Understanding Epidemic Contexts within the Cluster-Randomized SEARCH Trial: a Clustering Approach to Grouping Rural Communities in According to Epidemic Characteristics

Britta Jewell,1,2 Adam Akullian,2 Carol Camlin,3 Maneet Kaur,3 Tamara Clark,3 Edwin Charlebois,3 Moses Kamya,4 Maya Petersen,1 Diane Havlir,3 and Anna Bershteyn2

1 2 3 UC Berkeley, Berkeley, CA, Institute for Disease Modeling, Seattle, Washington UCSF, San 4 Francisco, CA, Makere University, Kampala,

BACKGROUND METHODS

SEARCH is a community-randomized trial of the “test-and-treat” strategy for HIV We explored community-level adult HIV prevalence and factors including prevention (NCT01864603) in 32 communities spanning three geographic regions demographic structure, mobility (defined as the proportion of adults spending ≥1 in East Africa – Southwest Uganda, East Uganda, and Western . Baseline night away from home in the past month), and traditional male circumcision (TMC) HIV prevalence across the communities varied widely, ranging from a minimum of rates as potential factors for community clustering. Data about demographic 1.5% in a community in East Uganda to 24.6% in a community in Western characteristics, traditional male circumcision rates, and mobility were collected at Kenya1. Using baseline HIV prevalence and household census data (N=146,874 baseline of Phase I of the SEARCH trial from 2013-2014. K-means clustering was adults ≥15), we investigated the geospatial distribution of factors potentially used to categorize each of the communities (N=32) into six clusters (three control driving community-level HIV prevalence at baseline, including traditional male and three intervention clusters, divided into low, medium, and high prevalence circumcision rates, mobility, and population age distribution. Historically, clusters in each arm of the trial), according to epidemic context. Clustering also communities with high rates traditional male circumcision have been associated identified which communities differed from others in their local region. with lower HIV prevalence,2 and greater mobility has been associated with higher HIV prevalence.3 We clustered communities for use in a mathematical model of SEARCH communities using EMOD, developed by the Institute for Disease Modeling (IDM). (A) (B) FIGURE 1

n 100%

HIV l o

i 90%

a s

Prevalence n

i 80%

o c (A) Local HIV prevalence among adults aged i 70%

(15+) t East Uganda i

≥15 at baseline in the 32 SEARCH m 60% d

<5% u Southwest Uganda a communities in Uganda and Kenya c 50%

r Western Kenya

5-10% r i (B) HIV prevalence among adults aged ≥15 and T 40%

10-15% / C

30%

traditional male circumcision rates at w

15-20% e

baseline in East Uganda, Southwest l 20% a 20-25% % Uganda, and Western Kenya 10% M 0% (C)HIV prevalence among adults aged ≥15 and 0% 5% 10% 15% 20% 25% mobility (% of adults who traveled >1 night of HIV Prevalence (15+) the past month) at baseline in East Uganda, Southwest Uganda, and Western Kenya (C) (D) (D)HIV prevalence among adults aged ≥15, 60%

o 57%

t h

population age structure (% of population h

n 56%

g 50%

h

W o

aged <15), and k-means clusters of 32 i

t 55%

i

t

5

N n communities. Communities were grouped + Low prevalence cluster

54% a

40% 1

5 l

o East Uganda 1 < 53% Medium prevalence cluster

into a low prevalence cluster, medium 1

u

<

M Southwest Uganda

High prevalence cluster p 30% d

s 52%

prevalence cluster, and high prevalence t

d

e o

e Western Kenya s Cluster center

cluster. Black crosses indicate the cluster e 51%

g

P

l g

a 20%

e 50% f

A Community clustered w/ P

center. Communities circled in black are A

v

o f f 49% 10%

those that clustered with a different a different geographic region

o

o r

48% % geographic region. T 0% % 47% 0% 5% 10% 15% 20% 25% 0% 5% 10% 15% 20% 25% HIV Prevalence (15+) HIV Prevalence (15+) RESULTS CONCLUSIONS

High HIV prevalence in Western Kenya occurred in the context of low rates of Consistent with findings from DHS and AIS surveys, HIV prevalence in Western TMC and high population mobility relative to the two Ugandan regions. The only Kenya was substantially lower in a community where a majority of men are Kenyan SEARCH community with a majority of men circumcised had one-third the traditionally circumcised. However, this inverse relationship between TMC and HIV prevalence of other communities in Western Kenya (6.0% vs. 18.6%)1. In prevalence was not observed in Uganda. Western Kenya was also distinct in its contrast, TMC rates were low in Southwest Uganda despite far lower HIV high mobility; in contrast, East and Southwest Uganda had similar mobility but prevalence, and there was no relationship between TMC and community HIV different demographic structures. These factors make it possible to cluster prevalence in East Uganda despite wide variation in TMC. East and Southwest SEARCH communities into distinct epidemic patterns for use in a mathematical Uganda had similar mobility, but East Uganda had a younger population and model. lower HIV prevalence than Southwest Uganda. Demographic structure and HIV ACKNOWLEDGEMENTS prevalence clustered the communities into three distinct groups, mostly corresponding to geographic region. Three communities clustered with different Research reported in this poster was supported by Division of AIDS, NIAID of the National regions, including the lower prevalence community in Western Kenya, which Institutes of Health under award number U01AI099959 and in part by the President’s clustered with Southwest Uganda. One community in Southwest Uganda Emergency Plan for AIDS Relief, Bill and Melinda Gates Foundation, and Gilead clustered with East Uganda, and one community in East Uganda clustered with Sciences. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH, PEPFAR, Bill and Melinda Gates Foundation or Southwest Uganda. Gilead. The SEARCH project gratefully acknowledges the Ministries of Health of Uganda and Kenya, our research team, collaborators and advisory boards, and especially all REFERENCES communities and participants involved. 1. Jain, V., Petersen, M.L., Liegler, T., et al., “Population levels and geographical distribution of HIV RNA in rural Ugandan and Kenyan communities, including serodiscordant couples: a cross-sectional analysis,” The Lancet HIV 4:3 (March 2017), e122-e133. 2. Shaffer, D., Bautista, C.T., Sateren, W.B., et al., “The Protective Effect of Circumcision on HIV Incidence in Rural Low-Risk Men Circumcised Predominantly by Traditional Circumcisers in Kenya: Two-Year Follow-Up of the Kericho HIV Cohort Study,” JAIDS 45:4 (2007), 371-379. 3. Deane, K.D., Parkhurst, J.O., and Johnston, D., “Linking migration, mobility, and HIV,” Tropical Medicine and International Health 15:12 (December 2010), 1458-1463.