aJohns Hopkins University, Bloomberg School of Public Health, Baltimore, MD, USA; bColumbia University, Mailman School of Public Health, New York, USA; cRakai Project, Uganda Virus Research Institute, Entebbe, Uganda; dInstitute of Public Health, and eFaculty of Medicine, Makerere University, Kampala, Uganda.
Received: 11 October 2001; accepted: 30 October 2001.
Halperin et al.  questioned whether behavioural confounding played a significant role in the protective effects of male circumcision on HIV acquisition and transmission in our observational studies in Rakai . We attempted to conduct a careful and balanced epidemiological analysis to assess the potential efficacy of circumcision in HIV prevention, after adjustment for differences in religion and behaviour among circumcised and uncircumcised men . Our data strongly suggest that circumcision is protective against male HIV acquisition, and may also be protective against male-to-female HIV transmission. However, we wished to be measured in our conclusions by highlighting the complexity of interpreting these data. We essentially concluded that the magnitude of protective effects afforded by circumcision are difficult to quantify from observational data, and we emphasized the need for randomized trials in order to guide policy and programmes .
Halperin et al.  argued the case for a protective effect of circumcision by citing the findings from our data. We do not disagree with many points made, and believe that the evidence from our data is supportive of this position. However, we feel that one must be cautious in discounting the potential impact of confounding and self-selection bias inherent in observational studies. For example, we were concerned that circumcision and religion are highly correlated, and that in Rakai it is difficult to disentangle the effects of behaviours linked to the culture of Islam from the biological effects of circumcision. Halperin et al.  note that, among circumcised men, the lower HIV incidence in Muslims compared with non-Muslims was not statistically significant [relative risk (RR) 0.63, 95% confidence interval (CI) 0.2–2.7], and suggest that this implies that behaviours specific to Muslim men do not confound the associations between HIV acquisition and circumcision per se. However, they ignore the fact that in the 20–29 year age group, HIV incidence among Muslim men was significantly lower than among circumcised non-Muslims (RR 0.25, CI 0.1–0.8), and this is the age group with the highest HIV incidence. Similarly, Halperin et al.  refer to the absence of circumcision among the Luo of Kenya cited in another paper, but do not refer to the comment by the authors  that ‘uncircumcised status and ethnicity were so closely correlated, it was not possible to independently assess the effects of circumcision and ethnic origin'. We also pointed out that the protective effects of circumcision on male HIV acquisition were not consistent in all sub-groups examined in our study .For example, HIV incidence in circumcised men was comparable to that of uncircumcised men among unmarried adolescents, and these are individuals who are likely to be targeted by prevention programmes. These exceptions need to be considered when weighing the observational evidence, in order to define policy.
Halperin el al.  recommend that intervention programmes be initiated before the completion of randomized trials. We believe this recommendation is unwise, until the safety and efficacy of circumcision are clearly understood. Until then, wide-scale implementation could be deleterious for three important reasons. First, safe circumcision requires substantial investment in training and services, and it is important to ensure that this investment is likely to be cost-effective compared with other HIV prevention strategies, in order to avoid diverting resources from other potentially more useful interventions. Second, circumcision might concievably lead to greater risk behaviours (disinhibition) by engendering false expectations of protection. Such disinhibition could attenuate any potential benefit of circumcision, and may actually result in increased HIV incidence. The behavioural consequences of circumcision thus need to be rigorously assessed. Finally, in large scale programmes, HIV-positive men are unlikely to be excluded, so an assessment of circumcision safety in HIV-infected men and their partners needs to be evaluated.
Randomized trials are the sine qua non of evidence-based medicine. Findings from observational studies may be replicated by trials, but not infrequently observational data can be misleading, and may subsequently be refuted by randomized trials [4,5]. There are numerous examples of observational studies that have over-estimated the effects of interventions relative to the results subsequently reported from carefully controlled trials . We believe that it would be unwise to ignore the historical experience in this regard. We agree with Halperin et al.  that multiple randomized trials of male circumcision for HIV prevention must be conducted in different cultural and epidemiological contexts, and we plan to conduct such a trial in Rakai, Uganda. However, we believe that the promotion of circumcision for HIV control before evidence of efficacy is premature.
Ronald H. Graya
Maria J. Wawerb
Nelson K. Sewankamboe
Fred Wabwire Mangend
1. Halperin D, Weiss HA, Hayes R. et al
. Letter of response to Ronald Gray et al
.: Male circumcision and HIV acquisition and transmission: cohort studies in Rakai, Uganda [Letter]. AIDS 2002, 16: 810–812.
2. Gray RH, Kiwanuka N, Quinn TC. et al
. Male circumcision and HIV acquisition and transmission: cohort studies in Rakai, Uganda. AIDS 2000, 14: 2371–2381.
3. Laverys L, Rakwar JP, Thompson ML. et al
. Effect of circumcision on incidence of human immunodeficiency virus type 1 and other sexually transmitted diseases: a prospective cohort study of trucking company employees in Kenya. J Infect Dis 1999, 180: 330–336.
4. Pettiti DB. Hormone replacement therapy and heart disease prevention: experimentation trumps observation. JAMA 1998, 280: 650–651.
5. Piantadosi S. Clinical trials: a methodological perspective.
Wiley Series in Probability and Statistics. New York: Wiley-Interscience Publication, John Wiley and Sons, Inc.; 1997.
6. Schulz KF, Chalmers I, Hayes RJ, Altman DG. Empirical evidence of bias. Dimensions of methodological quality associated with estimates of treatment effects in controlled trials.
JAMA 1995, 273: 408–412.