Editor -- In January 2001 the Bulletin published my review of the Mwanza and Rakai community-based trials of sexually transmitted disease (STD) interventions, highlighting the role of behavioural factors in the effectiveness of STD treatment (1). My review was written as objectively as possible, though a literature search would show that I have been promoting behavioural intervention since 1988. In addition, my criticism of publications from Rakai has been more forthright than my comments on the Mwanza trial.
Now that the architects of the Mwanza trial have claimed that "most of [Hudson's] assertions are inconsistent with published data" (2), I feel free to express my true feelings about that trial. In my view the Mwanza trial was unethical in its timing, as it should have followed a similar trial of a behavioural intervention. The 1999 detailed publication of the trial makes the point that as late as 1993 "few schools were providing AIDS prevention education" (3).
Could it be that the Mwanza trial was designed to minimize the possibility that detractors would claim the effect was mediated by behavioural change? No sample size calculations were published to justify the interview of only one in eight participants about sexual behaviour. Why did the published papers not use baseline data on prevalence of STDs and HIV to assess bias?
Given that the delay in interviewing was as little as 4 months after the baseline survey in some communities (4), a remarkable 10% of the population had moved away from the area. The proportion was 11% (90/810) of women and 8% (53/688) of men (Yates X=4.62, P=0.03). Furthermore, only 16% (97/599) of females interviewed were aged 15-19 compared with 21% (110/518) of males (Yates X=4.35, P=0.04). In the baseline survey there were more young women than young men, and the strongest risk factor for HIV infection in women was travel outside of the district (5). Could it be that most high-risk women in Mwanza evaded the interviewers?
Furthermore, the significance levels of comparisons of behaviour between intervention and control communities were not given. It is possible that many were in the P=0.05 to P=0.10 range. Other important data were not published: given that a year elapsed between initiation of the first and last pairs in the trial (6), the authors could have looked for a time-dependent intervention effect for both follow-up STD prevalence and HIV incidence. …