Conceived and designed the experiments: PH NH BSM LA. Performed the experiments: PH LA. Analyzed the data: PH LA. Contributed reagents/materials/analysis tools: PH NH BM LA. Wrote the paper: Wrote the first draft and revision of the manuscript: PH NH. Provided revisions and comments to the first and revised manuscript: BM LA.
The authors have declared that no competing interests exist.
To date, male circumcision prevalence has been estimated using surveys of men self-reporting their circumcision status. HIV prevention trials and observational studies involving female participants also collect data on partners' circumcision status as a risk factor for HIV/STIs. A number of studies indicate that reports of circumcision status may be inaccurate. This study assessed different methods for improving self- and partner reporting of circumcision status.
The study was conducted in urban and rural Zambia and urban Swaziland. Men (N = 1264) aged 18–50 and their female partners (N = 1264), and boys (N = 840) aged 13–17 were enrolled. Participants were recruited from HIV counseling and testing sites, health centers, and surrounding communities. The study experimentally assessed methods for improving the reporting of circumcision status, including: a) a simple description of circumcision, b) a detailed description of circumcision, c) an illustration of a circumcised and uncircumcised penis, and d) computerized self-interviewing. Self-reports were compared to visual examination. For men, the error in reporting was largely unidirectional: uncircumcised men more often reported they were circumcised (2–7%), depending on setting. Fewer circumcised men misrepresented their status (0.05–5%). Misreporting by women was significantly higher (11–15%), with the error in both directions. A sizable number of women reported that they did not know their partner's circumcision status (3–8%). Computerized interviewing did not improve accuracy. Providing an illustration, particularly for illiterate participants, significantly improved reporting of circumcision status, decreasing misreporting among illiterate participants from 13% to 10%, although misreporting was not eliminated.
Study results suggest that the prevalence of circumcision may be overestimated in Zambia and Swaziland; the error in reporting is higher among women than among men. Improved reporting when a description or illustration is provided suggests that the source of the error is a lack of understanding of male circumcision.
Randomized controlled trials conducted in Africa have shown that male circumcision (MC) reduces the risk of HIV infection among heterosexual men by about 60 percent
Levels of MC status misreporting have been found to vary depending on the setting and context. For instance, Castellsagué and colleagues
More recently, Westercamp and colleagues
A number of other studies have found substantial discrepancies between self-reports of MC and clinical assessment. Lissouba and colleagues conducted a cross-sectional survey in Orange Farm, South Africa among a random sample of 1,198 men aged 15–49 to examine knowledge, attitudes, and beliefs about MC and to assess the association between MC and HIV
A community randomized trial in Mwanza, Tanzania conducted in 1998–2002, which evaluated the impact of an adolescent sexual health intervention, followed 5,083 adolescent males aged 14–18 and also assessed the reporting of circumcision status at baseline and at 18 months
Few studies have assessed the natural variation in the length of foreskin or the completeness of circumcision. In the Chogoria area in the eastern region of Kenya, Brown and colleagues
To our knowledge, there have been no quantitative assessments of the accuracy of reports by women of partner circumcision status. The study in Mwanza, Tanzania included qualitative interviews with adolescent girls and found that the majority did not know what circumcision was
To improve the accuracy of reporting, researchers need to identify and address the underlying reasons for misreporting, including lack of knowledge, misunderstanding the question, translation accuracy, reporting bias, and physical differences in circumcision. To compensate for inadequate knowledge, researchers have recommended the use of visual aids to improve comprehension
Social desirability bias may also negatively affect reporting. In settings where MC programs are expanding, respondents may feel increasing pressure to present themselves as circumcised in face-to-face interviews. On the other hand, if circumcision is associated with tribes that have minority or lower status or if MC is perceived as traditional or rustic, circumcision may be underreported. To address the issue of social desirability in surveys, studies in the U.S. and elsewhere have found that the use of computerized self-interviews can significantly improve the accuracy of reporting
This study was designed to assess how accurately males report their own status and females report that of their partners' in two countries in which MC is scaling up, Zambia and Swaziland. The study also sought to identify and address the potential causes of misreporting and experimentally evaluate methods for improving the reporting of MC status. The analysis that follows focuses on two of the possible reasons for misreporting: (1) lack of understanding, and (2) reporting bias due to social desirability, and finds that misreporting of MC status is largely due to lack of thorough understanding of circumcision.
The study was reviewed and approved by the Population Council Institutional Review Board (protocol number 454), the University of Zambia Biomedical Research Ethics Committee (reference number 003-05-09) and the Swaziland Scientific and Ethics Committee of the Ministry of Health and Social Welfare (reference MH/599B). Written informed consent was obtained from all adult participants. For participants under the age of 18, assent was obtained after obtaining written informed consent from the parent/guardian. In accordance with local IRB recommendations, male and female participants were given 50 Lilangeni ($6.25) in Swaziland and 20,000 Kwacha ($4.00) in Zambia as compensation for participation.
The study was conducted from July 2009 to May 2010 and was implemented in urban Zambia (Lusaka), urban Swaziland (Mbabane and Matsapha), and rural Zambia (selected wards within 20 kilometers of Lusaka). Men aged 18–50 and their female sexual partners, as well as adolescent boys aged 13–17, were eligible and were recruited from clients visiting HIV counseling and testing (CT) sites, health centers, and from the communities surrounding these clinics. It was not required that couples be married or cohabitate to participate. However, after confirming they were sexual partners, efforts were made to verify the relationship by separately asking each male participant and his female partner a series of questions about the other, e.g., when they first met, how many children they had, what meal they last shared, etc. Interviewers determined the particular questions and how many were necessary to confirm the relationship status of the couple. Couples whose status could not be verified were not permitted to participate; data were not collected on how many couples were not allowed to participate in the study.
A block 6 randomization scheme stratified by site was used to randomly assign participants (1,264 men, 1,264 females, and 840 adolescent males) to one of the three methods of interview. A list of study IDs and random assignment to group was generated for the desired sample size prior to data collection and managed by the site coordinator who assigned participants to study arm in order of intake at enrollment. Each couple in the study was assigned to the same experimental arm; the woman's assignment was based on the man's randomization.
Note that because fieldwork was conducted in phases, with the results of each phase informing the next, the study design was adapted over time and varied by setting.
With the exception of the control group in rural Zambia where
The illustrations used in the study depicted a circumcised and an uncircumcised penis and are shown in
Within the study experimental arms, after the description was read and an illustration shown (if applicable), male participants were asked, “Are you circumcised?” while female participants were asked, “Is the man you came to the clinic with circumcised?” To verify the reported circumcision status of participants, male participants were subsequently asked to undergo a visual examination conducted by a clinical officer or medical doctor who was trained in and had performed MCs in each respective country. Status was categorized as: 1) not circumcised (glans penis completely covered); 2) completely circumcised (glans penis fully exposed); and 3) partially circumcised (glans penis partly covered). In Lusaka and Swaziland, the study and examinations were conducted in a nonclinical HIV VCT site. In rural Zambia, the study and examinations were conducted in district health clinics. To avoid the possibility that prior knowledge of the clinical examination might affect the participant's reporting of his circumcision status, informed consent for the visual examination was requested only after the survey interview was completed.
In order to assess the literacy level of study participants, participants were asked to read a simple sentence. Their ability to read all, part, or none of the sentence was recorded. The sentence used and the method for assessing literacy were based on the approach employed in the Demographic and Health Surveys
Descriptive frequencies were generated by setting and study sample (adolescent male, adult male and female); chi-square tests for differences in proportions tests were used to determine significance when appropriate. Unadjusted and adjusted logistic regression models were used to assess the impact of the experimental method on the misreporting of circumcision status. Odds ratios and adjusted odds ratios are reported for ease of interpretation. A logistic regression on pooled data, combining observations across all sites and sex, was estimated specifically to assess the impact of the illustration on misreporting of MC status. Two different models are presented that assess the impact of the illustration, one was separately run for literate and illiterate participants to provide odds ratios with the conventional interpretation. Another regression model included an interaction term between literacy status and exposure to the illustration to statistically assess the non-linear relationship. This estimation used a computational approach suggested by Norton et al.
The characteristics of the participants are presented by study site in
Urban Zambia | Urban Swaziland | Rural Zambia |
||||||
Characteristic | Males 13–17 N = 438 | Males 18–50 N = 420 | Females 18–50 N = 420 | Males 13–17 N = 402 | Males 18–50 N = 401 | Females 18–50 N = 401 | Males 18–50 N = 443 | Females 18–50 N = 443 |
|
||||||||
Not circumcised | 88 | 77 | — | 86 | 78 | — | 86 | — |
Fully circumcised | 12 | 21 | — | 13 | 20 | — | 9 | — |
Partially circumcised | <1 | 1 | — | 1 | 2 | — | 5 | — |
Don't know (self-report) |
3 | 1 | 8 | <1 | <1 | 5 | <1 | 3 |
|
15.1 | 34.6 | 29.1 | 16.1 | 29.7 | 25.4 | 39.9 | 33.2 |
|
||||||||
None | 3 | 2 | 5 | 2 | 2 | 1 | 3 | 11 |
Some/compl. primary | 47 | 19 | 34 | 28 | 11 | 15 | 37 | 50 |
Some/compl. secondary | 48 | 52 | 45 | 67 | 59 | 60 | 48 | 38 |
Higher | 2 | 27 | 16 | 3 | 29 | 25 | 12 | 1 |
|
||||||||
Never married | 97 | 17 | 15 | 99 | 37 | 49 | 0 | .5 |
Currently married or living together | 3 | 81 | 82 | 1 | 62 | 51 | 100 | 99 |
Separated, divorced, widowed, other | 0 | 3 | 3 | 0 | 1 | 0 | 0 | .5 |
|
||||||||
Ngoni | 11 | 8 | 8 | — | — | — | 9 | 5 |
Tonga | 11 | 10 | 11 | — | — | — | 35 | 26 |
Bemba | 25 | 29 | 29 | — | — | — | 12 | 8 |
Swazi | — | — | — | 100 | 98 | 98 | — | — |
Other | 53 | 52 | 52 | 0 | 2 | 2 | 44 | 61 |
|
||||||||
Christian | 99 | 97 | 99 | 95 | 95 | 99 | 100 | 100 |
Other | 1 | 3 | 1 | 5 | 5 | 1 | 0 | 0 |
|
||||||||
Cannot read | 13 | 6 | 19 | 4 | 4 | 2 | 16 | 32 |
Able to read only part of sentence | 16 | 5 | 14 | 1 | 1 | 3 | 7 | 15 |
Able to read whole sentence | 71 | 89 | 67 | 94 | 95 | 95 | 77 | 53 |
|
73 | 90 | 82 | 93 | 96 | 93 | 90 | 88 |
|
33 | 62 | 53 | 66 | 76 | 77 | 82 | 67 |
|
12 | 92 | 82 | 21 | 94 | 94 | 83 | 68 |
Adolescent sample was dropped in rural Zambia due to insufficient differences observed in urban Zambia and Swaziland.
Don't know and partially circumcised cases dropped from analysis of misreporting of MC status.
Defined as answering correctly that reducing partners, using condoms, and abstinence are HIV prevention methods; that a person cannot become infected by sharing food with someone who has AIDS; and that it is possible for a person who looks healthy to have AIDS
Although adolescents in Lusaka were slightly more likely to report not knowing their MC status, most adult men in these settings provided a response to the question of circumcision status. That said, the prevalence of “don't know” responses among men and boys in Lusaka was significantly higher in ACASI interviews than in FTFIs (5% versus <1%, p<.01), likely because respondents were less willing to reveal ignorance of status in the FTFIs. Women were much more likely to report that they did not know the MC status of their partner in all three settings. Further, “don't know” responses were significantly more prevalent in ACASI than in FTFIs in urban Swaziland (9% versus 3%, p<.01) and rural Zambia (6% versus 2%, p<.05)); for Lusaka, the data are directionally consistent but not significant.
The mean age of the adolescent sample was 15.1 years in Lusaka and 16.1 years in Swaziland. For adult men, the mean ages were 34.6, 29.7, and 39.9 in urban Zambia, urban Swaziland, and rural Zambia, respectively. The mean ages for women were 29.1, 25.4, and 33.2. As expected, the adult sample was slightly more educated than the adolescent sample, since a number of adolescent males were still attending school and had not completed their education. Males had higher levels of education than females in Zambia, with greater educational parity between the sexes in Swaziland. Almost all adolescent males were unmarried. In both urban Zambia and Swaziland, more than half of the couples were in formal unions; a substantial percentage of discordant reporting of formal union status was observed between males and females in Swaziland. Tribal affiliation varied in Zambia, reflecting the ethnic diversity within Lusaka district. Almost all participants indicated their religious affiliation as Christian in both countries. The ability to read a simple sentence in English was moderately high among males in Zambia, but substantially lower for women, particularly in rural areas. In Swaziland, literacy was nearly universal, but in that country respondents were allowed to choose to read either a sentence in SiSwati or in English for the literacy evaluation, which may account for the higher rates. Self-reported knowledge of MC was quite high in both countries, although relatively lower among adolescent males and among females in urban Zambia. Demonstrated comprehensive knowledge of HIV was higher than that observed in the Zambian 2007 DHS or the Swaziland 2006 DHS, but it is not possible to ascertain if this is due to trends in the indicator or the differences in the representativeness of the samples. Finally, ever use of condoms was high for adults in both countries, although relatively lower in rural Zambia, while adolescents had a substantially lower prevalence of condom use.
As seen in
Males 13–17 | Males 18–50 | |||||
Interviewed | Examined | Refused Exam | Interviewed | Examined | Refused Exam | |
|
438 | 311 | 29% | 420 | 318 | 24% |
|
402 | 381 | 5% | 401 | 371 | 8% |
|
— | — | — | 443 | 439 | 1% |
As can be observed in
The figure further illustrates that misreporting among women is significantly higher than among men and runs in both directions. The highest misreporting (13%) was found among Swazi women who have uncircumcised partners. Further, as indicated in
To assess whether introducing different methods of describing and illustrating MC significantly reduced the prevalence of misreporting after adjusting for potential confounding factors, logistic regression models were estimated. The dependent variable was coded 1 if the participant misreported circumcision status (in either direction) and 0 if reported circumcision status was consistent with the clinician's assessment. All characteristics shown in
Males | Females | |||
OR | AOR | OR | AOR | |
|
N = 615 | N = 591 | N = 293 | N = 284 |
FTFI Simple Description (Ref) | 1.0 | 1.0 | 1.0 | 1.0 |
FTFI SD + Illustration | 1.3 (.53–3.4) | 1.5 |
.91 (.38–2.2) | .83 (.62–1.1) |
ACASI SD + Illustration | 1.7 (.69–4.3) | 1.8 |
.94 (.38–2.4) | .79 (.51–1.2) |
|
N = 734 | N = 700 | N = 343 | N = 332 |
FTFI Detailed Description (Ref) | 1.0 | 1.0 | 1.0 | 1.0 |
FTFI DD + Illustration | .98 (.40–2.4) | 1.0 (.96–1.1) | .81 (.39–1.6) | .80 |
ACASI DD + Illustration | 2.3 |
2.3 |
1.1 (.52–2.2) | 1.1 (.80–1.5) |
|
N = 416 | N = 402 | N = 403 | N = 398 |
FTFI No description (Ref) | 1.0 | 1.0 | 1.0 | 1.0 |
FTFI Detailed Descript | 1.0 (.20–5.2) | .61 |
.77 (.33–1.8) | .63 |
FTFI DD + Illustration | 1.3 (.30–6.1) | 1.1 (.68–1.6) | .78 (.33–1.9) | .91 (.82–1.02) |
p<.10,
p<.05,
p<.01.
Dropped cases in which clinician indicated partial circumcision (n = 44) – see text.
The unadjusted and adjusted odds ratios presented in
Few demographic or other characteristics were found to be consistently significantly associated with misreporting: being older lowered the misreporting for males in both urban and rural Zambia (p<.05), while illiteracy increased misreporting (p<.01) among females in rural Zambia; also, being married increased the odds of misreporting (p<.01) among females in Lusaka (data otherwise not shown).
Although the tools tested in the different experimental arms did not consistently reduce misreporting, one additional step was undertaken to determine whether the illustration decoupled from the interview method reduced misreporting, particularly for those who were not able to read a simple sentence. To investigate this question, the data were combined across all sites and by sex, and two regression models were estimated with study site and sex included as covariates. The first model estimates the impact of the illustration separately for literate and illiterate participants (top panel of
Model 1: Separate model | Literate Participants | Illiterate Participants | ||
OR (N = 2226) | AOR (N = 2197) | OR (N = 544) | AOR (N = 538) | |
|
1.0 (.65–1.6) | 1.1 |
.66 (.32–1.3) | .62 |
ACASI | 1.5 |
1.4 |
1.2 (.55–2.5) | 1.3 |
Study site: Urban Zambia | .52 |
2.3 (.32–16.3) | ||
Study site: Rural Zambia | .28 |
2.1 |
||
Female | 2.5 |
4.6 |
||
Age (continuous) | .98 |
.97 |
||
Attended primary or lower | 1.2 (.48–3.0) | .84 (.31–2.2) | ||
Married or living with partner | 1.3 (.78–2.2) | .42 |
||
Comprehensive HIV Knowledge | .82 (.49–1.4) | .77 |
||
Ever heard of MC | .67 (.39–1.2) | 1.5 (.78–2.7) | ||
Ever used condom | 1.2 (.95–1.5) | 1.4 (.78–2.7) |
Model 2: Interaction Model |
Coefficient | SE | p-value |
Interaction: Illiterate |
−.045 (−.12–−.01) | .01 | p<.01 |
|
|||
Illiterate and No Illustration | .13 | ||
Illiterate and Illustration | .10 | ||
Literate and No Illustration | .06 | ||
Literate and Illustration | .08 |
p<.10,
p<.05,
p<.01;
Statistical computation based on approach by Norton et al.
As seen in the top panel of
The top panel of
The primary objective of this study was to provide evidence-based recommendations for the collection of self-reported data on MC status to researchers and program managers interested in measuring the prevalence of male circumcision in a general or study population. It also sought to inform HIV prevention trials and observational studies involving female participants, which rely on women to identify the circumcision status of their partners. The study assessed various tools for improving the reporting of circumcision status, including a) a simple and a more detailed description of male circumcision, b) illustrations of a circumcised and an uncircumcised penis, and c) computerized self-interviewing technology. Reporting of MC status was validated by visual examination.
A high participant refusal rate for visual examination of circumcision status occurred in the Lusaka study site. The high refusal rate for Lusaka suggests that visual examinations to validate self-reporting of MC status may be difficult to implement in some settings, replicating similar findings elsewhere
Between 2 and 7% of males in the study misreported their circumcision status according to the clinical exam. For males the error in reporting of MC status is largely unidirectional, with uncircumcised men reporting that they are circumcised; few circumcised men misrepresented their MC status. The results of this study suggest that national estimates likely overstate actual MC prevalence. Further, in assessments of the influence of MC on HIV incidence, estimates of the impact of MC are likely to be attenuated given misreporting of MC status. These results demonstrate that inaccurate self-reports of MC status are a concern in Zambia and Swaziland, paralleling findings from other countries (e.g., Weiss et al.
Between 11 and 15% of women inaccurately report the circumcision status of their partners, with the error in reporting in both directions. Clinical trials testing potential HIV prevention technologies and behavioral interventions using partner's MC status to control for confounding may be inaccurate if measurement error in MC status is correlated with the misreporting of other self-reported indicators, e.g., adherence to product use in clinical trials, socioeconomic status, and sexual or other risk behaviors (alcohol and drug use)
The study results indicate that audio computer-assisted self-interviewing (ACASI) did not improve, and likely compromised, the self-reporting of MC status. The poor performance of ACASI suggests that participants felt a greater obligation to respond honestly to an interviewer, implying that social desirability bias was probably not a factor in misreporting circumcision. As MC programs are scaled-up and mass media messaging becomes pervasive, social desirability bias may become more pronounced. The face-to-face interviews also likely provided a greater opportunity for the interviewer and participant to discuss the meaning of male circumcision.
The study found that providing an illustration for illiterate participants improved reporting of MC status: misreporting among illiterate participants declined from 13% without an illustration to 10% when one was provided. Counterintuitive results indicate that misreporting was slightly more common among literate participants when they were given an illustration; although the higher level of misreporting was not as substantial. Moreover, this anomaly disappears when the data from urban Zambia—where a potentially ambiguous illustration was used—are dropped from the analysis. The overall conclusion to be drawn is that for studies that rely on self-reports of MC status detailed descriptions and/or illustrations provide a useful method for improving the reporting of MC status by both males and females, but should be pilot-tested for appropriateness. Note, while this should improve reporting, it will not eliminate misreports of MC status.
There are limitations to this study that should be considered when interpreting results. A key concern is that the sample is not representative of the Zambian and Swazi populations, and therefore caution is needed in extrapolating the data to prevalence estimates of MC in each country. A second consideration is that as MC programs scale up and messages about the benefits of MC reach a larger proportion of people, there may be changes to misreporting: on the one hand, a potential decrease in misreporting resulting from poor comprehension; on the other hand, a potential increase in misreporting because of increased social desirability bias. A final limitation is that the study did not directly address the issue of partial circumcision. Partial circumcisions were rarely observed in the study; however, one of the study clinicians classified circumcisions as partial more often than the other clinicians did. Since only complete circumcisions are thought to effectively reduce HIV infection for men, more research needs to be done to understand the implications of variations in foreskin length. As Weiss
We gratefully acknowledge the contributions of the study coordinators Alfred Adams (Swaziland) and Kelvin Munjile (Zambia) and the data collection team members in each country. The research team is especially grateful to the participants, parents, and guardians who generously gave their time to participate in the study. Special thanks to the staff at the participating health centers for facilitating study implementation. The research team would like to acknowledge collaborating institutions in the MC Partnership for their active technical assistance in implementing the project: Steve Gesuale, Hayden Hawry, John Manda, Nicholas Shiliya (Population Services International - Society for Family Health, Zambia); Jessi Greene (PSI, Swaziland); Marie Stopes International, Zambia; and Jhpiego. We would also like to thank the Zambia Ministry of Health for their support and review of the study findings. In addition significant support was provided by Population Council staff, including Barbara Miller, Samir Souidi, Stan Mierzwa, Michael Vosika and Kavitha Valasa. Finally, we would like to thank the three anonymous reviewers for their thoughtful and constructive comments.