AIDS and Behavior

, Volume 18, Issue 9, pp 1764–1775

Risk Compensation Following Male Circumcision: Results from a Two-Year Prospective Cohort Study of Recently Circumcised and Uncircumcised Men in Nyanza Province, Kenya

Authors

    • Division of Epidemiology and Biostatistics, School of Public HealthUniversity of Illinois at Chicago
  • Kawango Agot
    • Impact Research and Development Organization
  • Walter Jaoko
    • Department of Medical MicrobiologyUniversity of Nairobi
  • Robert C. Bailey
    • Division of Epidemiology and Biostatistics, School of Public HealthUniversity of Illinois at Chicago
Original Paper

DOI: 10.1007/s10461-014-0846-4

Cite this article as:
Westercamp, N., Agot, K., Jaoko, W. et al. AIDS Behav (2014) 18: 1764. doi:10.1007/s10461-014-0846-4

Abstract

We present the results of the first study of longitudinal change in HIV-associated risk behaviors in men before and after circumcision in the context of a population-level voluntary medical male circumcision (VMMC) program. The behaviors of 1,588 newly circumcised men and 1,598 age-matched uncircumcised controls were assessed at baseline, 6, 12, 18 and 24 months of follow-up. Despite the precipitous decline in perception of high HIV risk among circumcised men (30–14 vs. 24–21 % in controls) and increased sexual activity among the youngest participants (18–24 years; p-time < 0.0001, p-group = 0.96), all specific risk behaviors decreased over time similarly in both groups. The proportion of men reporting condom use at last sex increased for both groups, with a greater increase among circumcised men (30 vs. 6 %). We found no evidence of risk compensation in men following circumcision. Concerns about risk compensation should not impede the widespread scale-up of VMMC initiatives.

Keywords

Risk compensationBehavioral disinhibitionMale circumcisionHIV/AIDSSexual behaviorAfrica

Introduction

Three randomized controlled trials (RCTs) in sub-Saharan African populations have demonstrated the efficacy of male circumcision (MC) in reducing the risk of female-to-male HIV transmission by approximately 60 % [14]. Following these results, the United Nations Joint Programme on HIV/AIDS (UNAIDS) and the World Health Organization (WHO) recommended MC as an important additional strategy for the prevention of heterosexually acquired HIV infection in men, and 14 countries in eastern and southern Africa, including Kenya, with high HIV prevalence and low levels of MC were set as priority areas for MC scale-up [4].

Circumcision, when performed by trained practitioners, has been shown to be safe, cost-effective, and acceptable in a variety of non-circumcising communities across Africa [13, 57]. Several modeling studies found that the long-term population-level impact of widespread implementation and scale-up of voluntary medical male circumcision (VMMC) services will result in substantial reductions in HIV incidence for both men and women [812]. As a result, ten of the fourteen high-priority countries are actively engaged in national VMMC program scale-up [1315]. However, questions remain about whether the promotion of VMMC as an HIV prevention intervention will translate into a decline in HIV incidence in the general population.

Risk compensation, also sometimes referred to as behavioral disinhibition [16], is defined as an increase in risky behavior in response to the perceived risk reduction following an intervention. Risk compensation is an important possible mechanism that could negatively impact the effectiveness of VMMC programs [17, 18]. If operating, risk compensation could reduce the protective effect of circumcision against HIV and, if of sufficient magnitude, completely negate the protection [19]. Epidemiological modeling studies suggest that, at the population level, only extreme levels of increased risk behavior will offset the protection offered by circumcision to men [6, 9, 10, 1925]; however, only moderate levels of risk compensation in men could result in increased HIV risk for women, especially in the short term [2022, 24].

The current evidence of risk compensation following MC is limited to hypothetical models and behavioral evaluations and extended follow-up in the RCT populations [13, 2628]. The Rakai trial in Uganda found no consistent evidence of risk compensation during the initial [2] and extended follow-up of the trial cohort, including the newly circumcised control participants [29]. In Orange Farm, South Africa, circumcised participants reported a higher mean number of sexual contacts during 4–12 months and 13–21 months of follow-up [3]. Despite this higher risk behavior, the protective effect of circumcision was not changed after adjustment for sex behavior and was remarkably consistent with results of the two other trials [30]. In Kisumu, Kenya [1], risky behaviors declined in both circumcised and uncircumcised trial participants over time, but differences in the rate of the decline were observed—most notably, in the proportion of men reporting two or more sexual partners. More specifically, the proportion of men reporting two or more partners in the past 6 months declined steadily over the course of the trial among controls, but among circumcised men it dropped in the first 6 months of participation and then remained stable. An in-depth study specifically addressing risk compensation in a subset of Kisumu trial participants found no evidence of increase in sexual risk behavior over 1 year of follow-up based on a composite scale of 18 risk behaviors and differences in laboratory diagnosed STIs [28]. Lastly, an independent non-randomized prospective cohort study conducted in a rural community near Kisumu before trial results were available also found no evidence of risk compensation up to 1 year after the procedure [26].

The absence of risk compensation reported by these studies is encouraging, but should be interpreted with caution. HIV risk perception and associated behavior modification could change considerably in the context of large-scale VMMC implementation in which counseling and follow-up are less intense, the protective effect of MC against HIV is proven and known, and promotional messages are designed to increase demand. The assessment of risk compensation outside the experimental setting has been identified as an operations research priority [5, 31]. However, due to slower than anticipated scale-up, little empirical data on risk compensation related to a wide-scale VMMC implementation has been available to date.

Here we report the results of the first longitudinal study of risk compensation associated with MC in the context of a successful national VMMC initiative. Our main study objective was to assess change in HIV risk perception and sexual risk behavior in men before and after being circumcised, and to compare these newly circumcised men to uncircumcised controls over 24 months of follow-up.

Methods

Participants

By all measures, Kenya’s VMMC program for HIV prevention is the world’s most successful [5, 31]. Started in 2008, the VMMC program was initially focused on the non-circumcising Luo population of Nyanza Province. Nyanza suffers the highest HIV prevalence in Kenya, and contributes approximately one-third of the nation’s new infections [32]. Following just 2 years of program activity over 390,000 circumcisions were performed, 82 % of which were done in Nyanza province alone. As a result, the prevalence of circumcised Luo males in Kisumu, the urban center of the province, rose from just 11 % in 2006 to 38 % in 2011 [3335].

The current study took place in two rural (Nyando, Kisumu West) and one urban (Kisumu East) districts of Nyanza Province. Study information was distributed as posters at nine participating governmental health facilities, by word of mouth, in community outreach, and by community sensitization through chief’s barazas (community meetings) and other local meetings. Men seeking VMMC services at study health facilities were recruited into the circumcision (intervention) group and went through screening, written consenting, enrollment and baseline visit procedures before circumcision. Following baseline study procedures, men proceeded through the normal VMMC process, including surgical informed consent, risk reduction counseling, and the circumcision procedure itself carried out by specialized VMMC surgical teams operating at the respective health facilities. Uncircumcised controls were recruited mainly through community outreach where a group of young men was approached and provided information about MC for HIV prevention. When services were offered, men who chose to be circumcised were invited to enroll in the circumcision group and men who declined the procedure were invited to enroll as controls. Whether recruited through community outreach, word of mouth, personal referrals, or other means, all controls were offered the procedure but decided to remain uncircumcised. Controls were frequency-matched on age and residence (community) to circumcision group participants. To be eligible for participation in either group, men had to be 18–35 years of age, be uncircumcised at enrollment, and reside within the study area with no plans to relocate within the next 2 years. Eligibility was not restricted by HIV status or level of sexual activity. For controls, intention to become circumcised was assessed. If a potential control expressed intent to become circumcised within the next 6 months, he was encouraged not to enroll as a control but to join the circumcision group when scheduled for the procedure. Any participant who crossed over from one study group to the other was requested to continue active participation in the study.

Recruitment began in late November 2008, concurrent with the launch of the VMMC program, and ended in April 2010. Participants were consented in their language of choice (English, Dholuo or Kiswahili) by male research assistants certified in ethical research conduct and trained in ascertainment of MC status through visual examination. Participants received travel/income reimbursement of 200 Kenyan shillings (about $2.50) for each completed study visit. Study follow-up was finished in January 2012. Ethical approval was obtained from the Kenyatta National Hospital/University of Nairobi Ethics and Research Committee (protocol #P336/11/2007) and the University of Illinois at Chicago Institutional Review Board #3 (protocol #2007-0914).

Procedures

Study participants were asked to return to the respective health facility for follow-up at 6, 12, 18, and 24 months after enrollment. At all study visits participants in both groups underwent visual examination to confirm circumcision status, completed the study questionnaire, were encouraged to attend HIV testing and counseling (HTC) services available at the facility-based VCT center at each health facility, and were exposed to HIV educational videos playing in the general waiting bay when returning for a study visit at the respective health facility. No direct risk reduction counseling was provided by study staff during a study visit. The window period for each follow-up was ±3 months, with visits considered missed 3 months after the scheduled study visit date. Tracing procedures were initiated for any participant more than 1 week late for a follow-up visit. In the event of tracing, study procedures were carried out at the participants’ homes, places of employment, or other convenient location in the community.

Study questionnaires were administered through audio computer-assisted self-interview (ACASI) modules, developed in three languages (English, Dholuo, and Kiswahili). An equivalent paper-based questionnaire was used at participant request or in cases of power outages at study facilities. In total, 30 % of questionnaires were completed on paper requiring database entry, and approximately 75 % of these were double-entered for quality control.

Statistical Analyses

Our targeted sample size of 3,200 (1,600 in each group) allowed for the detection of effect sizes between 0.1 and 0.2, corresponding to 5–10 % difference in sexual behaviors between groups. This assumes 2 years of follow-up under constant group effect and under group by linear time interaction scenarios. We allowed for 10 % attrition at each follow-up, 20 % crossover (non-adherence to group self-assignment), and repeated measures correlation of 0.5–0.7 [36]. Significance was considered at p 0.05 with 80 % power and the covariance structure was assumed constant. Sample size and power calculations were done in RMASS2 statistical power analysis program [37].

Behavioral outcomes were assessed in longitudinal analyses with random intercept mixed-effect models and included: having sex in the past 6 months, condom use at last sex, last sex with a casual partner, condom use at last sex if it was with a casual partner, multiple sexual partners (>1) in the past 6 months, multiple partners (>1) within the same 30-day period, and exchanging sex for money or gifts (transactional sex) in the past 6 months. Self-assessed risk perception was determined by asking: “What do you think are your chances of getting HIV/AIDS?” Men choosing “moderate” or “great” were categorized as having high HIV risk perception.

All outcomes were binary, and each outcome model included the sequential visit number (treated as continuous time), circumcision status (group) and the interaction of time and group. Circumcision status at each study visit was treated as time-variant covariate, with baseline status set according to self-selected group assignment at enrollment. Men who were enrolled in the control group but became circumcised during their study participation were considered crossovers to the circumcision group (i.e., circumcised crossovers). Men who were enrolled in the circumcision group, but did not undergo circumcision during their time in the study were considered crossovers to the control group (i.e., uncircumcised crossovers).

We used Pearson χ2 tests for categorical variables and Kolmogorov–Smirnov two-sample test for non-normally distributed continuous variables to detect baseline differences between circumcision and control group participants, between men who did and did not return for follow-up after the baseline visit, and between crossovers and non-crossovers. Sensitivity analyses to determine the effect of missed visits on behavioral outcomes were performed through Poisson and logistic regressions modeling the predictors of missingness, as well as pattern-mixture mixed-effect models to assess differences in outcomes by groups with different missed visit patterns, compared to men with complete follow-up. Fixed covariates were included in mixed-effect models to allow consideration of baseline adjustments to the group effect. Age and demographic variables that significantly differed by study group (i.e., marital status, education, income, ethnicity, and employment status) were considered independently for association with sexual behaviors, as well as adjustments to the group effect. Final model selection was done using backwards elimination, with retention of time, group effect and age in the model regardless of statistical significance and adjustment for other independent predictors if significant at p < 0.05. All analyses were done with SAS version 9.2 using PROC NLMIXED for binary mixed-effect models [38].

Results

Of the 3,627 men who presented for screening, 91 % (3,299/3,627) were eligible for participation and 97 % (3,186/3,299) agreed to participate. By design, our study groups were balanced with 1,588 men initially self-selecting into the intervention (circumcision) group and 1,598 enrolling as controls (Fig. 1).
https://static-content.springer.com/image/art%3A10.1007%2Fs10461-014-0846-4/MediaObjects/10461_2014_846_Fig1_HTML.gif
Fig. 1

Study profile. For each follow-up visit, participants were classified as “no further follow-up” if they were eligible for that study visit but passed the window period and did not return for any subsequent visit. Participants were classified as “missed visit” if they passed the window period for that visit, but returned for later follow-up visits. Participants who were expected for the last study visit, but were unable to return for follow-up due to the termination of the data collection were classified under “truncated participation”

Approximately 5 % of each group interviewed at baseline (intervention: 79/1,588, control: 74/1,598) did not return for any follow-up and were excluded from longitudinal analysis. These participants were less likely to be Luo (p = 0.03), to have ever had sex (p = 0.01), and to have been sexually active in the 6 months prior to interview (p = 0.02) compared to men returning for at least one follow-up visit. Participants missing a study visit could continue study participation at a subsequent visit resulting in a complex study flow (Fig. 1). Basic follow-up rates by study visit were 70 % (6 months), 81 % (12 months), 82 % (18 months), and 84 % (24 months).

Among men enrolled in the control group, 21 % (332/1,598) chose to become circumcised during study follow-up (circumcised crossovers). Among the intervention group, 8 % (133/1,588) did not become circumcised (uncircumcised crossovers). The baseline comparison of men initially selecting into the intervention group and men initially refusing, but later accepting circumcision, offers a comparison between VMMC program early and later adopters. Sixty-five percent (65 %) of later adopters became circumcised within 12 months of enrollment in the study. Compared to early adopters, later adopters tended to be younger (median age 19 vs. 20, p = 0.001), perceived themselves at less risk of HIV (26 vs. 32 %: high risk, p = 0.04), were more likely to have used a condom at last sex (57 vs. 47 %, p = 0.005), and were more likely to have sex that was transactional (30 vs. 22 %, p = 0.03).

A sensitivity analysis comparing models restricting crossovers from consideration showed no significant differences with analyses using the full dataset. All models presented here are based on the full sample with circumcision treated as a time-varying covariate. Additional sensitivity analyses showed that controlling for patterns of missed study visits changed neither the magnitude nor the significance of association between circumcision and behavioral outcomes.

Demographically, men choosing not to become circumcised (controls) at baseline were significantly more likely to be Luo (p < 0.001), less educated (p < 0.001), and more likely to be currently employed (p < 0.001) and married (p < 0.001) (Table 1). Multivariable models reported in Table 2 include these demographic characteristics to control for the baseline differences. Although there were baseline differences in marital status, remarkably the two groups did not differ in any measure of sexual history or HIV risk behavior.
Table 1

Enrollment characteristics of study participants

 

Circumcision group N = 1,588

Control group N = 1,598

p

Age in years (IQR; range)

20 (19–24; 18–35)

20 (19–24; 18–35)

0.08

Ethnic group

  

<0.001

 Luo

1,547 (97 %)

1,585 (99 %)

 

 Other

41 (3 %)

13 (1 %)

 

Educational level

  

<0.001

 Primary and less

367 (23 %)

510 (32 %)

 

 Any secondary or higher

1,221(77 %)

1,088 (68 %)

 

Employment status

  

<0.001

 Employed

421 (27 %)

584 (37 %)

 

 Unemployed

1,167 (73 %)

1,014 (63 %)

 

Marital status

  

<0.001

 Single

1,097 (69 %)

994 (62 %)

 

 Married or living as married

491 (31 %)

604 (38 %)

 

Ever had sex

  

0.13

 Yes

1,382 (87 %)

1,419 (89 %)

 

 No

206 (13 %)

179 (11 %)

 

Age at first sex (years)

16 (15–18; 9–30; 1,380)

16 (15–18; 9–29; 1,417)

0.66

Sexual intercourse in past 6 months (ever sexually active only)

0.54

 Yes

1,032 (75 %)

1,074 (76 %)

 

 No

350 (25 %)

345 (24 %)

 

Number of partners in past 6 months (ever sexually active only)

0.13

 None

350 (25 %)

345 (24 %)

 

 One

502 (36 %)

564 (40 %)

 

 2+

393 (29 %)

398 (28 %)

 

 Unsure/refused to answer

137 (10 %)

112 (8 %)

 

 Number of partners lifetime

3 (2–6; 1–123; 1,193)

3 (2–6; 1–122; 1,270)

0.73

Gave gifts or money to a woman in exchange for sex in the past 6 months

0.14

 Yes

246 (24 %)

284 (26 %)

 

 No

784 (76 %)

790 (74 %)

 

 Refused to answer

2 (0 %)

0 (0 %)

 

Had sex with 2 or more partners in the same 30-day period in the past 6 months

0.36

 Yes

279 (27 %)

261 (24 %)

 

 No

750 (73 %)

810 (76 %)

 

 Refused to answer

3 (0 %)

3 (0 %)

 

Used condom at last time having sexual intercourse (ever sexually active only)

0.52

 Yes

613 (44 %)

660 (47 %)

 

 No

693 (50 %)

694 (49 %)

 

 Unsure/refused to answer

75 (6 %)

65 (4 %)

 

Last sexual intercourse was with regular partner (ever sexually active only)

0.10

 Yes

1,014 (73 %)

1,082 (76 %)

 

 No

282 (20 %)

271 (19 %)

 

 Unsure/refused to answer

86 (5 %)

66 (5 %)

 

Self-perceived chances of getting HIV

 

<0.001

 No chance or small chance

1,118 (70 %)

1,209 (76 %)

 

 Moderate or great chance

470 (30 %)

389 (24 %)

 

Sample sizes vary in questions based on past or recent sexual activity. Data are median (IQR; range) for continuous data, or n (%) for categorical data. P values are based on Kolmogorov–Smirnov two-sample test for non-normally distributed continuous data and Pearson’s χ2 test for categorical data for comparison of circumcision and control group

Table 2

Change in HIV risk perception and sexual behavior among circumcised and uncircumcised men over 24 months of follow-up: results of adjusted mixed-effect models

Covariates

Estimate (β)

SE

p

Self-perception of high or moderate chance of acquiring HIV

 Time (visit number)

−0.054

0.025

0.032

 Group (circumcision status)

0.477

0.121

<0.001

 Time by group interaction

−0.231

0.036

<0.001

 Age (continuous)

0.033

0.010

<0.001

 Employment (yes/no)

−0.204

0.067

0.002

 Marriage (yes/no)

0.159

0.067

0.017

Sexually active in the past 6 months

  

 Time (visit number)

0.133

0.017

<0.001

 Group (circumcision status)

−0.799

0.431

0.064

 Age (continuous)

0.094

0.015

<0.001

 Age by group interaction

0.047

0.020

0.020

 Employment (yes/no)

0.797

0.070

<0.001

 Marriage (yes/no)

1.381

0.072

<0.001

Used condoms last time had sexa

   

 Time (visit number)

0.102

0.024

<0.001

 Group (circumcision status)

−0.376

0.120

0.002

 Time by group interaction

0.140

0.034

<0.001

 Age (continuous)

−0.101

0.010

<0.001

 Education (any secondary or above)

0.696

0.085

<0.001

 Employment (yes/no)

−0.424

0.062

<0.001

 Marriage (yes/no)

−1.036

0.062

<0.001

Last sex was with a casual partnera

  

 Time (visit number)

−0.152

0.020

<0.001

 Group (circumcision status)

0.107

0.071

0.131

 Age (continuous)

−0.023

0.010

0.027

 Marriage (yes/no)

−1.295

0.079

<0.001

Used a condom during last sex with a casual partnerb

 

 Time (visit number)

0.275

0.047

<0.001

 Group (circumcision status)

−0.175

0.147

0.234

 Age (continuous)

0.094

0.024

<0.001

 Education (any secondary or above)

0.740

0.181

<0.001

Sex with ≥2 partners in the same 30-day period in the past 6 monthsc

 Time (visit number)

−0.176

0.021

<0.001

 Group (circumcision status)

0.108

0.077

0.163

 Age (continuous)

0.016

0.010

0.098

Exchanging money or gifts for sex in the past 6 monthsc

 

 Time (visit number)

−0.285

0.023

<0.001

 Group (circumcision status)

−0.100

0.085

0.237

 Age (continuous)

−0.005

0.012

0.664

 Education (any secondary or above)

−0.575

0.103

<0.001

 Employment (yes/no)

0.298

0.081

<0.001

 Marriage (yes/no)

−0.258

0.082

0.002

Two or more partners in the past 6 monthsc

  

 Time (visit number)

−0.141

0.018

<0.001

 Group (circumcision status)

−0.073

0.068

0.285

 Age (continuous)

−0.017

0.009

0.059

 Education (any secondary or above)

−0.222

0.081

0.006

aSexually active men only

bMen reporting last sex with casual partner

cMen sexually active in the past 6 months only

The proportion of men reporting sexual activity in the past 6 months increased equally in both the intervention and control groups over the follow-up period (group effect p = 0.86, time effect p < 0.02, Fig. 2a). However, a significant interaction of study group and age (p = 0.02; Table 2) indicated possible differences in this relationship by age group. Stratifying by age (groups: 18–24 years, 25–29 years, and 30–35 years), we note an equivalent increase (time: p < 0.001) in sexual activity among both circumcised and uncircumcised men in the youngest group (group: p = 0.96); no significant increase in 25–29 year-olds (time: p = 0.46); and an increase in sexual activity in circumcised men only (group: p = 0.03, time: p = 0.04) in the oldest, 30–35 year-old, group (Fig. 2b–d).
https://static-content.springer.com/image/art%3A10.1007%2Fs10461-014-0846-4/MediaObjects/10461_2014_846_Fig2_HTML.gif
Fig. 2

Observed proportions of men reporting sexual activity in the past 6 months, by circumcision and age. Notep(T) is the p value for the linear time trend; p(G) is the p value for the group effect (overall difference between the circumcised and uncircumcised men); and p(G × T) is the p value for the group by time interaction, when applicable. All p values are based on unadjusted analysis. Below the horizontal axis labels are the raw numbers corresponding to the graphically represented proportions (MC circumcised men, C uncircumcised men or controls reporting behavior)

Despite this overall increase in sexual activity, all other sexual risk behaviors declined (Fig. 3b,d–f) and condom use increased (Fig. 3a, c) over 24 months of follow-up. The most dramatic declines were observed in transactional sex in the last 6 months (26–12 %), most recent sex with a casual partner (20–12 %), and having multiple sex partners (within a 30-day window) in the last 6 months (26–16 %). These declines were not associated with circumcision status (Table 2). Overall condom use at last sex, regardless of partner type, increased significantly in both groups (Fig. 3a), but showed more pronounced gains in those circumcised (relative increase of 30 % in circumcised vs. 6 % in uncircumcised; group by time interaction p < 0.001). This increase was more pronounced when restricted to condom use with casual partners, but did not differ by study group (Fig. 3c; Table 2).
https://static-content.springer.com/image/art%3A10.1007%2Fs10461-014-0846-4/MediaObjects/10461_2014_846_Fig3_HTML.gif
Fig. 3

Observed trends of examined behavioral variables by circumcision status over 24 months of follow-up. Note Explanations for the notation and p value calculations are the same as in the legend for Fig. 2

The proportion of men reporting last sex with a casual partner decreased over time in both groups (Fig. 3b; Table 2). Reflecting the relative youth of study participants, the proportion of men becoming married increased significantly over the follow-up period. Adjusted for age, circumcised men had relatively greater increase in proportion married (26 vs. 21 % increase; p < 0.001), but gains in both groups were significant (time: p < 0.001; group by time interaction: p = 0.0002).

At baseline, a greater proportion of men seeking circumcision services (intervention group) considered themselves at high risk of HIV than men in the control group (30 vs. 24 %; p = 0.001, Table 1). Over the 2 years of follow-up, uncircumcised men had relatively stable self-perception of HIV risk (Fig. 4). Men who became circumcised, however, had a precipitous decline in perceived risk from 30 % considering themselves at high risk to just 14 % by study exit (group by time interaction: p = 0.001: Table 2).
https://static-content.springer.com/image/art%3A10.1007%2Fs10461-014-0846-4/MediaObjects/10461_2014_846_Fig4_HTML.gif
Fig. 4

Observed change in risk perception by circumcision status over 24 months of follow-up. Note Explanations for the notation and p-value calculations are the same as in the legend for Fig. 2

Discussion

Risk compensation could significantly reduce the impact of VMMC programs [17, 18] and, if of sufficient magnitude, has the potential to completely negate the protective effect of circumcision against HIV [19]. Research designed to monitor post-circumcision risk compensation over time, in the context of active promotion of VMMC as an HIV prevention strategy, has been set as a high priority [5, 19]. This is the first study to examine longitudinal change in HIV-associated risk behaviors in men before and after circumcision in the context of a large population-level VMMC program. We observed no evidence of behavioral risk compensation over 24 months of follow-up. Further, there is evidence that men exposed to the VMMC program, both as circumcised clients and through informational messages as study controls, meaningfully shifted towards safer behaviors. This behavioral reduction in risk was noted in all sexual risk behaviors examined, including increased condom use. The only behavior that showed an increase over time was sexual activity in the past 6 months, largely driven by the youngest age group (18–24) that had similar increases in both circumcised and uncircumcised men.

That men are likely to increase their risk-taking behavior in response to a perceived reduction in risk following circumcision has formed a prevalent assumption for a wide variety of stakeholders. In addition to the HIV prevention community [4, 17, 18, 39], the potential dangers of risk compensation have been expressed by politicians [40], healthcare providers [41], those involved with international VMMC organizations [42], and HIV advocacy groups [43]. At the community level, MC acceptability and feasibility studies often noted that concerns of risk compensation—expressed as the fear that circumcision will lead to sexual promiscuity, adultery, and decrease gains in condom use—could act as a barrier to community acceptance of the intervention [4447]. The results of our study are not consistent with these concerns or with assumptions of moderate to high risk compensation used in the modeling studies projecting the long-term effect of VMMC on HIV at the population level [9, 10, 19, 24, 25, 48].

Our results do strongly support previous empirical findings of a lack of risk compensation, including research done as part of the MC RCTs [13], in detailed behavioral evaluation of trial participant subgroups [28], in extended trial follow-up [27], and in a non-randomized study before trial conclusion [26]. The consistent lack of risk compensation found to date has been considered inconclusive due to the following limitations: 1) rigorous risk reduction counseling provided as part of trial design that is unrealistic in operational settings; and 2) lack of certainty about the protective effect of MC against HIV when these studies were done. While our study participants may have gained a higher degree of HIV awareness than the general population as a result of greater exposure to HIV education, our findings are not likely to be subject to these limitations, and confirm the lack of significant increases in HIV risk behavior among newly circumcised men in the context of a promoted and visible VMMC program.

Decline in condom use is the most consistently expressed concern regarding VMMC promotion and uptake. Specifically, the perception is that circumcised men, considering themselves armed with an “invisible condom”, will be less inclined to practice, and women less able to negotiate, condom use [18, 21, 49, 50]. In contrast, qualitative research in newly circumcised men has repeatedly revealed the view that circumcised men find condoms easier and more comfortable to use once the foreskin has been removed [51], and a modest increase in condom use was observed in the Kisumu trial [1]. Similarly, no changes in condom use were noted in newly circumcised cohorts in other studies [2, 26, 27, 29]. In our study, we observed that both newly circumcised men and uncircumcised controls increased their condom use as the study progressed, with moderately more significant increase in those circumcised. This suggests that men circumcised through the VMMC program were no less likely, and perhaps more likely, to use a condom when compared to age-matched uncircumcised controls from the same community.

Two cross-sectional random-household surveys completed in 2008 and 2011 assessed population level impacts of the VMMC program in Kisumu. At the population level, HIV-related risk behaviors showed no significant increase, and measures of condom use did increase between 2008 and 2011 [34]. These results are remarkably consistent with our findings and suggest that a wider secular change may explain in part the patterns of decreased risk behaviors found in our study. Additionally, surveys found no association between VMMC uptake and HIV high-risk behaviors, also consistent with our findings [34].

Over twenty percent of men initially declining circumcision (control group) ultimately became circumcised during the 2 years of study follow-up. This expected crossover phenomenon was built into our study design, and can be seen as a reflection of secular increases in VMMC mobilization efforts and overall program success during the course of the study. Compared to crossovers, the earliest adopters tended to be older and perceive themselves at higher risk of HIV, consistent with another study in this region [52]. This suggests that men motivated to early adoption of VMMC may represent a higher risk group. Planners and implementers of VMMC programs should anticipate this and ensure that high-quality HIV counseling is a priority throughout the commencement and often-rapid initial scale-up of services.

Drawing conclusions regarding the mechanisms responsible for our findings is outside of the scope of this study. However, several possibilities can be suggested. Decreases in high-risk behaviors may relate directly to the VMMC counseling provided to clients as part of the integrated HIV prevention package [53]. This would primarily apply to the circumcised men; however, all participants in our study were exposed to some HIV education through their participation. A high acceptance (nearly 90 % in 2012) of HTC services among men undergoing circumcision in Kenya [33] demonstrates the VMMC programs represent a potential avenue to increase access and uptake of HTC, as well as linking those who are HIV-positive to care, in male populations known for low utilization of HTC services [32, 54]. This may be an important additional benefit of VMMC programs in light of growing evidence that HTC can reduce HIV risk [55].

It is also possible that the behavioral changes observed in circumcised men may reflect a form of cognitive dissonance [56]—the psychological state of conflict between attitudes, beliefs or behaviors resulting in realignment to decrease discomfort caused by the conflict—in which men reevaluate their behaviors in light of the personal investment involved in getting circumcised. In our study, men choosing to become circumcised perceived themselves at a greater risk of HIV than men choosing to remain uncircumcised. Taking such a drastic and life-altering step as a genital surgery to reduce the risk of HIV is likely to lead to a behavioral reevaluation, with the previous risky practices coming into a conflict with the decision to become circumcised and changes towards safer behavior to follow. These ideas have been noted in qualitative interviews with newly circumcised men [51, 57], and may have value in leveraging further engagement of newly circumcised men in their sexual/reproductive health.

The differential social pressure for behavioral change possibly experienced by the circumcision group may also partially explain the differences in the reported behavioral trends between the circumcised and uncircumcised men. The decrease in sexual risk behaviors in uncircumcised participants that did not benefit from the counseling provided during the VMMC procedure and did not receive the intervention may reflect either a social desirability bias or a secular change in the population generally exposed to the VMMC campaign, as confirmed by the population-level survey in Kisumu discussed earlier in this section [34]. Such persistent and repeated HIV-related messaging designed for VMMC promotion may act to bring HIV back to the fore and reinvigorate the risk reduction efforts in population as a whole.

There are a number of limitations in this study that should be considered. Both circumcision and control group participants were self-selected to enroll in the study. While we utilized age and residence matching to enhance the comparability of the two groups, it is possible that the motivation to become circumcised represents fundamental differences between our study groups, limiting comparability. Comparing the behaviors and perceptions of newly circumcised men to uncircumcised controls over time is an important part of this study, checking our results against possible secular changes in the community. The control group also enabled us to study the cross-overs and compare the early adopters (circumcision group) and later adopters (cross-overs) of circumcision in terms of the demographic, behavioral and HIV risk characteristics. However, risk compensation is defined as changes in behavior after an intervention due to perceived change in risk and the most important findings presented here are the changes in risky behaviors over time among the circumcised participants, based on comparison of circumcised men before and after circumcision. While our study participants had a considerably lower exposure to risk reduction counseling than men participating in the RCTs of MC, it is possible that they had gained a higher awareness of HIV compared to other men circumcised in the Kenyan VMMC program or other men in general population through (1) the repeated encouragement to attend VCT services; and (2) exposure to HIV educational videos at the health facility waiting bay. While we were unable to track the actual VCT utilization or the extent of the exposure to the HIV educational videos, they were likely attenuated by the sporadic availability of facility-based VCT services and the off-site study visits due to the increasingly intensive tracing efforts as the study follow-up progressed.

Our study started concurrently with the initiation of the VMMC scale-up in Kenya, with study enrollment taking place only during the first year of program activities. Therefore, the circumcised men are likely to be representative of relatively early adopters of MC in Nyanza and may not reflect the experiences of men circumcised later. Given our analysis of study group crossovers, it may be that the HIV risk profile of participants will differ as the VMMC program matures and strives to reach broader participation. All behavior and sexual history information were by self-report, and were subject to social desirability and recall biases. We attempted to limit these biases through the use of computer assisted self-interviewing and study staff trained in sensitive face-to-face interview techniques [5861]. Exposure to HIV and risk reduction education may have also introduced social desirability bias. We attempted to limit this bias by encouraging participants to utilize the third-party facility-based VCT services after the completion of the study procedures, rather than providing the risk reduction counseling as a part of the study. However, exposure to the educational videos at the clinic’s waiting bay may have resulted in the reporting bias or behavioral change. Additionally, as stated earlier, exposure to VCT services and educational videos likely decreased over time as the need to trace participants for follow-up became increasingly more intense as the study progressed. This resulted in the majority of follow-up visits being carried out in the field rather than at the health facility. Another important limitation is that no biological tests that could corroborate the reported behaviors, including tests for STIs and HIV, were done as a part of the study. Lastly, men were followed for 2 years after circumcision/enrollment; longer-term behavior changes were not assessed.

Conclusion

Our large prospective study carried out concurrently with the scale-up of the Kenya national VMMC program found no evidence of risk compensation in circumcised men. To the contrary, both circumcised and uncircumcised men significantly reduced their HIV risk behaviors over 24 months of follow-up. In light of our results and those of previous studies in varying populations [13, 2628, 34], concerns about risk compensation in the context of VMMC programs for HIV prevention should not impede the widespread scale-up of the VMMC services. Previously, modeling of the impact of widespread VMMC programs has included sensitivity analyses of the hypothetical effect of only increases in risk behaviors after circumcision [6, 912, 1925]. It would now be most prudent for models to include scenarios of safer sexual behaviors occurring in the context of VMMC programs, as modeling often drives national policy debate and funding projections [11, 25]. However, in VMMC promotion, as with any partially protective intervention, the educational messages about the effectiveness of the intervention must be carefully balanced with emphasis on continuing overall risk reducing practices.

Acknowledgments

Support for this study was provided by a Grant to FHI360 from the Bill and Melinda Gates Foundation to support the Male Circumcision Consortium, a partnership between FHI360, EngenderHealth, and University of Illinois at Chicago working closely with the Nyanza Reproductive Health Society (Grant #47394). Robert C. Bailey received support from the Chicago Developmental Center for AIDS Research, an NIH funded program (P30 AI 082151). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

We thank all of the participants, without whom this work would not have been possible. We are grateful to Timothy Okeyo, Cosam Ang’awa, Yusto Okembia, Kevine Amolloh, Kelvin Akoth, Kennedy Otieno, Danstan Ochieng’, Victor Odula, George Kidi, David Ang’awa, Evans Otieno, Erik Ogutu, George Ong’eng’a, and Richard Okello for their dedication in recruitment, data collection, tracing, and overall commitment to the study; to Nicholas Obwama and Joseph Abuya for their tireless data entry and cleaning efforts; to Matthew Westercamp for his invaluable input and for reading and editing multiple versions of this manuscript; to Christine L. Mattson for inspiration; and to the entire NRHS staff for their assistance in making this study a success.

Copyright information

© Springer Science+Business Media New York 2014