Documenting and explaining the HIV decline in east Zimbabwe: the Manicaland General Population Cohort

Purpose The Manicaland cohort was established to provide robust scientific data on HIV prevalence and incidence, patterns of sexual risk behaviour and the demographic impact of HIV in a sub-Saharan African population subject to a generalised HIV epidemic. The aims were later broadened to include provision of data on the coverage and effectiveness of national HIV control programmes including antiretroviral therapy (ART). Participants General population open cohort located in 12 sites in Manicaland, east Zimbabwe, representing 4 major socioeconomic strata (small towns, agricultural estates, roadside settlements and subsistence farming areas). 9,109 of 11,453 (79.5%) eligible adults (men 17-54 years; women 15–44 years) were recruited in a phased household census between July 1998 and January 2000. Five rounds of follow-up of the prospective household census and the open cohort were conducted at 2-year or 3-year intervals between July 2001 and November 2013. Follow-up rates among surviving residents ranged between 77.0% (over 3 years) and 96.4% (2 years). Findings to date HIV prevalence was 25.1% at baseline and had a substantial demographic impact with 10-fold higher mortality in HIV-infected adults than in uninfected adults and a reduction in the growth rate in the worst affected areas (towns) from 2.9% to 1.0%pa. HIV infection rates have been highest in young adults with earlier commencement of sexual activity and in those with older sexual partners and larger numbers of lifetime partners. HIV prevalence has since fallen to 15.8% and HIV incidence has also declined from 2.1% (1998-2003) to 0.63% (2009-2013) largely due to reduced sexual risk behaviour. HIV-associated mortality fell substantially after 2009 with increased availability of ART. Future plans We plan to extend the cohort to measure the effects on the epidemic of current and future HIV prevention and treatment programmes. Proposals for access to these data and for collaboration are welcome.

► The Manicaland cohort is one of a handful of longrunning, large-scale general population HIV serosurveys conducted in countries in sub-Saharan Africa with widespread epidemics that constitute a key resource for evaluating the populationlevel impact of HIV prevention and treatment programmes. ► The current data span the period 1998 to 2013 during which Zimbabwe experienced one of the largest HIV epidemics in the world and was almost unique in sub-Saharan Africa in achieving a substantial national decline in HIV prevalence largely caused by reductions in sexual risk behaviour. The study data also cover periods prior to, during and following the roll-out of prevention of mother-tochild transmission services (introduced in Zimbabwe from 2002) and antiretroviral treatment services (from 2004 with rapid scale-up from 2009). ► The study data include comprehensive and consistent measurements of trends in HIV prevalence, HIV incidence, HIV-associated and all-cause mortality, sexual risk behaviours, healthseeking behaviours, and in the coverage and effects of national HIV control programmes over time. The study also includes parallel measurement of trends in HIV prevalence among pregnant women attending local antenatal clinics which permits assessment of biases in the primary source of routine HIV surveillance data used by countries and Joint United Nations Programme on HIV/AIDS to produce national and regional HIV estimates. ► Findings from the study are generalisable to Zimbabwe as a whole and data are available on their wider generalisability. ► Limitations of the cohort include the age limit (55 years) for participation, changes in eligibility criteria across rounds, and long intervals (2-3 years) between rounds of follow-up such that short-term migrants may be missed and measurement of some key variables including mortality can be subject to recall and misclassification bias.

Open Access
Imperial College London and the Biomedical Research and Training Institute (BRTI) with funding from the Wellcome Trust. Findings from an earlier study (1993)(1994)(1995)(1996) had shown that HIV was spreading extensively in rural areas of eastern Zimbabwe, and was associated with large increases in mortality. 1 The new cohort was established to provide robust scientific data on HIV prevalence and incidence within a general population sample, on local patterns of sexual behaviour and their role in the spread of HIV, and on the mortality and wider demographic impact of HIV in a range of different settings in Manicaland, Zimbabwe's eastern province.
Following an initial pilot study, 2 in the first two rounds of the cohort study, a two-arm cluster-randomised controlled trial was conducted of a peer education, condom distribution and syndromic management of sexually transmitted infections (STIs) intervention in female sex workers and male clients to reduce the spread of HIV infection. The trial found that this intervention was not effective in reducing HIV incidence within the general population. 3 In subsequent rounds of the cohort survey, the research aims were extended to include investigation of the temporal dynamics of the HIV epidemic, the social determinants of HIV, and the coverage and effectiveness of national HIV control programmes, including antiretroviral therapy (ART) services introduced in the mid-2000s.

cohort descrIPtIon study design and location
The study is designed as a stratified General Population Open Cohort HIV Sero-Survey and is located in three districts (Mutasa, Makoni and Nyanga) of Manicaland province, which runs along Zimbabwe's eastern border with Mozambique ( figure 1A).
To accommodate the two-arm cluster-randomised trial, a stratified design was chosen with six pairs of sites (figure 1B) matched on socioeconomic criteria. Consequently, the Manicaland cohort was drawn from two small towns (Nyanga and Nyazura), four agricultural estates (Katiyo and Eastern Highlands tea estates and Selborne and Sheba forestry plantations), two roadside settlements (Watsomba and Nyabadza/Nyahukwe), and four subsistence farming areas (Bonda, Honde, St. Theresa's and St. Killian's missions).
The central coordinates of the component study locations (rural village markets, estate compounds and urban locations) have been mapped using handheld global positioning system devices ( figure 1C) The study areas are located in the Eastern Highlands region of Zimbabwe (average altitude approximately 1300 m) and are predominantly rural but benefit from a temperate climate (quite hot with rains between October and March; cool and dry from May to August) with generally relatively good rainfall (average temperature and annual rainfall c25° and c1000 mm) and fertile soils. The principal crops include maize, sorghum, finger millet, yams, cotton, tea, bananas, avocados, sugarcane and other fruits. Most local people also grow vegetables and keep cattle, goats and chicken. eligibility criteria and participation and follow-up rates The baseline census and survey were conducted in a phased manner (one site at a time) between July 1998 and January 2000. In the census, a household was defined as a group of people who regularly eat together from the same cooking pot. Regular household members aged 17-54 years for men and 15-44 years for women-the ages of expected highest HIV incidence-were eligible for enrolment into the cohort (figure 2). However, participation in the cohort was restricted to a maximum of one member of each marital group (ie, a man and his wife or wives), selected at random, in order to maximise statistical power for the trial of the peer education and STI treatment intervention. Local village community workers were employed as guides to assist in locating participants. Where eligible individuals were unavailable for interview at the first household visit, appointments and up to two additional visits were made.
The second round of the open cohort survey was conducted between August 2001 and July 2003. All baseline respondents and individuals who had aged into the qualifying age range were eligible for this round. In-migrants in the 3-year intersurvey period and visitors were eligible for enrolment in the last eight sites. The third round ran from August 2003 to August 2005. Eligibility for the cohort was extended to include all men and women aged 15-54 years and the restriction to one member of each marital group was lifted. The fourth round ran from August 2006 to November 2008. All households were eligible for enumeration in the census and the same age criteria were used but follow-up and recruitment into the cohort were limited to members of a random sample of two-thirds of households due to funding constraints. The same eligibility criteria were applied in round 5, which ran from October 2009 to July 2011, and in round 6, which ran from July 2012 to November 2013. In round 6, four sites were dropped from the study, again due to funding constraints.
In the household censuses, the overall response rate was 98.  mentary table S1). This estimate for the response rate in the baseline census is an overestimate because fears that the researchers were Satanists-because they were asking for blood specimens thought to be used by Satanistscaused difficulties in identifying households. From the third round onwards, a steady increase in numbers of households has been observed reflecting reductions in these Satanist fears and growth in the population. The escalating economic crisis and a government initiative in 2005 to clean up the urban areas contributed to the increase in household numbers in round 4 (2006-2008). No difference in response rates was seen between households selected and not selected for individual interviews in rounds 4 to 6 (see online supplementary table S1).
At baseline, 11 453 individuals were eligible for the study, of whom 79.5% (9109) participated (figure 2, see online supplementary figure S2). In subsequent rounds, using the same age ranges for each sex for comparison (17-54 years for men; 15-44 years for women), overall participation rates have been similar except in round 6 when the overall rate fell to 73.0%. The cohort size has varied between 6269 in round 2 (2001)(2002)(2003) and 13 196 in round 3 (2003-2005) reflecting, primarily, the changes in eligibility criteria between rounds. Participation rates generally have been higher in women than in men (see online supplementary table S2). Direct refusal rates are consistently low (<5%), most non-participation being due to temporary absences from the household reflecting the high population mobility found in Zimbabwe.
The follow-up rate among all members of the cohort in the preceding round has varied between rounds from 47.0%  table  S1). However, most loss to follow-up comprises previous members of the cohort who ceased to be eligible due to  2) In rounds 1 and 2, a maximum of one member per marital grouping was selected at random for interview; this restriction was dropped from round 3 onwards.
(3) In round 2, individuals who migrated into a household since round 1 were only treated as eligible from site 5 (out of 12 sites). (4) From round 4, eligibility for individual interviews was restricted to individuals from a random sample of two-thirds of enumerated households. (5) In round 6, the number of study sites was reduced from 12 to 8 (2 agricultural estates and 2 subsistence farming areas were dropped). R1, R2, … indicate round numbers in the cohort survey.

Questionnaire data
In the household census questionnaire, the location and identity of each household is recorded (table 1). Basic sociodemographic information is collected for each member of the household including information on the eligibility criteria for inclusion in the cohort for adults. In households enumerated in previous rounds of the census, details of individuals who stayed in the household at or following the last visit are recorded (even if they have since left the household) and details of their survival status and date of leaving the household (where applicable) are recorded. Information on moveable and immoveable household assets is also collected for use in measuring socioeconomic status using wealth indices. 4 The individual respondent questionnaires used for the Manicaland cohort comprise sections on the respondent's own sociodemographic characteristics, the characteristics of up to four current spouses, the respondent's psychological health (since round 5-including all questions from the Shona Symptom Questionnaire, 5 a locally validated common mental health inventory and WHO's Self-Report Questionnaire 6 ), sexual relationships, health and access to treatment (since round 3), HIV knowledge and awareness (including exposure to HIV control programmes), and fertility and pregnancy histories (table 1). Dried blood spot (DBS) specimens have been collected for anonymised HIV testing for research purposes only as a requirement for participation at each round of the cohort. Free parallel voluntary HIV counselling and testing services were made available locally for cohort members during survey visits. 7 For cohort members who passed away between rounds of the survey, verbal autopsy interviews were conducted with the deceased's primary caregiver. The questionnaire included questions on accidents, medical conditions and symptoms during the final illness, 8 and on social and financial circumstances surrounding the death 9 (table 1).
Unusual features of the Manicaland cohort include parallel HIV surveys among pregnant women attending antenatal (ANC) check-ups at local health clinics, conducted to obtain information in biases in routine HIV surveillance data 10 (see online supplementary table S4); and six rounds of facility surveys conducted (2010-2016) to measure trends in local availability of HIV services. 11 In the first five rounds of the Manicaland cohort, all interviews were conducted using paper questionnaires. In round 6, the questionnaires for household census and individual cohort interviews were administered using HTC Smartphones using EpiCollect software. 12 Copies of the study questionnaires are available from the Manicaland Centre for Public Health Research website (http://www. manicalandhivproject. org/ questionnaires). The data from all rounds of the study are held in an SQL relational database developed for use in Microsoft Access. † Eligibility for the adult individual general population cohort: round 1: regular household members who had slept in the household at least four nights in the last month and had been resident in the household at the same time 1 year earlier, men aged 17-54 years and women aged 15-44 years limited to one member of a marital union selected at random (to maximise power in the embedded community randomised controlled trial of HIV prevention interventions); round 2: same criteria as in round 1 except that in-migrants were not eligible in the first four sites (Katiyo tea estate, Eastern Highlands tea estates, Bonda Mission, Honde Mission). In the remaining eight sites (and in all sites in subsequent rounds), individuals who stayed in households in the study areas the night before the census visit but who had not met the round 1 residence tests were treated as eligible for participation in the cohort; round 3: eligible age ranges extended to 15-54 years for men and women; restriction to one member of each marital union dropped and residence criteria extended to all persons who slept in the household the previous night rounds 4 and 5: same criteria as in round 3 except that eligibility was limited to adults in a random sample of two-thirds of households in the household census; round 6: same criteria as in rounds 4 and 5 but restricted to eight sites: Eastern Highlands tea estate, Bonda Mission, Honde Mission, Selborne forestry estate, Nyazura, Nyanga, Watsomba, Nyabadza/Nyahukwe. ART, antiretroviral therapy; PMTCT, prevention of mother-to-child transmission of HIV infection; STDs, sexually transmitted diseases. The HIV incidence rates for each intersurvey period reported in this paper were estimated assuming that new infections between rounds occurred midway between the first and second interview dates.

characteristics of the study population
The study population is comprised primarily of people who speak the Manyika dialect of Zimbabwe's majority Shona language. Most are also Christian, belonging to a large number of different Mission, Apostolic, Pentecostal and other spiritual churches. 14 Customary marriage, based on payment of bride-wealth, is almost universal, and is often followed by a church wedding. Polygyny remains common in some Apostolic churches and people who practice traditional religion. 15 As elsewhere in Zimbabwe, education levels are high compared with other countries in sub-Saharan Africa. 16 17 Just over half of the cohort is female reflecting the predominantly rural study areas (table 2). Over time, the cohort has aged somewhat (from a median of 25 years in round 1 to 27 years in round 5) and the proportion living on agricultural estates has fallen due to retrenchments on these estates reflecting increased mechanisation and the economic decline. The latter is also reflected in the large increase in unemployment between round 2 (32.8%) and round 4 (55.5%). However, education levels 17 and the proportion of the cohort who are married have both increased due, in part, to the ageing of the cohort.

FIndIngs to dAte hIV surveillance in a high hIV prevalence setting
The pilot study provided important early evidence that, by the early 1990s, HIV prevalence had reached high levels (23.3%) in the general population in rural areas of Zimbabwe. 2 Up to this point, HIV prevalence had been found to be much higher in urban areas than in rural areas in most sub-Saharan African countries and the high prevalence in rural areas meant that Zimbabwe was faced with one of the world's largest HIV epidemics. The study also found extremely high prevalence in young women aged 15-24 years (20.8%). 2 In the Manicaland cohort, HIV prevalence was 25.1% at baseline (1998)(1999)(2000) and fell steadily to 16.7% in round 5 and 15.8% in round 6 (table 2, figure 3). To provide a consistent comparison across rounds of the cohort survey, these data are shown throughout for men aged 17-54 years and women aged 15-44 years who were regular household members and stayed in the household for at least 4 nights in the last month before the interview. *In round 2, individuals who had migrated into a household in the study areas since baseline were only treated as eligible from site 5 (out of 12 sites). †In round 6, the number of study sites was reduced from 12 to 8 (2 agricultural estates and 2 subsistence farming areas were dropped). ‡Sixty missing cases due to indeterminate HIV test results. In the late 1980s, Zimbabwe established a national HIV surveillance system based on unlinked anonymous testing of pregnant women attending ANC check-ups. 18 19 The Manicaland Study, using data from its parallel general population cohort and ANC HIV prevalence surveys, has contributed information on the extent and causes of bias in ANC surveillance data on levels and trends in HIV prevalence 10 20-22 which have been used to develop the methods used in Zimbabwe and internationally to produce national HIV estimates. [23][24][25][26] sexual behaviour, migration and the spread of hIV infection Understanding the role of sexual behaviour in the spread of HIV infection has been hampered by reporting biases in the data. 27 In the Manicaland cohort, we developed an informal confidential voting interview method to reduce social desirability bias which has produced epidemiologically plausible results. 28 Data from the cohort were used to provide a detailed description of patterns of sexual risk behaviour in eastern Zimbabwe and their associations with HIV infection 13 29 30 and of changes in behaviour over time (table 3). In particular, the data showed that large age differences between sexual partners were common in the study population (median difference 6 years for women aged 15-24 years; IQR 4-9 years) and were associated with increased risk of HIV infection in young people. 29 Using a mathematical model, we found that age differences between men and women in sexual partnerships are unlikely to affect the scale of HIV epidemics but can explain the large female-male ratios of HIV infection found in young adults in sub-Saharan African populations. 31 Data from the cohort provided evidence that medical injections are not a major contributor to new HIV infections in generalised epidemics. 32 The Manicaland cohort provided the first evidence for declines in HIV prevalence occurring within the general population in Zimbabwe associated with reductions in sexual risk behaviours. 13 In other studies, data from the cohort have been used to assess the effectiveness of national HIV prevention programmes in reducing sexual risk behaviour. [33][34][35] The inter-relationships between migration and HIV in Manicaland are complex and not fully understood. 36 37 However, while high internal population mobility may have driven the early spread of HIV infection into and within rural areas of Zimbabwe 2and therefore contributed to the size of the national epidemic-the subsequent extensive out-migration from Zimbabwe was probably not a major factor in the decline in HIV prevalence that occurred from the late 1990s. 16 38 the demographic impact of a generalised hIV epidemic Data from the Manicaland cohort were used to test early mathematical model predictions that HIV epidemics could eliminate the high rates of natural population increase (≥3% per annum) seen in sub-Saharan African countries in the 1980s. [39][40][41] By the late 1990s, adult mortality in Manicaland was much higher among HIV-infected individuals (82/1000 person-years) than in uninfected individuals (7.2/1000 person-years) and the demographic impact was dramatic. However, even in the worst affected areas (towns with HIV prevalence of 33%), population growth remained positive, falling by two-thirds from 2.9% to 1.0%. 42 Using the cohort data, we demonstrated substantial reductions in fertility among HIV-infected women 43 and large increases in orphanhood 44 (and associated risks of HIV infection and poor health in orphaned adolescents) [45][46][47]  the role of social capital in hIV control in Zimbabwe Several studies have used data from the Manicaland cohort and, in some cases, qualitative data from the same populations to improve understanding of the underlying socioeconomic drivers of the spread of HIV infection. These have included studies on poverty and the influence of economic crises on patterns of HIV risk, 4 49 and on patterns and effects of HIV stigma, 50 masculinity, 51 52 religion, 14 53 and female sex work. 54 An unusual feature has been the innovative mixed-methods research done to describe local patterns of social capital. Strong evidence was found for associations between female participation in a range of different types of local community groups and reductions in HIV risk 55 and faster uptake of new services including HIV testing and prevention of mother-to-child transmission of HIV services. 56 The research developed the notion of HIV-competent communities 57 and highlighted the importance of community leadership and participation as a key factor in the success of HIV control interventions. 58 59 A full list of publications is available from the Manicaland Centre for Public Health Research website (see above). Open Access

strengths And lImItAtIons
The Manicaland cohort is one of the handful of long-running, large-scale, General Population HIV Sero-Surveys conducted in sub-Saharan African countries with widespread epidemics that constitute a major resource for evaluating the population-level impact of HIV control strategies. 60 A major strength of the Manicaland cohort is its comprehensive and consistent approach to measurement of trends in HIV prevalence, HIV incidence, HIV-associated mortality and all-cause mortality, sexual risk behaviours and health-seeking behaviours, and the coverage and effects of national HIV control programmes over time. The study is also unusual in its inclusion of a parallel ANC Survey, 22 in its use of validated methods to improve the quality of data on sexual behaviour 61 and in its use of mathematical models in interpreting the wider implications of the findings. 31 62 A weakness of studies that focus on localised areas can be difficulty in establishing whether findings are generalisable to the national population. In the Manicaland cohort, this has been addressed partially by including four of the main socioeconomic strata found in Zimbabwe. Triangulation of results with data from national sources shows that the overall levels and trends in the HIV epidemic observed in the study sites have been similar to those seen nationally. 16 63 64 The broader generalisability of the data on sexual behaviour patterns and trends has been explored in collaborative work with other general population studies in sub-Saharan Africa in the ALPHA network 65 66 and through studies using mathematical models. 67 Specific weaknesses of the cohort that we hope to address in the future include the age limit (55 years) for participation, and the length of (18-24 months) and time intervals between (2-3 years) rounds of follow-up. The latter means that individuals who move into and out of the study areas between rounds of the survey, may be missed and that measurement of some key variables including mortality can be subject to recall and misclassification bias.
collAborAtIons And Future dIrectIons The Manicaland cohort has provided a valuable resource and platform for the design and implementation of a number of trials of HIV control interventions and collaborative projects led by independent researchers. These include trials of peer education among female sex workers and their clients (1998)(1999)(2000)(2001)(2002)(2003) 3 and conditional and unconditional cash transfers to support orphans and vulnerable children (2009-2011), 68 studies on HIV and migration, 37 69 and innovative studies on HIV competent schools 70 71 and on patterns of social contacts that influence the spread of infectious diseases in children. 72 Subject to funding availability, we plan to extend the cohort to provide data on the implementation and impact of current and future HIV control programmes including primary prevention interventions and programmes to address major comorbidities associated with the ageing of HIV epidemics. We would welcome proposals for further collaborative projects related to this work.
contributors All authors read and contributed to this manuscript. SG, GPG, OM and CAN designed the study and raised the funding. CAN, RM and SG collected data for the study and enrolled participants. CC and MS designed and conducted the qualitative studies. PRM, JMut and JMay conducted the laboratory procedures. AT, RR and JE undertook data management. CC, JE, SG, TH, PM, SM, RR, LS, MS and RS analysed the data. SG wrote the first draft of the report, to which all authors contributed.
Funding Core funding for the Manicaland Cohort has been provided by the Wellcome Trust (084401/Z/07/B and 069516/Z/02/Z). UNAIDS, WHO, the European Union, the World Bank, and the Bill and Melinda Gates Foundation have provided funds for aspects of the work.
competing interests SG has shares in AstraZeneca and GSK.
ethics approval Medical Research Council of Zimbabwe; Imperial College Research Ethics Committee.
Provenance and peer review Not commissioned; externally peer reviewed. data sharing statement Data from the Manicaland Cohort can be obtained from the project website (http://www. manicalandhivproject. org/ data-access. html). Here we provide a core data set which contains a sample of sociodemographic, sexual behaviour and HIV testing variables from all six rounds of the main survey, as well as data used in the production of recent academic publications. If further data are required, a data request form must be completed (available to download from our website) and submitted to s. gregson@ imperial. ac. uk. If the proposal is approved, we will send a data sharing agreement which must be agreed upon before we release the requested data.
open Access This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http:// creativecommons. org/ licenses/ by/ 4. 0/