Table 2. Studies creating continuous surface maps of HIV.

AUTHOR	COUNTRY	TYPE OF SPATIAL PREDICTION	SIZE	KEY FINDINGS
BARANKANIRA ET AL. [18]	Burundi	Kernel density smoothing	8,086	Spatial heterogeneity independent of administrative boundaries. Identified locations in need of HIV resources.
CARREL ET AL. [35]	Democratic Republic of Congo	Bayesian kriging of 2007 and 2013 HIV data, subtracting the maps to show areas of greatest difference.	9275 (2007), 18,257 (2013)	HIV prevalence decreased in urban locations and increased in rural locations, but areas of high difference were relatively small.
CHANG ET AL. [41]	Uganda	Bayesian modeling of percent and number of people living with HIV (PLHIV) per km2	17,119	High HIV prevalence along Lake Victoria and patchy prevalence in district interior. Areas with highest number of PLHIV were inland in high population-density trading centers.
COBURN ET AL. [38]	Lesotho	Inverse distance weighted (IDW) mapping combined with population density map to display the number of HIV-positive persons per km2.	7,099	Density of infection is significantly higher in urban areas, but the majority of HIV-positive people live dispersed in rural areas.
CUADROS & ABU-RADDAD [17]	Cameroon, Ethiopia, Kenya, Lesotho, Malawi, Mali, Rwanda, Senegal, Tanzania, Zimbabwe	IDW mapping for visualizing differences in spatial distribution of HIV between time periods.	10 countries	HIV prevalence within high-prevalence clusters either did not decline or increased, even if national prevalence declined.
CUADROS ET AL. [37]	Tanzania	Kriging of HIV prevalence and male circumcision rates to assess their spatial correlation.	2003–04: 12,522 2007–08: 16,318 2011–12: 18,809	Areas of low male circumcision overlap with areas high HIV prevalence, and vice versa.
CUADROS & ABU-RADDAD [28]	Cameroon, Kenya, Lesotho, Tanzania, Malawi, Zambia, Zimbabwe	IDW mapping of HIV and sero-discordancy prevalence to assess their spatial correlation.	16,140	No spatial pattern for sero-discordancy independent of HIV prevalence patterns.
KALIPENI & ZULU [34]	Continental	IDW and kriging interpolation of international HIV prevalence for country-level estimates. Model epidemic curves for each country and project future trends.	1,442 sentinel sites over 18 years	Differences between UNAIDS estimates vs. kriging- and IDW-generated national estimates were statistically insignificant. Nearly all countries have reached maturity level of epidemic curve.
KLEINSCHMIDT ET AL. [36]	South Africa	Bayesian kriging map of HIV prevalence among youth.	11,758	Variation in HIV prevalence independent of provincial boundaries, highest in the east and for women.
LARMARANGE & BENDAUD [33]	17 countries	Kernel density estimation with adaptive bandwidths (prevR) to generate sub-national HIV estimates.		Continuity of HIV estimates across borders. Certainty of estimates varied depending on total sampling size, total number of administrative units, distribution of survey clusters across area.
MESSINA ET AL. [19]	Democratic Republic of Congo	IDW HIV prevalence maps by sex to create regional-level estimates.	9,755	Spatial variation in HIV, distribution and intensity varied by sex.
SARTORIUS ET AL. [40]	South Africa	Bayesian kriging of all-cause and cause-specific child mortality risk.	46,675	Two geographic foci of high mortality, matching areas of high HIV/TB mortality.
SARTORIUS ET AL. [39]	South Africa	Bayesian kriging of all-cause and cause-specific adult mortality risk.	104,969	Five geographic foci of high mortality, correlating to areas of high HIV/TB mortality.
SARTORIUS [27]	South Africa	Bayesian kriging of age-specific all-cause and HIV/TB mortality risk.	1,110,166 person-years	Spatial distribution of all-cause mortality risk varied by age group, reflecting spatial trends in HIV/TB mortality.
SCHAEFER ET AL. [24]	Zimbabwe	Kriging of HIV prevalence and uptake of HIV testing and counseling (HTC).	8,092	HIV prevalence higher in two urban areas for men and women, but HTC uptake lower in those areas and in one other.
SUBNATIONAL ESTIMATES WORKING GROUP [32]	Tanzania, Kenya, Malawi	Comparison of six methods. Pixel-level estimates: - Kernel density estimation with adaptive bandwidths (prevR) - Bayesian model-based geostatistics - Kriging of each covariate with regression to combine of layers Administrative unit-level estimates: - Shared spatial component model - Regression kriging at aggregated scale - Bayesian geo-additive mixed model.		All methods revealed within-country variations and were similar in accuracy, but Bayesian geostatistical approach slightly better.
TANSER ET AL. [23]	South Africa	Kernel density smoothing to estimate spatial distribution of HIV.	12,221	Spatial variation in HIV prevalence with highest prevalence in urban settlements near the National Road.
ZULU ET AL. [29]	Malawi	IDW HIV prevalence maps for eight years to compare trends over time.	19 ANCs for time trends	Spatial variation independent of district boundaries, shifting spatial patterns over time.
GONESE ET AL. [44]	Zimbabwe	Compare ANC surveillance with geographically proximate DHS data.	7,202 (ANC) 13,049 (DHS)	ANC and DHS similar for most populations, but ANC estimates were lower for women within 30km of ANC site.
MUSINGUZI ET AL. [43]	Uganda	Compare HIV prevalence rates between ANC surveillance sites and national population survey clusters within 30km.	16,936 (UHSBS); 9,668 (ANC)	Overall estimate similar. ANC-based was higher in ages 15–19, lower for those aged 30+, and in urban areas.