in the eld. Detailed interviewer's guide with other instructional manuals and eld supervisors complemented the effectiveness of data collection. (6,7) The dependent variable was circumcision. It was indicated by responding yes or no to a question that asked if the respondent had ever been circumcised. Circumcision was described to them as the complete or partial removal of the foreskin of the penis. Those who responded 'don't know' were considered uncircumcised for this study.
The independent variables included risky sexual behaviours (RSB) among sexually active circumcised and uncircumcised men: Age at rst sex <15: Having the rst sexual intercourse below the age of 15 was considered risky sexual behaviour. This was because the mean age of rst sex in 2005 was 15 and 17 years in 2014/15 survey. The age of rst sexual intercourse was categorized into two, those below 15 years and those 15 years and above.
Ever paid for sex: This was comparable in both survey years. Men responded yes or no to the question if they ever gave money in exchange for sex in their lifetime.
Condom Use at Paid Sex: The only condom use that was comparable from both surveys was condom use every time the men had paid for sex. They responded yes or no to a question that asked if they always used a condom every time they gave money in exchange for sex. Risky sexual behaviour was de ned by those who responded no.
Extramarital Sex: This was de ned as having sex with a non-marital/non-cohabiting partner in the previous 12 months before the surveys. Respondents simply answered yes or no if in the previous 12 months they had had sex with someone who was not their spouse nor a live-in partner.
Having more than three lifetime number of sex partners. This number was arrived at as the average lifetime number was approximately three. So going above the average was considered to be riskier.
Most recent Sex Partner was Casual/Commercial: This indicator was derived from responses to a question that asked respondents to state their relationship with the most recent sex partner. Those whose most recent partners were casual and commercial were grouped together and considered risky sexual behaviour. Married/cohabiting partners, girlfriends and ancée were considered non-risky.
Marital Status: The marital status was regrouped into married/living together, those divorced, separated or widowed formed another category and then those who had never married or lived with a partner.
Level of education: De ned as the highest level of education attained. They were categorized into; no education, primary, secondary and higher. Residence: The residence of the men were considered the way they were originally grouped into rural and urban dwellings.
Wealth Index: Wealth categories were left the way they were originally classi ed; Poorest, Poorer, Middle, Richer and Richest.
Occupation: Occupation was also left the way they were originally classi ed; Cigarette Smoking: the original categories were also used. Respondents answered yes or no to a question that asked if they smoked cigarettes or not.

Data Analysis Procedures
Using STATA version 13.0, variables of interest were rst of all extracted to form two new datasets. The variables were each regrouped and re-named according to the operational de nitions of the study and to make them comparable within and between the years of survey; 2005 and 2014/15. The statistical analyses were done separately and with combined datasets. Descriptive statistics were conducted and presented on tables and gures. To determine the association of circumcision status with risky sexual behaviour indicators, odd ratios of each risky sexual behaviour (ever paid sex, extramarital sex, more than 3 lifetime sex partners, sex before age 15, and most recent sex partner casual/commercial) were estimated for circumcised and uncircumcised men using bivariate logistic regression analysis, at a signi cance level of 0.05. In the multivariate analysis, the associations were adjusted for demographic and other health characteristics (age, marital status, level of education, province, residence, wealth index, occupation and cigarette smoking). Then, after combining the datasets, the difference of the RSBs for all men (circumcised and uncircumcised) were estimated between the survey years by conducting bivariate and multivariate logistic regression analysis as well. The nal regression model of survey year were obtained by adjusting for circumcision and the demographic variables to determine other predictors of RSBs.

Data Management and Ethical Considerations
The les containing the personal information of study participants were not requested from DHS Program when applying for the dataset. They remained secure with the DHS Program so that con dentiality and privacy of personal information was maintained. Moreso, the results have been presented as aggregated data and not as personal information. The raw and analyzed datasets were submitted to the University of Rwanda school of public health library archives.
Meanwhile, during the primary data collection, authorization to conduct the survey was given by the Rwanda Ministry of Health to the national institute for statistics (NISR). Ethics approval was granted by the National Ethics Committee of Rwanda and during data collection, consent was sought to participate in the survey. Special consent was gotten from the parents or guardians of children 0-59 years and for testing anemia and HIV. Personal information was not linked to HIV test samples. Further details can be found in the original RDHS 2005 and 2014/15 reports. (7,23)

Characteristics of Respondents
Data of a total of 11,037 men who completed interviews in both survey years were analyzed. The percentage of men who reported being circumcised in 2014/15 almost tripled those of 2005; from 11% (532) to 29% (1821). In both surveys majority of men were of the younger age groups and the percentage decreased as the population aged. The marital status was distributed in similar ways in 2005 and 2015, with half of the population reporting being married or living with a partner. Also two-thirds of the population were of rural residence (76% in 2005 and 75% in 2014/15). There was an increase in the number of men reporting secondary and higher levels of education (9% and 3% increase respectively). The distribution of wealth was similar in both surveys. (See Table 1) decrease in having more than three lifetime sex partners as well as in having rst sex below 15. Having a recent sex partner that was casual or commercial decreased by 4% while paying for sex without using condoms decreased by 2% (see Table 2).
On the other hand, there were differences in the prevalence of risky sexual behaviours between the circumcised and uncircumcised men in both surveys that did not follow the pattern of the behaviour changes for all men. While paid sex increased among all men in 2015 by 2% (from 5.9% to7.8%), there was a 2% decrease of this behaviour among the circumcised (from 13.2-11.5%). Also while rst sex below sixteen decreased in all men, there was rather a 3% increase (from 35.0-38.3%) in circumcised men; from 35-38.3%. Another change in prevalence noted among circumcised and uncircumcised men was in ever paying for sex without using a condom. Circumcised men who were paying for sex without using condoms almost tripled between 2005 and 2015 (from 12.5-35.3%), while the uncircumcised reduced that risky behaviour by more than a half (from 40.0%-24.1%). Having more than three total lifetime partner also decreased by 14% (from 37.6-23.4%).
In general, the prevalence of the different risky sexual behaviours were higher for the circumcised than the uncircumcised in both survey years except in using condoms when paying for sex and in sexual debut where the reverse is true in 2005.  Table 3. The number of responses on condom use at each paid sex was not su cient to predict a signi cant association between the circumcised and uncircumcised men in both years. (Table 3) Increasing education also signi cantly increased the probability to pay for sex and have more partners but was protective Also religion contributed to extramarital sex and having more lifetime sex partners in 2014/15; being Muslim was 60% more likely to contribute to these behaviours. Lastly, having any kind of occupation was protective of early sexual debut but generally provoked more partners and extramarital sex with the army being the highest. (See Table 5)  -The shaded areas represent factors that were not part of the models for the speci c risky sexual behaviours.   -The shaded areas represent factors that were not part of the models for the speci c risky sexual behaviours.

Discussion
Increases in risky sexual behaviours as a result of decrease in perceived risk in men who are circumcised have been one of the public health concerns since the rapid uptake of male circumcision as an HIV prevention intervention. (4,12,24,25) The research design for the investigation of risk compensation shifted from randomized controlled trials due to possible in uences of the strict routine HIV counseling sessions, undergone by trial participants, on the research results. Results from real world settings have since been sought to show whether risk compensation can occur in men who undergo circumcision under less strict HIV prevention counseling conditions (13,20) and a longer time-span for observation of sexual behavioral changes following circumcision. (11) This study analyzed two nationally representative surveys of Rwandan men aged 15-59 years over a ten-year time-span, to compare men's sexual risk behaviours before and after the national scale up of the VMMC programme in 2008. The hypothesis was that if risk compensation had occurred in Rwanda, the prevalence of risky sexual behaviours among men who reported being circumcised in 2014/15 would be signi cantly higher than those of 2005 and there may be no change among the uncircumcised. Thus the prevalence-ratios of RSBs between the circumcised and uncircumcised men will be higher in 2014/15 than in 2005.
Firstly, the prevalence of most of the risky sexual behaviours studied decreased over time among all men, showing a general shift towards safer behaviours. This decrease was observed in total lifetime number of sex partners, non-condom use, most recent sex with a casual/commercial partner and rst sex below age 15. The only risky behaviour with a general increase was in ever paid sex largely driven by an increase among uncircumcised men. Non-condom use drastically increased (23% increase) among circumcised men in 2014/15. Extramarital sex was more or less stable although higher for the circumcised between surveys. The circumcised men generally had a higher prevalence than the uncircumcised for most of the behaviours in both surveys.
Similar ndings of general shifts towards safer behaviours were obtained in Uganda (26) for multiple partnerships, and extramarital sex. and in Kenya (20) for paid sex, most recent sex with casual partner, more than 2 partners and noncondom use for both groups over a 2-year timespan.
The increase in non-condom use among circumcised men has been one of the greatest public health and community concerns of promoting increasing uptake of MC/VMMC program. (16,(27)(28)(29). Our study found the prevalence of noncondom use at every paid sex to have tripled from 2005-2014/15 whereas that of the uncircumcised decreased by more than a half. These ndings differ with the Kenyan longitudinal study where condom use was reported to increase among the circumcised over time. (20) Other prospective studies however did not nd any signi cant difference in condom use among men who had been circumcised compared to their partners who chose not to be circumcised in Uganda (2,13,14), Kenya (11,20) and Zimbabwe (21) However the recent Ugandan comparative analysis of men's data before and after VMMC showed a signi cant 13% increase in prevalence of this risky behaviour. (26) Moreover, our study found circumcision status to be signi cantly associated with all risky sexual behaviours in both surveys except condom use that had the least number of observations. Circumcision status was positively associated with paying for sex, having more lifetime partners and extramarital sex, while being protective of sexual intercourse before age fteen after controlling for demographic and other factors. The strength of the associations signi cantly reduced from 2005 to 2014/15 rejecting the hypothesis of signi cant increase in risky sexual behaviours among the circumcised and thus no behavioral disinhibition.
Similar results were obtained in Uganda were sex with non-marital partners, multiple sexual partners and condom use were all signi cantly associated with circumcision status in the survey before and after the safe circumcision program. The strength of the associations also decreased with time. (26) Longitudinal studies in Kenya also found decrease in risk behaviours over time in both groups (20) and among the circumcised arm of the cohort (11) Chikutsa (2013) however found no association between risky sexual behaviours and circumcision status in Zimbabwe after about two years of circumcision programme implementation. Gray et al (2012) also found no difference in a posttrial follow up study in Uganda suggesting that longer time-spans are needed to observe behavioural changes.
Our ndings showed that increasing age, education and wealth increased the probability of engaging in more than 3 lifetime partners, paying for sex or having extramarital partners. Having extramarital partners however declined from age 35-39. Being Muslim positively affected having more lifetime partners and extramarital sex. Also, apart from province that positively in uenced having rst sex below 15 years, all other demographic and health factors were protective of this behaviour; all provinces out of Kigali had almost a double risk of having rst sex below age 15 than Kigali.
Similar results have ensued from Malawi (30) where secondary education, age, Muslim religion, and work-related migration in uence number of partners and condom use positively while those with more household wealth were less likely to pay for sex. Being of younger age, having primary education and above and being unmarried was also found to predict risky sexual behaviour among adolescents in Tanzania (31). To also support our nding of no difference in the way age affected rst sex below age fteen in both surveys, an analysis of sexual behaviour data from 59 countries by WHO (32) did not show any universal trends towards earlier sexual intercourse.
In our study, having tested for HIV interestingly increased the risk of risky sexual behaviours except early sexual debut. A qualitative study in Tanzania (33) however found no evidence that adolescent sexual behaviour was affected by HIV/AIDS awareness. On the contrary, a quantitative study (Erick Gong, 2014) found an increase in risky behaviours following an HIV test in East Africa.
That cigarette smoking increased all risky behaviours except starting sex early can be understood as the younger age groups with the most probability of starting sex earlier are restrained by law in Rwanda not to smoke. (34)

Conclusions
The study showed that the prevalence of risky sexual behaviour among circumcised men did not signi cantly increase 7-8 years after the VMMC programme program implementation in Rwanda. On the contrary, there was a general decrease in probabilities to engage in these behaviours among the circumcised and the uncircumcised. Other demographic and health factors played signi cant roles in the outcome of each man's sexual behaviour, other than his circumcision status. We conclude by saying that this study found no risk compensation in the HIV high-risk behaviours of paying for sex, increased lifetime number of sex partners, extramarital sex and having casual partners among men in the era of increasing uptake of male circumcision in Rwanda. However, more subjects on condom use are needed to determine a true relationship between condom use and circumcision. Also investigation of sexual debut in young people out of Kigali are needed to ascertain their compromise by the male circumcision program for HIV prevention.
VMMC packages should intensify counselling on safer sex, particularly regarding correct and consistent use of male and/or female condoms especially at higher risk sex and for men who are circumcised. Also, to intensify HIV counseling to the adolescents and younger adults, 15-24, undergoing circumcision especially in the provinces out of Kigali and in the lower wealth index to know that circumcision does not provide them full protection against HIV/STIs. During the primary data collection, ethics approval was granted by the National Ethics Committee of Rwanda and during data collection, consent was sought to participate in the survey. Special consent was gotten from the parents or guardians of children 0-59 years and for testing anemia and HIV. Personal information was not linked to HIV test samples. Further details can be found in the original RDHS 2005 and 2014/15 reports. (7,23) In the current analysis, the les containing the personal information of study participants were not requested from DHS Program when applying for the dataset. They remained secured with the DHS Program so that con dentiality and privacy of personal information was maintained. More so, the results have been presented as aggregated data and not as personal information.

Consent for Publication
Not applicable

Availability of Data and Materials
The datasets analyzed during the current study are available in the DHS Program repository, [https://dhsprogram.com/data/new-user-registration.cfm]

Competing Interests
The authors declare that they have no competing interests.