Participants and Randomisation Procedure
The study communities comprised six pairs of communities matched by socio-economic type—small town, tea/coffee estate, forestry plantation, roadside trading settlement, and subsistence farming area (two pairs) (). Each community included at least one Government or Mission health centre. It was anticipated that HIV-1 incidence would be similar within each pair of communities. Within each pair, one community was assigned at random (un-blinded coin toss by a Ministry of Health official witnessed by programme and research personnel) to receive the additional intervention and the other to be the control. These procedures were designed to ensure that Mission, non-governmental organisation, and private sector programmes (for details, please refer to the following section) would be distributed evenly between intervention and control sites.
Location of Intervention and Control Communities in Manicaland Province, Eastern Zimbabwe
We assessed the effect of the intervention using results from laboratory tests for HIV-1 infection and questionnaire data collected in the baseline and 3-y follow-up rounds of a population-based, closed-cohort survey. The 12 study communities were enumerated in a phased manner, with paired communities being enumerated consecutively to minimise the effects of any seasonal factors. HIV-1–prevention activities were commenced in each intervention community shortly after completion of the baseline survey in that community. In each community, individuals eligible for the study were identified in the first round using data from household listings prepared in an initial census. All males and females aged 17–54 y and 15–44 y at last birthday (the age groups expected to have the highest incidence of HIV infection), respectively, who had slept in a household in the community for at least four nights in the previous month, and who had also done so at the same time 1 y earlier, were considered eligible for the study. In heterosexually driven HIV-1 epidemics, risk of infection can be correlated amongst marital partners [13
]. Therefore, to maximise statistical power to detect differences in HIV-1 incidence, enrolment was restricted to one randomly selected member per marital group.
Intervention and control communities were to receive standard Government services including basic syndromic STI management, condom distribution from health clinics and Zimbabwe National Family Planning Council outlets, home-based care, and limited HIV/AIDS–focussed IEC activities (e.g., occasional AIDS-awareness meetings and distribution of posters and leaflets). In addition, social marketing of male and female condoms would be provided through an ongoing national programme [14
The intervention comprised targeted and population-level strategies to promote safer sexual behaviour and to improve treatment of STIs that facilitate HIV-1 transmission. The intervention strategies were implemented by two local non-governmental organisations (Family AIDS Caring Trust and the Biomedical Research and Training Institute) and the Zimbabwe Ministry of Health and Child Welfare through an integrated programme of community- and clinic-based activities. Integration of the individual programme components was achieved through the joint involvement of the participating agencies in the planning and implementation of activities and through the inclusion of biomedical and behavioural aspects within each component. The programme design comprised three key components: (1) peer education and condom distribution amongst commercial sex workers and male clients at workplaces and in the general community, supported by income-generating projects; (2) strengthened syndromic management of STI services at local health centres; and (3) open days with HIV/AIDS IEC activities at health centres to promote safer sexual behaviour and to increase the uptake of local STI treatment services.
The peer-education component was based on a model which had been developed by the Project Support Group at the University of Zimbabwe [7
] and which had been widely implemented within Zimbabwe and neighbouring countries. Activities were held weekly at workplaces and at locations within the general community (e.g., beer halls and markets) where casual relationships were most frequently formed [15
]. The target population comprised sex workers and male clients who form a bridge population in HIV transmission [16
] between sex workers and the monogamous (or serial monogamous) majority of women [17
]. It was posited that the high HIV-1 incidence observed amongst young women could be reduced by altering the behaviour of their older male partners whose own behaviour was intrinsically more risky [19
]. The behavioural component would be reinforced in counselling sessions with STI patients and through micro-credit income-generating projects to reduce unmarried women's dependence on commercial sex work. The micro-credit scheme consisted of small interest-free loans repayable over 10 mo, provided to groups and to individuals together with training in small-business management. The targeted activities would be extended to the general population through open days held at local health centres.
Besides providing basic HIV/AIDS information, it was envisaged that programme meetings and activities, by their continuous nature, would sustain high levels of awareness of the risks of HIV transmission and would facilitate renegotiation of community social norms, making safer behaviours easier to adopt. The key messages of the programme were: (1) remain faithful to one regular sexual partner; (2) use condoms consistently with any casual sexual partners; and (3) seek prompt and effective treatment for any STIs.
Syndromic management of STIs at primary healthcare centres was first introduced in Zimbabwe in the 1980s [20
] and formed the basis of STI diagnosis and treatment services at baseline in the intervention and control communities. It was envisaged that these services could be strengthened and made more effective through a programme of regular classroom training and on-site supervision of nursing staff, through the introduction of training in systemic counselling for STI patients, and through the provision of small quantities of treatment drugs to cover delays in routine supplies.
Quality-assurance procedures applied in the intervention communities included pre- and post-training tests for peer educators and, for nursing staff, attending the syndromic STI management and systemic counselling courses, regular on-site supervision (including random spot checks) and training, refresher courses, routine planning meetings and monitoring of service statistics, and quarterly workshops where detailed programme procedures were reviewed and updated. An interim qualitative process evaluation of intervention activities was conducted during the inter-survey period, and a report on the findings was provided to the implementing organisations.
Outcome and Process Measures
The primary outcome of the study was HIV-1 incidence at the community level amongst individuals who were uninfected at baseline. Blood was collected onto Whatman No. 3 filter paper and transported to the Biomedical Research and Training Institute laboratory in Harare. Blood spots were air dried at 4 °C and, for long-term (>1 mo) storage, were kept at −20 °C. For baseline studies, blood was eluted into phosphate-buffered saline, and antibodies to HIV were detected using a dipstick dot EIA (ICL-HIV-1/HIV-2 Dipstick, [PATH, http://www.path.org
; produced locally in Thailand]) and a standard protocol [21
]. All positive results and a 10% sample of negative results were confirmed using a plate EIA (Abbott Third-Generation HIV-1/HIV-2 EIA [http://www.abbott.com
] or Genelavia MIXT HIV-1/HIV-2 [Sanofi Diagnostics Pasteur, Marnes La Coquette, France]). At follow-up, a similar protocol was followed. Only the samples from those participants recorded as being HIV seronegative at baseline were tested at follow-up, again using a dot EIA (ICL-HIV-1/HIV-2 Dipstick, [PATH, produced locally in India]). Where seroconversion was indicated, the frozen stored baseline sample was retested to confirm the original negative result using the same dot EIA test. Where the baseline result remained negative, the Abbott EIA test was used to confirm both baseline and follow-up results. The change in place of manufacture of the dot EIA and the exclusive use of Abbott test kits to confirm positive sera at follow-up was due only to changes in the supply of test reagents, and not to perceived changes in sensitivity or specificity [23
]. Apart from the principal investigators (based in Harare, London and Oxford) and those nurses given permission by participants requesting voluntary counselling and testing (VCT), all research personnel remained blind to the HIV-1 status of individual participants.
Secondary outcomes, measured at the community and individual level, were self-reported genital ulcers and urethral or vaginal discharge in the past year (STI cases), STI treatment effectiveness (self-reported cessation of symptoms), indicators of sexual and health-seeking behaviour change, and HIV/AIDS knowledge. The behaviour-change variables assessed were sexual debut, sexual partner change in the past year, non-regular partnerships in the past month, and unprotected sex with regular and casual partners in the past 3 y. The data on sexual partnerships and condom use were collected using the Informal Confidential Voting Interview method for 75% of respondents selected at random in the first round of the survey. This method includes procedures to build rapport, ensure a non-judgemental interview approach, and provide reassurance that there are no right or wrong answers to questions of a personal nature, and uses a simple secret voting-box system to reduce embarrassment and guarantee confidentiality in low-development settings [18
]. Its use has been shown to be associated with greater disclosure of socially proscribed behaviour in the study population [24
Process indicators examined comprised changes in knowledge and psychosocial status and indicators of programme coverage and quality.
Initial sample-size calculations assumed 20% HIV-1 prevalence at baseline, 30% loss to follow-up after 2 y, and 80% power to detect a 40% reduction in HIV-1 incidence in the intervention communities compared with control communities, assuming a background yearly incidence of 2%. Based on six pairs of communities and a co-efficient of variation between communities of 0.15, the required sample size in each community was 1,000. Funding constraints and slower than anticipated implementation of intervention activities led to revisions of the sample size for each community to 800 and the length of follow-up to 3 y, respectively. Assuming a proportionate increase in loss to follow-up to 41%, these arrangements also yielded 80% power to detect a 40% reduction in HIV-1 incidence.
To test the randomisation with small numbers of communities, HIV-1 prevalence, STI history, and socio-demographic characteristics were compared at baseline for study participants in the intervention and control communities, together with uptake of STI treatment and VCT services offered at baseline.
Outcome and process indicators were compared for intervention versus control communities. Analysis of the primary outcome was on an intention-to-treat basis. Incident events and person-years at risk of seroconversion were used to calculate HIV-1 incidence rates and unadjusted and adjusted incidence rate ratios (IRR) with 95% confidence intervals (CIs) for each pair of communities. Adjustment was made for sex, 3-y age group, and community-level baseline HIV prevalence. The overall IRRs (unadjusted and adjusted) were taken to be the geometric means of the IRRs for the six pairs of communities. We calculated 95% CIs for each geometric mean as geometric mean ± 1.96 × standard error of the geometric mean. Paired student t-
tests on the logarithms of the pair-specific IRRs were used to test whether these differed significantly from unity [25
]. The coefficient of variation between communities was calculated based on baseline HIV prevalence using a standard procedure for pair-matched studies [26
Analyses of prevalence for secondary outcome and process variables were conducted separately for male and female respondents seen at both survey rounds by fitting logistic regression models to the individual-level data and adjusting for community pair and, where available, value of variable at baseline.
Since most programme activities were targeted and overall coverage of programme activities was therefore limited, sub-group analyses, adjusted for community pair, were done for HIV-1 incidence and behavioural outcomes to assess the individual-level effects of attendance at programme meetings.
All study participants in the intervention and control communities were offered free VCT for HIV-1, an information sheet on HIV/AIDS, results from a diagnostic test for Trichomonas vaginalis
] (done at baseline only), and free treatment for T. vaginalis
and other STIs from a research nurse. Testing and treatment for T. vaginalis
was provided because the prevalence of other curable STIs was low in the study areas [22
]. Antibodies reactive with T. vaginalis
were detected in DBS eluates following a previously described procedure [27
Written informed consent was sought as a condition of enrolment and continuation in the study. Prior ethical approval was obtained from the Research Council of Zimbabwe, number 02187; the Applied and Qualitative Research Ethics Committee in Oxford, United Kingdom, N97.039; and the UNAIDS Research Ethics Committee, ERC 98/03.