Recent declines in HIV prevalence and incidence in Magu DSS, 1994-2007 Mark Urassa, Raphael Isingo, Milalu Ndege, Milly Marston, Julius Mngara, Basia Zaba and John Changalucha INDEPTH conference, Dar-es-Salaam, September 22 nd 26 th, 2008
Structure of presentation 1. Introduction to Magu DSS 2. Methods: fieldwork and statistical ti ti analysis 3. Results: prevalence and incidence trends 4. Discussion and policy implications
Magu DSS covers just one ward in Magu district, the eastern part of Mwanza region in NW Tanzania. City Cty
Kisesa ward and its population Population About 20,000 people p in 1994, fourteen years later 28,000; growth 2.4% per year, faster in and near trading centre Ethnicity: 95% are Sukuma Religion: 74% Christian, 23% Traditional, 3% Islam Education: 14 primary schools (11 public, 3 private) and 2 secondary schools (1 public, 1 private) Health: 7 health facilities (4 public, 3 private) including VCT service and new ART clinic run by the research project Economy Per capita income below $120 per year Farming is main source of income, petty trading common, including selling agricultural produce in Mwanza city
Objectives of the cohort study To improve understanding of the dynamics of the HIV epidemic To asses the demographic, social and economic impacts of the HIV/AIDS epidemic To evaluate the effects of national prevention, treatment and care interventions as implemented in Kisesa Ward To measure child and adult mortality and fertility in the general population and by HIV status To asses the leading causes of death through verbal autopsy To asses changes in the family structure due to HIV epidemic To provide reliable data for district health planning
Types of research 1 Demographic surveys 2 Serological og surveys house to house, whole pop adults invited to village clinics keep track of births, deaths and HIV & other lab tests movements every 3 years twice a year long questionnaire about simple, quick questions behaviour & other risks 3 In depth enquiries specially selected individuals sensitive topics timing as required some open-ended questions &/or qualitative interviews 4 Operations research community groups, village leaders, health services help national surveillance timing continuous / as required IEC activities and feedback
Timing of study components Demography Serology ANC surveillance VCT service 0 Jan-94 Jan-95 Jan-96 Jan-97 Jan-98 Jan-99 Jan-00 Jan-01 Jan-02 Jan-03 Jan-04 Jan-05 Jan-06 Jan-07 Jan-08 Jan-
Demographic surveillance Visit every household and collect information from household head or other adult member Ask about births, pregnancies, deaths, moves in or out, update information on school attendance Link children to mothers and fathers, adults to their spouses Follow-up death reports with verbal autopsy interview Use demographic data to identify eligible persons for sero-survey and special topic surveys
Our very simple DSS data collection form 6 columns of pre-printed identifying information ex-members of household are shown: dead in black, absent in dark grey. Returning migrants get their old line numbers back pale grey shading shows which fields should not be filled
Sero-surveys Resident adults 15+ invited to special village clinics Clinics i provide treatment t t for common illnesses, family members can also be treated Long questionnaire on socio-demographic background, sexual behaviour, other risk factors, knowledge and attitudes Finger prick blood specimens tested anonymously at NIMR laboratory (lab and field staff cannot link test results to personal identifiers) VCT services offered during latest surveys, using separate samples of venous blood
A few pages from our 19-page very complicated Sero-survey questionnaire!
Sero-survey questionnaire covers: Registration & consent Identification Socio-economics Residence & mobility Condom knowledge First sex Pregnancy history Marital history Sex within marriage Polygamy Non marital partners High risk sex Family planning STI symptoms & treatment HIV knowledge Injections & blood transfusion Stigma & HIV attitude Use of health services Experience of VCT It is linked to other information using pre-printed stickers
Sero-survey participation,1994-2007 1994-1995 15-44 years 5,672 attended (74%) 1996-1997 15-46 years 6,174 attended (75%) 69% follow-up 1999-2000 15+ years 5,650 attended (68%) 52% follow-up 2003-2004 15+ years 6,943 attended (66%) 50% follow-up Survey 5: provisional data includes ineligible temporary visitors 2006-2007 15+ years 9,334 attended (~73%) Surveys 1-4: numbers exclude ineligible temporary visitors ~52% follow-up
Statistical methods: (i) allowing for effects of participation i bias on prevalence Assumptions made about eligible non-attenders: 1. Anyone earlier testing positive was positive, anyone later testing negative was negative 2. Anyone earlier testing negative with no later test was assumed still negative, anyone later testing positive with no earlier test was assumed already positive 3. Those who had never tested were assigned the average prevalence for their age, sex and residence area (this last assumption allows for the effect of the age, sex and residence structure on the crude prevalence estimates) The bias limits following our prevalence estimates show the maximum and minimum possible effects of combining assumption 1 with assumption 2 and/or assumption 3.
Statistical methods: (ii) allowing for interval censoring in incidence id estimates t If sero-surveys are 2 years apart, it is reasonable to assume all infections occur halfway between surveys. Our surveys are ~3 years apart and take ~1 year to complete halfway assumption leads to heaping of infections in non-survey years those who move between villages may have inter-survey intervals between 2 and 4 years Multiple imputation technique assign individual infections to random date between last negative and first positive test, calculate age, sex and region specific rates repeat 10,000 times, average the age, sex and region specific rates for best estimates; use minimum and maximum rates as bounds chose random dates in the run that was closest to average result in future analyses that need individual infection dates (e.g. survival analysis)
Statistical methods: (iii) defining a summary incidence measure: Life Time Risk of Infection = 1 risk of staying uninfected Kaplan-Meier failure estimate.5.4 Kisesa, 1994-2004 Life Time Risk of HIV infection = 40% (by age 65).1.2.3 Kisesa, 1994-2004 Average HIV prevalence = 7.3% 0 15 20 25 30 35 40 45 50 55 60 65 age 95% CI Failure function
Trends in HIV infection in Kisesa Prevalence, ages 15-44, crude proportion: p p sero 1 6.0% (m 5.1%, f 6.9%); 5,645 eligibles tested sero 2 6.7% (m 5.1%, f 8.0%); 6,164 eligibles tested sero 3 8.3% (m 6.9%, f 9.3%); 5,639 eligibles tested sero 4 8.2% (m 7.5%, f 8.8%); 6,937 eligibles tested sero 5 7.3% (all ages, no sex breakdown yet); 8,638 test results Incidence, ages 15-44, crude rates: sero1tosero2 0.81%; based on 8,288 person-years sero 2 to sero 3 1.24%; based on 10,220 person-years sero 3 to sero 4 1.14%; based on 13,416 person-years
HIV prevalence trends in Kisesa 16 14 HIV prevalen nce 12 10 8 6 4 2 Roadside Women Roadside Men Remote Women Remote Men 0 1994 1996 1998 2000 2002 2004 year
Is prevalence change a reliable guide to the course of the epidemic? 500 ber infected Cha ange in num 400 300 200 100 0-100 -200???? pos Enter pos Neg pos Pos died Pos left Pos???????? pos Enter pos Neg pos Pos died Pos left???? pos Enter pos Neg pos Pos died Pos left largest changes are due to migration, not sero-conversion or death -300 Pos???? Pos???? -400 sero1 to 2 sero2 to 3 sero3 to 4
Trends in HIV incidence: Life Time Risk by inter-survey period Cumu lated % ri sk of infe ction to a ge 45 50 40 30 20 10 0 1996 1998 2000 2002 year Roadside Men Roadside Women Remote Men Remote Women
Trends in HIV Crude incidence: incidence Crude hazard Hazard by Rates calendar by year calendar year.02 Incidence hazards with uncertainty ranges.025 Rural males.025 Rural females.015.01 0.00 05.01.015.02 0.00 05.01.015.02 1994 1996 1998 2000 2002 2004 uncertainty range crude incidence hazard 1994 1996 1998 2000 2002 2004 uncertainty range crude incidence hazard.005 Roadside males Roadside females.015.02.025.015.02.025 1994 1996 1998 2000 2002 2004 0.005.01 Male rural Male roadside 1994 1996 1998 2000 2002 2004 0.005.01 Female rural Female roadside 1994 1996 1998 2000 2002 2004 uncertainty range crude incidence hazard uncertainty range crude incidence hazard
Trends in age patterns of incidence
Full details of HIV incidence age pattern emerge when all data analysed together othed hazard Smo.02.015..01 Incidence peaks later for men and then continues at a higher rate than for women.005 0 20 30 40 50 60 age There are secondary incidence peaks at older ages for both men and women males females poster 5
Provisional results from sero5 (unlinked data) Participants as percent of invited: 62% (invitations issued to adults resident in DSS 19 or DSS 20; true participation rate calculated after DSS 21, expected to be ~ 73%) Overall HIV prevalence = 7.3% (down from 8.2%) Cannot calculate incidence prior to linking Proportion accessing VCT = 17.3% (up from 9.2%)
Do we suffer from the Hawthorne effect? 14 ANC 12 urban 10 8 6 4 2 ANC roadside ANC rural Kisesa roadside Kisesa rural DHS Mwanza 0 1994 1996 1998 2000 2002 2004 2006 2008
Overall conclusions There are encouraging signs of decline in incidence which appears to have started in the late 1990s. When we compare incidence in men and women we do not see the kind of female disadvantage that prevalence data suggest. Women become infected at earlier ages, but their overall life time risk is similar to that of men BUT It is possible that the incidence decline was part of the natural epidemic dynamic, rather than a response to prevention campaigns It is worrying that incidence and prevalence in remote rural areas are now catching up with the roadside settlements and trading centre A i l ti f f t h i th i t f ART ll t A crucial question for future research is the impact of ART roll-out on HIV incidence