| COVARIATE(S) MEASURED WITH GIS |
| BRDAR ET AL [42] |
Côte d’Ivoire |
Mobile phone usage data relating to social connectivity, spatial location, migration and movement, and activity. |
Predictive Ridge and Support Vector regression models |
5 million mobile phone users |
Night-time connectivity and activity, area covered by users and overall migrations are strongly linked to HIV prevalence. Models based on spatial features were highly predictive of HIV. |
| BRODISH AND SINGH [57] |
Mozambique |
S. haematobium exposure (distance to high-endemic areas) |
Regression analysis |
8,847 |
Exposure to S. haematobium increased the odds of HIV by three times, controlling for demographic and sexual risk factors. |
| CARREL ET AL [35] |
Democratic Republic of Congo |
Distance to the nearest city |
Poisson mixed effects regression comparing two time periods |
9275 (2007), 18,257 (2013) |
Urban HIV prevalence decreased and rural HIV increased between 2007 and 2013. Protective effect of distance to city disappeared. |
| MESSINA ET AL [19] |
Democratic Republic of Congo |
Distance to cities, rivers, refugee camps, conflict sites |
Regression analysis |
9,755 |
Proximity to city and distance to river (for women) associated with HIV. |
| TANSER ET AL [58] |
South Africa |
Mean distance from household to major road |
Regression analysis |
16,583 |
Distance to major road strongly correlated with HIV prevalence. |
| ZULU ET AL [29] |
Malawi |
Distance/time to roads, public transport and health facilities, proximity to cities, and elevation |
Regression analysis and mapping of clusters and outliers of selected risk factors relative to HIV prevalence (local Moran's I and Getis-Ord Gi*) |
54 ANCs for risk analysis |
Mean travel time to public transport for ages 30–44 associated with HIV. Distance to main road protective. Hotspots and coldspots of relationship between risk factors and HIV identified in different areas. |
| SPATIAL ANALYSIS OF RISK FACTORS |
| ABIODUN ET AL [64] |
Nigeria |
Early sexual debut |
Bayesian spatial Cox hazards model for spatial analysis of early sexual debut. |
4,301 |
Northern states significantly earlier sexual debut after controlling for other factors. |
| AKULLIAN ET AL [65] |
Kenya |
HIV stigma |
Describe spatial patterns of HIV stigma using difference of K-function cluster analysis and spatial regression. |
373 |
Spatial trend and clustering in external stigma (blame) but not internal stigma (shame). |
| AKULLIAN ET AL [60] |
Kenya |
Male circumcision |
Smoothed map of circumcision in 2008 and 2014. |
484 (2008); 1649 (2014) |
Clear boundary in circumcision prevalence between traditionally circumcising areas in 2008, diminished in 2014 after VMMC program implementation. |
| CUADROS ET AL [63] |
Kenya, Malawi, Tanzania |
Malaria |
Smoothed map (model-based geostatistics) of malaria prevalence to calculate covariate in logistic regression. |
19,735 |
People living in high malaria prevalence areas were nearly twice as likely to be HIV positive as those living in low malaria areas. |
| CUADROS ET AL [37] |
Tanzania |
Male circumcision |
Compare Kuldorff clusters and LISA hotspots of male circumcision (MC) and HIV. Compare HIV incidence by gender inside and outside MC cold spots. |
2003–04: 12,522; 2007–08: 16,318; 2011–12: 18,809 |
Outside of low-MC clusters, females at greater risk than males, but inside low-MC clusters, males and females at equal risk. |
| CUADROS AND ABU-RADDAD [28] |
Cameroon, Kenya, Lesotho, Tanzania, Malawi, Zambia, Zimbabwe |
Sero-discordant partnerships |
Compare Kuldorff clusters of sero-discordant couples and HIV prevalence. Compare epidemiologic measures of discordancy inside and outside clusters. |
16,140 |
No spatial pattern for sero-discordancy independent of HIV prevalence patterns. HIV prevalence correlated with proportion of couples that were sero-discordant. |
| PALK AND BLOWER [62] |
Lesotho |
Couples with one member temporarily living away from home |
Kriging maps of divided household by absent member (husband vs. wife) and their temporary residence (within country vs. South Africa). Regression on HIV status and extramarital partnerships. |
2,026 couples |
Spatial patterns of divided households differed based on where the absent partner was. No significant association between divided household and HIV. Absent wives increased the risk of extramarital partners for men. |
| SARTORIUS [27] |
South Africa |
Clusters of age-specific mortality |
Comparing Kuldorff clusters of high and low mortality rates |
1,110,166 person-years |
Multiple social and demographic characteristics identified that significantly differed between high and low mortality clusters |
| TANSER ET AL [23] |
South Africa |
Clusters of high and low HIV |
Compare characteristics of high and low HIV clusters. |
12,221 |
High prevalence clusters have high education, household wealth, employment, lower marriage and migrants. |
| WAND AND RAMJEE [61] |
South Africa |
Education, age at sexual debut, cohabitation with partner, number of recent partners, transactional sex |
Geo-additive spatial regression of risk factors and HIV risk at two clinics. |
3,462 |
Women at Botha's Hill clinic had higher education, more sexual partners and less marriage. Total risk score showed higher impact on Botha's Hill women than Umkomaas. |
| WESTERCAMP ET AL [59] |
Kenya |
Sexual behaviors and STI history |
Kuldorff cluster detection for STIs and sexual behaviors among young men |
649 |
No clusters detected other than condom use. |
| ZULU ET AL [29] |
Malawi |
Distance to main roads, travel time to public transport, ever having tested for HIV, education, syphilis |
Mapping of clusters and outliers of selected risk factors relative to HIV prevalence (local Moran's I and Getis-Ord Gi*) |
19 ANCs for time trends; 54 ANCs for risk analysis |
Hotspots and coldspots of each explanatory variable relative to HIV identified in different areas. |
