Assessing Spatial Patterns of HIV Knowledge in Rural Mozambique Using Geographic Information Systems

Tropical Medicine and International Health doi:10.1111/tmi.12437 volume 20 no 3 pp 353–364 march 2015

Assessing spatial patterns of HIV knowledge in rural Mozambique using geographic information systems

Charlotte P. Buehler1, Meridith Blevins1,2, Ezequiel B. Ossemane3,Lazaro Gonzalez-Calvo 3, Elisee Ndatimana3, Sten H. Vermund1,3,4, Mohsin Sidat5, Omo Olupona6 and Troy D. Moon1,3,4

1 Vanderbilt Institute for Global Health, Nashville, TN, USA 2 Department of Biostatistics, Vanderbilt University School of Medicine, Nashville, TN, USA 3 Friends in Global Health, Maputo, Mozambique 4 Department of Pediatrics, Vanderbilt University School of Medicine, Nashville, TN, USA 5 Department of Community Health, Faculty of Medicine, University Eduardo Mondlane, Maputo, Mozambique 6 World Vision Mozambique, Maputo, Mozambique

Abstract objectives To conduct a cross-sectional mapping analysis of HIV knowledge in Zambezia Province, Mozambique, and to examine spatial patterns of HIV knowledge and associated household characteristics. methods A population-based cluster survey was administered in 2010; data were analysed from 201 enumeration areas in three geographically diverse districts: Alto Molocu e, Morrumbala and Namacurra. We assessed HIV knowledge scores (0–9 points) using previously validated assessment tools. Using geographic information systems (GIS), we mapped hot spots of high and low HIV knowledge. Our multivariable linear regression model estimated HIV knowledge associations with distance to nearest clinic offering antiretroviral therapy, respondent age, education, household size, number of children under ﬁve, numeracy, literacy and district of residence. results We found little overall HIV knowledge in all three districts. People in Alto Molocu e knew comparatively most about HIV, with a median score of 3 (IQR 2–5) and 22 of 51 (43%) enumeration areas scoring ≥4 of 9 points. Namacurra district, closest to the capital city and expected to have the best HIV knowledge levels, had a median score of 1 (IQR 0–3) and only 3 of 57 (5%) enumeration areas scoring ≥4 points. More HIV knowledge was associated with more education, age, household size, numeracy and proximity to a health facility offering antiretroviral therapy. conclusions HIV knowledge is critical for its prevention and treatment. By pinpointing areas of poor HIV knowledge, programme planners can prioritize educational resources and outreach initiatives within the context of antiretroviral therapy expansion.

keywords HIV knowledge, geographic information systems, hot spot analysis, Mozambique, Getis-Ord Gi*, rural health

estimate of 11.5% ranks it among the most heavily HIV- Introduction afﬂicted nations in the world (INSIDA 2010; UNAIDS Since 2004, the global community has invested signiﬁ- 2013). cantly in HIV prevention and treatment through interna- Knowledge of HIV infection is seen as a critical compo- tional programmes such as the US President0s Emergency nent to its prevention and treatment, where low knowl- Plan for AIDS Relief (PEPFAR) and the Global Fund to edge often represents a barrier to engaging in care and the Fight AIDS, Tuberculosis and Malaria (De Cock et al. subsequent cascade of events needed to initiate antiretrovi- 2011; Moon et al. 2010, 2011; El-Sadr et al. 2012; Ver- ral therapy (ART) (Ciampa et al. 2012a,b). In SSA, young mund et al. 2012; UNAIDS 2013). Yet, the burden of adults aged 15–24 represented 74% of global HIV infec- HIV remains disproportionate, with 68% of all people tions in 2010, yet only 26% of women and 33% of men living with HIV residing in sub-Saharan Africa (SSA) had correct and comprehensive knowledge about HIV pre- (UNAIDS 2013) and the highest HIV prevalence noted vention (Global HIV/AIDS Response 2011). among women of reproductive age (UNAIDS 2013). In In previous studies in rural Mozambique, we found Mozambique, the latest 2009 adult HIV prevalence that while knowledge of basic HIV concepts, such as

C. P. Buehler et al. Spatial patterns of HIV knowledge

sexual or maternal transmission, was relatively common, gene–environmental interactions (Torbick et al. 2014). significant misperceptions existed which may impact HIV GIS allows one to make sense of geographic patterns in prevention and care delivery, including a belief that space and time, and help identify and analyse correlates washing after sexual contact can protect against transmis- of health behaviour and outcomes (Cromley & McLaffer- sion and that HIV results from a curse (Ciampa et al. ty 2012). 2012a,b). These misperceptions have been subsequently In this study, we examine the spatial patterns and clus- linked to negative health seeking behaviours, such as tering of HIV knowledge within three districts of delayed access to care or avoiding conventional HIV care Zambezia Province, Mozambique. We hypothesise that all together in favour of traditional healers (Audet et al. geographic patterns of higher and lower HIV knowledge 2012). Conversely, increased HIV knowledge has been can be identified, ultimately illuminating areas where correlated with reduced stigma and increased uptake in enhanced HIV education interventions are needed. HIV testing (Audet et al. 2012; Mukolo et al. 2013a,b). Health outcomes can correlate with geography (Crom- ley & McLafferty 2012; Richardson et al. 2013). Past Methods research suggests that proximity to health services is asso- Study context ciated with primary care utilisation (Tanser et al. 2006), contraceptive use (Brauner-Otto et al. 2007) and higher Zambezia Province, located in north-central Mozambique vaccine rates (Al-Taiar et al. 2010; Blanford et al. 2012). (Figure 1), is predominantly rural and depends almost A study in rural Malawi found that as distance from entirely on subsistence farming and fishing (Moon et al. health facilities increased, people were less likely to 2010). Of 11 provinces, Zambezia is one of the country’s receive their HIV test results (Thornton 2008). A similar poorest performing province in terms of health indicators study in Mozambique found that proximity to sexual and and continues to have high rates of tuberculosis (TB) and reproductive health services was associated with increased malaria, poor maternal and child health (MCH) indices, service utilisation (Yao et al. 2012). Our study is the first high levels of malnutrition and low literacy rates (Ciampa in Mozambique in which HIV knowledge has been stud- et al. 2012a,b; Instituto Nacional de Estatıstica Min- ied with respect to proximity to health facilities. isterio da Saude 2013). In 2009, Zambezia also had the As scaling-up of global HIV/AIDS activities continues, highest estimated number of persons living with HIV in efficient and evidence-based health planning and evalua- Mozambique (nearly 20% of the HIV-infected adult pop- tion tools are needed. Recent developments within geo- ulation) (INSIDA 2010). Additionally, a 16-year civil war spatial technologies enable the creation of maps with (1976–1992), fought disproportionately in Zambezia spatial and temporal data relevant to health planning, Province, resulted in destruction of many of its hospitals implementation and research (Richardson et al. 2013). and clinics. In 2009, World Vision United States was Geographic information systems (GIS) can characterise awarded a United States Agency for International Devel- locations with geographic data, allowing users the ability opment (USAID) grant called Strengthening Communities to conduct spatial analyses that provide insights into geo- through Integrated Programming (SCIP), which is a 5- graphic relationships. For example, spatial distribution of year multisector programme aimed at improving the HIV infection was recently mapped across SSA (Cuadros health and livelihoods of children, women and families in et al. 2013) as part of the concept of ‘Know your epi- Zambezia. Known locally as Ogumaniha, which means demic, know your response’ (UNAIDS 2007). This study ‘united for a common purpose’, SCIP is a consortium of found evidence of clustering of HIV infection at a mi- five international organizations. crogeographic scale (Cuadros et al. 2013). Additionally, the incidence of sexually transmitted infections (STIs) has Survey data been investigated and found to be associated with geographic areas of high HIV prevalence (Ramjee & Wand At programme initiation, Ogumaniha conducted a base- 2014). Analyses using GIS, such as spatial clustering line population-based survey from August 8 to September methods, have been used for disease and epidemic sur- 25, 2010. The protocol for data collection was approved veillance to not only better monitor HIV infection (Cuad- by the Mozambican National Bioethics Committee for ros et al. 2013), but also malaria incidence and Health (Comite Nacional de Bioetica em Saude [CNBS]) prevalence (Yeshiwondim et al. 2009; Haque et al. and the Institutional Review Board of Vanderbilt Univer- 2014), depression prevalence (Salinas-Perez et al. 2012), sity. The survey collected information on >500 variables drug adherence (Hoang et al. 2011), obesity (Turi et al. in eight dimensions and was designed to gather house- 2013) and create eco-epidemiological models to analyse hold information from the female heads of households.

C. P. Buehler et al. Spatial patterns of HIV knowledge

Alto Molócuè

Morrumbala

Namacurra Figure 1 Map of Zambezia Province, Mozambique, with location of the Quelimane provincial capital Quelimane and three focus districts: Alto Molocu e, Morrumbala and Namacurra. Black polygons represent the enumeration areas sampled for each of the three focus 0 20 40 80 120 160 districts. km

The design and implementation of this survey are detailed enterprises. The national highway, one of the few paved elsewhere: http://globalhealth.vanderbilt.edu/manage/ roads in the country, connects Alto Molocu e with com- wp-content/uploads/phase1and2_report.pdf. Briefly, sur- mercial centres to the south and the deep-sea ports of vey teams completed 3749 interviews in 259 enumeration Nampula Province to the north. The capital has better areas (EAs) across Zambezia Province. A large represen- established HIV care and treatment services (since 2006) tative sample (201 EAs) from three diverse districts (Alto that are housed within the main district referral hospital. Molocu e, Namacurra and Morrumbala) in Zambezia was Nauela, a much smaller peripheral health facility, began obtained to increase the precision of survey results while offering HIV services in January 2010, approximately minimising cost. This survey will be repeated at the pro- 7 months before survey administration. The national gramme’s end in mid-2014, allowing for an interrupted highway is integral to Zambezia’s commerce, and as time-series analysis of the impact of this large-scale, mul- such, Alto Molocu e inherently receives more commercial tisector intervention. exchange because of its geographic location relative to Of the 43 health facilities in our three focus districts, the national highway and its proximity to neighbouring 15 were in Alto Molocu e, 11 in Namacurra and 17 in commercial centres. Morrumbala. At the time of survey administration, Alto Molocu e and Namacurra had only two facilities offering District of Namacurra a complete package of HIV care and treatment services and Morrumbala had one. This complete package Namacurra district is located in the south of Zambezia includes pre-ART and ART care, prevention of mother- Province, approximately 45 min by car on the national to-child transmission (PMTCT), care for HIV-exposed highway from the provincial capital Quelimane. The dis- children, voluntary counselling and testing (VCT), pro- trict hospital in the capital, Vila da Namacurra, has been vider-initiated testing and counselling (PITC), care for offering HIV services since 2007. Macuze, the district’s TB/HIV coinfected patients, ART adherence support and second facility was capacitated with ART services in a variety of psychosocial support activities including HIV 2009. The port at Macuze is one of only a handful of education. natural deep water ports on Africa’s eastern coastline and is slated for large investments in infrastructure over the next 5–10 years, and as such, the surrounding areas have District of Alto Molocu e received significant focus in both development and health Alto Molocu e is one of the more economically devel- interventions over the last decade. The southern and east- oped districts of Zambezia with several commercial ern regions of the district can be difficult to reach with

C. P. Buehler et al. Spatial patterns of HIV knowledge

roads that flood and suffer significant washout during the language measure of reading and arithmetic (Ciampa rainy season. et al. 2012a,b). Scores for literacy ranged from 0 to 57. Respondents were characterised as having low literacy if they had scores 0–15 (US grade level: kindergarten or District of Morrumbala less), intermediate literacy if their scores were 16–56, and Morrumbala district lies on the western border of high literacy if they had scores of 57, corresponding to Zambezia Province. At the time of survey administration, reading the entire word list properly (Ciampa et al. only Vila da Morrumbala was offering HIV services (ini- 2012a,b). The numeracy subscale consisted of 15 orally tiated 2006). The national highway crosses Morrumbala’s administered items augmented by a visual aid card, scores south-western tip, which allows for significant commer- ranged from 0 to 15. A perfect score of 15 was the US cial exchange for persons living in proximity. Geographi- equivalent of 1st grade mathematics skills. Both items cally, much of the district is isolated from government were field tested prior to the start of data collection (Ci- investments because of poor road accessibility. However, ampa et al. 2012a,b). the north-western edge of the district shares a border with Malawi, which hosts substantial commercial Spatial analysis exchanges for this region. It is not uncommon for â Mozambicans to seek health services across the border in Mapping analyses were conducted using ArcGIS version Malawi where health facilities are better resourced. 10.2 (ESRI, Redlands, CA, USA). Data were projected into Moznet_UTM_Zone_36S. District boundaries, administrative units and road data were obtained from Household characteristics and calculation of HIV DIVA-GIS Free Spatial Data (www.diva-gis.org/gdata). knowledge score These data were originally obtained from Global Admin- Household characteristics and data pertaining to HIV istrative Areas (GADM) (www.gadm.org) and the knowledge from the three focus districts were analysed. Defense Mapping Agency (DMA) Digital Chart of the HIV knowledge scores were measured using a previously World, Fairfax, Virginia, 1992. validated HIV knowledge assessment tool with five ques- A database containing aggregated household character- tions (Ciampa et al. 2012a,b; Mukolo et al. 2013a,b). istics and median HIV knowledge scores from each EA Scale domains included adult-to-adult and mother-to- was joined to the EA layer polygon shape file. Records child transmission routes as well as potential transmission were joined based on each EAs unique identification by casual contact (Mukolo et al. 2013a,b). HIV knowl- code. Distance to nearest facility offering ART was calcu- edge scores ranged from 0 to 9 points (Table 1). For GIS lated as minimum Euclidean distance between each EA mapping, data were aggregated at the EA-level to main- feature and the nearest ART facility. This was calculated tain respondent anonymity. Median values of HIV with the ‘Near’ tool in the ‘Analysis’ tab of ArcToolbox. knowledge scores were calculated for each EA. Only health facilities offering complete HIV care and treatment services were included in this analysis. Clustering of HIV knowledge scores was assessed using Determination of literacy and numeracy score Optimized Hot Spot analysis within the ‘Spatial Statistics’ Literacy and numeracy scores were determined using a tool of ArcToolbox. The Optimized Hot Spot analysis modified version of the WRAT-3, a validated English uses the Getis-Ord Gi* statistic, which has been used in a

Table 1 Ogumaniha HIV Knowledge Scale: Five questions and origin of 9-point scoring system

Question Possible Score

In what ways can one adult man or woman transmit HIV to another man or woman? 0 = DK/none, 1 = 1 correct, 2 = 2 correct In what ways can a woman with HIV pass it to her baby? 0 = DK/none, 1 = 1 correct, 2 = 2 correct How can HIV transmission from an adult man or woman to another be prevented? 0 = DK/none, 1 ==1 correct, 2 = 2 correct How can HIV transmission from mother to a child be prevented? 0 = DK/none, 1 = 1 correct, 2 = 2 correct Do you think there is a cure for HIV/AIDS? 0 = DK/no, 1 = correct Total Possible Range 0–9

HIV knowledge was measured with ﬁve items adapted from the Demographic Health Survey V that assess knowledge of HIV transmission, transmission prevention and HIV treatment (Ciampa et al. 2012a,b). Scores ranged from 0 to 9. DK, do not know.

C. P. Buehler et al. Spatial patterns of HIV knowledge

variety of public health studies to determine spatial clus- analyse socio-demographic characteristics which may tering (Ord & Getis 1995; Yeshiwondim et al. 2009; Ho- influence one0s health knowledge (Ciampa et al. 2012a, ang et al. 2011; Turi et al. 2013; Haque et al. 2014). b). R-software 3.0.2 (www.r-project.org) was used for The full technical aspects of this model are also described statistical analyses. elsewhere (ESRI. ArcGIS Resources: Optimized Hot Spot Analysis (Spatial Statistics) 2014; Ord & Getis 1995) Results Briefly, locational outliers more than three standard devi- ations away from their closest non-coincident neighbour Alto Molocu e had the highest median HIV knowledge were identified and removed from analysis. A standard score of the three districts, with 22 of 51 EAs (43%) hav- distance calculation was used to determine the optimal ing median HIV knowledge ≥4 points. Two EAs had distance band based on the average distance to 30 nearest median scores of 6.0 and were located <1 km from the neighbours (after correction for outliers). To determine main capital health facility Vila da Alto Molocu e. Signifi- whether features with high values or features with low cant clustering (P < 0.05) of higher HIV knowledge exists values cluster in a geographic area, each feature, within within the 10-km radius of Vila da Alto Molocu e the context of its neighbouring features, was analysed. (Figure 2). In the south-west region of the district, one The local sum for a feature and its neighbours was com- distal EA (about 50 km away by road from Vila da Alto pared proportionally to the sum of all features. When the Molocu e) had a median score of 1.0. This EA was identi- local sum was much different than the expected local fied as a significant cold spot. No significant clustering of sum, the feature was characterised as a hot spot (or a EAs with either higher or lower HIV knowledge scores cold spot) and a large Z-score was produced (Ord & Ge- was located near the Nauela health facility. tis 1995). Large Z-scores of 1.96 and 2.58 correspond In comparison, only three of 57 EAs (5%) in Nam- to 95 (P < 0.05) and 99 (P < 0.01) percentiles, respec- acurra (compared to 43% in Alto Molocu e) had HIV tively, suggesting that an observed pattern is unlikely to knowledge scores ≥4 points. These higher scoring EAs be due to random chance. were located within 2 km of the Macuze health facility. Hot spots and cold spots of higher and lower HIV Unsurprisingly, significant hot spots (P < 0.05) were also knowledge scores and their proximity to the health facili- identified within the 10-km radius of Mucuze (Figure 3). ties offering ART are represented in Figures 2–4. EAs The most frequently occurring knowledge scores in Nam- with larger positive Z-scores (hot spots) signify a more acurra were between 0.0 and 1.0 (n = 31) with 15 of 57 intense clustering of higher HIV knowledge score values EAs (25%) have knowledge scores of 0.0. A majority of (red). Conversely, larger negative Z-scores (cold spots) EAs with scores ≤3.0 were isolated in the south-east indicate a more intense clustering of lower HIV knowl- region of the district. This region was also identified as edge score values (blue). For scale, a 10-km radius an area containing significant cold spots (P < 0.05) of around each ART facility represents the travel distance lower HIV knowledge scores. No significant clustering of by foot in Mozambique that is thought acceptable for either higher or lower knowledge scores was located near purposes of ART adherence and retention in care. the main district health facility at Vila da Namacurra. Within Morrumbala, eight of 93 EAs (9%) had median HIV knowledge scores ≥4 points. Of those eight EAs, five Statistical analysis EAs were located in the north-west corner of the district, Multivariable linear regression allowed for covariate- two EAs were located <2 km from the district capital adjusted estimation of the relationship between distance health facility, and one EA was located west of the dis- to nearest clinic offering ART and HIV knowledge. Co- trict health facility on the border of the Zambezi River. variates included district, respondent age, years of educa- Almost half (43%) of the EAs in Morrumbala (40 of 93) tion, number of children <5 years living in the house, had median HIV knowledge scores of 0.0. Hot spots household size, literacy and numeracy score. Robust (P < 0.05) of higher HIV knowledge scores are in prox- covariance matrix estimates were used to correct for cor- imity to the 10-km radius around the district capital and related responses within EA. Missing values of covariates along the national highway (Figure 4). Hot spots of were multiply imputed to prevent casewise deletion (Ru- higher HIV knowledge were also located in the north- bin & Little 1987). To account for non-linearity, respon- west region of the district, where it is known that resi- dent age, years of education, children <5 living in home dents seek HIV services across the border in Malawi. and distance to health facility were included in the model Of 2920 female heads of household interviewed in the using restricted cubic splines (Harrell 2001). Covariates three focus districts, the median age was 32 years, the in the multivariable model were chosen a priori to median household size was four persons, and the median

C. P. Buehler et al. Spatial patterns of HIV knowledge

Nauela

Vila da Alto Molócuè

HIV knowledge score clustering Gi* Z-score Health facility offering ART Figure 2 Clusters of Higher and Lower National highway HIV Knowledge Scores, District of Alto Unpaved road Molocu e. EAs in red indicate a hot spot Gizscore of higher HIV knowledge scores. EAs in >–2.58 blue indicate a cold spot of lower HIV –2.57 – –1.96 knowledge scores. A 10-km radius around –1.95 – 1.96 each health facility shows walking 1.961 – 2.58 0 5 10 20 km distance considered appropriate for ART >2.58 care and retention. number of children <5 living in the home was one. HIV knowledge increased 0.05 points [0.05 (95% CI: Respondents from Alto Molocu e had higher educational 0.01–0.10)]. attainment as well as higher literacy and numeracy scores While literacy scores were not signiﬁcantly correlated compared with Namacurra and Morrumbala (Table 2). with higher HIV knowledge, numeracy scores were Of the women interviewed in our study, age, higher (P < 0.001). For every 5-point increment in the numeracy education, increasing household size and closer proximity score, HIV knowledge increased almost half a point [0.41 to a health facility correlated with higher HIV knowl- (95% CI: 0.30–0.51)]. Respondents in EAs closer to a edge. A respondent aged 30 years scored on average a health facility offering ART were signiﬁcantly half a point higher on the HIV knowledge scale com- (P < 0.001) more likely to have higher HIV knowledge. pared with a 20-year-old respondent [0.43 (95% CI: There is a sharp decrease in HIV knowledge as travel dis- 0.20–0.66)]. Compared with no education, a respondent tance to a health facility offering HIV services increased with an eighth-grade education scored on average 1.3 from 0 to 25 km. At a distance of 5 km from a health points higher. For every one person in the household, facility, respondents scored half a point higher on the

C. P. Buehler et al. Spatial patterns of HIV knowledge

Vila da Namacurra

Macuze HIV knowledge score clustering Gi* Z-score Figure 3 Clusters of Higher and Lower Health facility offering ART HIV Knowledge Scores, District of National highway Namacurra. EAs in red indicate a hot Unpaved road spot of higher HIV knowledge scores. Gizscore EAs in blue indicate a cold spot of lower >–2.58 HIV knowledge scores. A 10-km radius –2.57 – –1.96 around each health facility shows walking –1.95 – 1.96 distance considered appropriate for ART 1.961 – 2.58 0 3 6 12 km care and retention. >2.58

HIV knowledge scale compared to those living over certain by any means. Understanding how an individual’s 35 km [0.59 (95% CI: 0.36–0.82)] (Table 3; Figure 5). knowledge about HIV helps them engage in prevention Misperceptions about HIV have been linked to nega- and care, both within and outside of hot spots (higher tive health seeking behaviours, such as delay in seeking HIV knowledge) and cold spots (lower HIV knowledge), testing and care. Pinpointing cold spots with lower HIV remains an implementation science research priority. knowledge allows planners to better target specific communities for interventions. We found HIV knowledge to Discussion decrease with increasing travel distance from a health facility offering HIV services. Rural and/or difficult-to- We used ArcGIS to characterise the spatial patterns of reach areas have been highlighted by Ogumaniha for HIV knowledge with data from 201 enumeration areas HIV education interventions and community-based HIV collected in a 2010 population-based cluster survey. HIV testing. Interventions that target the highest priority knowledge scores (0–9 points) were determined for each demographic subgroups in the most vulnerable communi- EA. Hot spot analysis identified significant clusters of ties might have the largest impact, although this is not higher and lower HIV knowledge. Higher HIV knowl-

C. P. Buehler et al. Spatial patterns of HIV knowledge

Vila da Morrumbala

HIV knowledge score clustering Gi* Z-score Health facility offering ART Figure 4 Clustering of Higher and Lower National highway HIV Knowledge Scores, District of Unpaved road Morrumbala. EAs in red indicate a hot Gizscore spot of higher HIV knowledge scores. >–2.58 EAs in blue indicate a cold spot of lower –2.57 – –1.96 HIV knowledge scores. A 10-km radius –1.95 – 1.96 around each health facility shows walking 1.961 – 2.58 01020 40km distance considered appropriate for ART >2.58 care and retention. edge scores tended to be in closer proximity to a health about HIV/AIDS as of survey administration in 2010. facility offering ART. A multivariable linear regression Misperceptions about HIV can affect risk behaviours model estimated HIV knowledge associations with dis- related to HIV transmission and can represent barriers tance to nearest clinic offering antiretroviral therapy, to health care with delays in HIV testing and enrol- respondent age, education, household size, number of ment into care. children under five, numeracy, literacy and district of res- Household characteristics associated with higher HIV idence. knowledge among our female respondents were not sur- It was disappointing to find very low overall HIV prising: larger household size, increased years of educa- knowledge in all three districts (score values ≤4 points tion and increased age. This may reflect increased of nine points). Low HIV knowledge scores at baseline involvement in antenatal care of this age group and a suggests that the investments in scaling-up HIV pro- source of HIV knowledge acquisition. Greater numeracy grammes in Zambezia Province, which began in 2006, skills and closer proximity to a health facility offering had not yet translated into widespread knowledge ART were also associated with higher knowledge.

C. P. Buehler et al. Spatial patterns of HIV knowledge

Table 2 Household and Respondent Characteristics by District in Zambezia Province

Alto Molocu e Namacurra Morrumbala Total Household characteristics median (IQR) (n = 815) (n = 810) (n = 1295) (n = 2920)

HIV Knowledge score 3 (2–5) 1 (0–3) 1 (0–4) 2 (0–4) Age of respondent 28 (22–37) 34 (25–43) 34 (27–45) 32 (24–41) Years of education 3 (2–5) 0 (0–3) 0 (0–2) 0 (0–3) Distance of EA from health facility (km) 17.5 (11.4–24.3) 12.3 (8.8–16.2) 38.2 (22.6–69.3) 20.3 (12–38.2) Number Children <51(0–2) 1 (0–1) 1 (0–2) 1 (0–2) Household size 4 (3–6) 4 (3–5) 5 (3–6) 4 (3–6) Literacy score (57-point scale) 10 (0–57) 0 (0–3) 0 (0–0) 0 (0–5) Numeracy score (15-point scale) 15 (11–15) 9 (6–13) 10 (6–11) 10 (7–14)

Continuous variables are reported as weighted estimates of median (IQR: interquartile range), with each observation being weighted by the inverse of the household sampling probability.

– Table 3 Effects from ordinary least squared modelling in assess- Other studies in SSA have found that women aged 15 ing predictors of HIV knowledge in rural Zambezia Province, 24 years had lower levels of accurate and comprehensive Mozambique (n = 2920) HIV knowledge than their male counterparts (UNAIDS 2013). Overall educational achievement among the Effect (95% CI) P-value women in our study was very low. Of women who Age (years) attended school, HIV knowledge increased with increased 20 (ref) 0 0.002 schooling. A previous study in Zambezia found that peo- 30 0.43 (0.20, 0.66) ple who learned about HIV prevention in school had the 40 0.38 (0.14, 0.62) highest odds of knowing multiple ways to avoid infection Education (years) (Audet et al. 2012). While schools are not the only venue 0 (ref) 0 <0.001 4 0.41 (0.16, 0.66) for learning about how to avoid HIV infection, it does 8 1.30 (0.90, 1.70) indicate that those who have less education are likely Children under five: 4 vs. 2 0.50 (0.94, 0.07) 0.045 missing out on important HIV educational opportunities. Household size 0.05 (0.01, 0.10) 0.019 The prevalence of illiteracy is substantial in Zambezia (per one person) (Ciampa et al. 2012a,b). While literacy per se showed no Literacy score 0.01 ( 0.04, 0.02) 0.610 association with HIV knowledge in our study, increasing (per five points) numeracy score, the basic understanding of numeric con- Numeracy score 0.41 (0.30, 0.51) <0.001 (per five points) cepts and computational skills, was found to be associ- Distance to nearest ART facility (km) ated with higher HIV knowledge. 0 0.87 (0.57, 1.17) <0.001 In a previous study, women who primarily spoke Echu- 5 0.59 (0.36, 0.82) abo, a local language, had lower levels of HIV knowledge 10 0.33 (0.11, 0.55) than Portuguese-speaking women, the national language 25 0.02 ( 0.09, 0.12) taught in schools (Ciampa et al. 2012a,b). HIV educational 35 (ref) 0 interventions that (i) focus on schools, (ii) are incorporated District Alto Molocu e (ref) <0.001 into income generation projects for women’s groups, (iii) Morrumbala 1.10 (1.33, 0.87) are incorporated into interventions for female sex workers Namacurra 0.97 (1.19, 0.75) and (iv) promote the scale-up of interventions aimed at increasing access to antenatal care, and PMTCT pro- This model accounts for correlated responses from individuals in grammes have been employed by Ogumaniha to increase the same enumeration areas using the Huber–White robust method to adjust the variance–covariance matrix (Rubin & Little HIV knowledge in this target population. It will be impor- 1987; Harrell 2001). tant to assess in the planned 2014 post-intervention survey whether these interventions have had an impact.

Community characteristics were also predictive; more Strengths and limitations remote areas that were susceptible to ﬂooding and/or isolation in the rainy season had residents who were less A study strength includes the representative sampling knowledgeable about HIV/AIDS. scheme of our survey. We used GIS to ascertain geo-

C. P. Buehler et al. Spatial patterns of HIV knowledge

Distance to HF Numeracy Age Education

0 20406080100 0 5 10 15 20 30 40 50 60 70 80 0246810

Children under 5 Districts Household size Literacy HIV knowledge score 4

02468ALM MOR NAM 24681012 0102030405060

Figure 5 Effects from Ordinary Least Squared Modelling in Assessing Predictors of HIV Knowledge in Rural Zambezia Province, Mozambique (n = 2920). HF, health facility offering HIV services; ALM, Alto Molocu e; MOR, Morrumbala; NAM, Namacurra. graphic clusters of high or low HIV knowledge and iden- important, as all districts have both spatial and non- tified household characteristics that were associated with spatial factors that easily act as obstacles to accessing higher or lower HIV knowledge. Contextualizing HIV health care. knowledge geographically allows planners to allocate We found poor overall HIV knowledge in all three rural resources and focus community-based HIV outreach ini- districts studied, suggesting little impact of prior HIV tiatives and ART expansion planning more strategically. treatment and prevention investments on overall HIV A study limitation includes our plausible but unvalidated knowledge. By establishing HIV knowledge scores at base- assumption that people utilise the nearest health facility line (this 2010 survey), our anticipated 2014 survey at the offering ART services within the district. Distance to end of the Ogumaniha interventions can help assess their health facility was calculated using Euclidian distance impact on HIV knowledge in these communities. between the EA and the health facility offering ART in one dimension. We did not consider some spatial factors Acknowledgements such as road type, road conditions or barriers such as mountains and rivers that influence true travel distance We are grateful for the contributions of the Ogumaniha- (Martin et al. 2008). Nonetheless, Euclidean distance has SCIP consortium members (led by World Vision) to the been a valid proxy for actual travel distance/time (Car- baseline survey and for permission to analyse the survey lucci et al. 2008; Jones et al. 2010; Cudnik et al. 2012). data. Disclaimer: The contents of this manuscript are the We also did not consider non-spatial barriers to health responsibility of the authors and do not necessarily reflect care such as socio-economic status. Future work looking the views of USAID, the United States Government or at both proximity and access (Yao et al. 2013) will be World Vision.

C. P. Buehler et al. Spatial patterns of HIV knowledge

References Harrell FE (2001) Regression Modeling Strategies: With Applica- tions to Linear Models, Logistic Regression, and Survival Al-Taiar A, Clark A, Longenecker JC & Whitty CJM (2010) Analysis. Springerr-Verlag, New York, Inc. Physical accessibility and utilization of health services in Hoang C, Kolenic G, Kline-Rogers E, Eagle KA & Erickson SR Yemen. International Journal of Health Geographics 9, 38. (2011) Mapping geographic areas of high and low drug adher- Audet CM, Sidat M, Blevins M, Moon TD, Vergara A & Ver- ence in patients prescribed continuing treatment for acute cor- mund SH (2012) HIV knowledge and health-seeking behavior onary syndrome after discharge. Pharmacotherapy: The in Zambezia Province, Mozambique. SAHARA-J: Journal of Journal of Human Pharmacology and Drug Therapy, 31, 927– – Social Aspects of HIV/AIDS, 9,41 46. 933. Blanford JI, Kumar S, Luo W & MacEachren AM (2012) It’s a INSIDA Demographic and Health Surveys Program 2010. long, long walk: accessibility to hospitals, maternity and inte- Inquerito Nacional de Prevalência, Riscos Comportamentais e grated health centers in Niger. International Journal of Health Informacßao~ sobre o HIV e SIDA em Mocßambique (2009) at – Geographics 11,1 15. http://dhsprogram.com/pubs/pdf/AIS8/AIS8.pdf. Brauner-Otto SR, Axinn WG & Ghimire DJ (2007) The spread Instituto Nacional de Estatıstica Ministerio da Saude (2013) of health services and fertility transition. Demography 44, Mozambique Inquerto Demograficio e de Saude 2011. Instituto – 747 770. Nacional de Estatıstica Ministerio da Saude, Maputo, Mozam- Carlucci JG, Kamanga A, Sheneberger R et al. (2008) Predictors bique. http://dhsprogram.com/pubs/pdf/FR266/FR266.pdf. of adherence to antiretroviral therapy in rural Zambia. Journal Jones SG, Ashby AJ, Momin SR & Naidoo A (2010) Spatial – of Acquired Immune Deficiency Syndromes 47, 615 622. implications associated with using euclidian distance measure- Ciampa PJ, Skinner SL, Patricio SR, Rothman RL, Vermund ments and geographic centroid imputation in health care SH & Audet CM (2012a) Comprehensive knowledge of research. Health Services Research. 45, 316–327. HIV among women in rural Mozambique: development and Martin D, Jordan H & Roderick P (2008) Taking the bus: incorpo- validation of the HIV knowledge 27 scale. PLoS ONE 7, rating public transport timetable data into health care accessibil- e48676. ity modelling. Environment and Planning A 40, 2510–2525. Ciampa PJ, Vaz LME, Blevins M et al. (2012b) The association Moon TD, Burlison JR, Sidat M et al. (2010) Lessons learned among literacy, numeracy, HIV knowledge and health-seeking while implementing an HIV/AIDS care and treatment program behavior: a population-based survey of women in rural in rural Mozambique. Retrovirology: Research and Treatment, Mozambique. PLoS ONE 7, e39391. 3,1–14. Cromley EK & McLafferty SL (2012) GIS and Public Health, Moon TD, Burlison JR, Blevins M et al. (2011) Enrolment and 2nd edn. Guilford Press, New York. programmatic trends and predictors of antiretroviral therapy Cuadros DF, Awad SF & Abu-Raddad LJ (2013) Mapping HIV initiation from President’s Emergency Plan for AIDS Relief clustering: a strategy for identifying populations at high risk of (PEPFAR)-supported public HIV care and treatment sites in HIV infection in sub-Saharan Africa. International Journal of rural Mozambique. International Journal of STD & AIDS 22, Health Geographics 12, 28. 621–627. Cudnik MT, Yao J, Zive D, Newgard C & Murray AT (2012) Mukolo A, Blevins M, Hinton N et al. (2013a) Negative labeling Surrogate markers of transport distance for out-of-hospital and social exclusion of people living with human immunodefi- – cardiac arrest patients. Prehospital Emergency Care 16, 266 ciency virus/acquired immune deficiency syndrome in the anti- 272. retroviral therapy era: insight from attitudes and behavioral De Cock KM, El-Sadr WM & Ghebreyesus TA (2011) Game intentions of female heads of households in Zambezia Prov- changers: why did the scale-up of HIV treatment work despite ince, Mozambique. AIDS Care, 26,1–6 doi:10.1080/ weak health systems? Journal of Acquired Immune Deficiency 09540121.2013.861570. – Syndromes 57(Suppl. 2), S61 S63. Mukolo A, Blevins M, Victor B, Vaz LM, Sidat M & Vergara A El-Sadr WM, Holmes CB, Mugyenyi P, Thirumurthy H, Eller- (2013b) Correlates of social exclusion and negative labeling brock T & Ferris R (2012) Scale-up of HIV treatment through and devaluation of people living with HIV/AIDS in rural set- PEPFAR: a historic public health achievement. Journal of tings: evidence from a general household survey in Zambezia Acquired Immune Deficiency Syndromes 60(Suppl. 3), S96. Province, Mozambique. PLoS ONE 8, e75744. ESRI. ArcGIS Resources: Optimized Hot Spot Analysis (Spatial Ord JK & Getis A (1995) Local spatial autocorrelation statistics: Statistics) (2014) (Version 10.2.2). Redlands, CA: ESRI. distributional issues and an application. Geographical Analysis Retrieved from http://resources.arcgis.com/en/help/main/10.2/ 27, 286–306. index.html#//005p00000058000000. Ramjee G & Wand H (2014) Geographical clustering of high Global HIV/AIDS Response: Epidemic update and health sector risk sexual behaviors in “hot-spots” for HIV and sexually progress towards universal access (2011). Retrieved from transmitted infections in Kwazulu-Natal, South Africa. AIDS http://www.who.int/hiv/pub/progress_report2011/en/. and Behavior 18, 317–322. Haque U, Overgaard HJ, Clements ACA et al. (2014) Malaria Richardson DB, Volkow ND, Kwan MP, Kaplan RM, Good- burden and control in Bangladesh and prospects for elimina- child MF & Croyle RT (2013) Spatial turn in health research. tion: an epidemiological and economic assessment. The Lancet Science 339, 1390–1392. Global Health, 2, e98–e105.

C. P. Buehler et al. Spatial patterns of HIV knowledge

Rubin D & Little R (1987) Statistical Analysis with Missing from http://data.unaids.org/pub/report/2008/jc1535_annual_re- Data. John Wiley and Sons, New York. port07_en.pdf. Salinas-Perez JA, Garcıa-Alonso CR, Molina-Parrilla C, Jorda- UNAIDS Joint United Nations Programme on HIV/AIDS (2013) Sampietro E & Salvador-Carulla L (2012) Identiﬁcation and UNAIDS report on the global AIDS epidemic 2013. Retrieved location of hot and cold spots of treated prevalence of depres- from http://www.unaids.org/en/resources/campaigns/globalre- sion in Catalonia (Spain). International Journal of Health port2013/globalreport/. Geographics 11, 36. Vermund SH, Sidat M, Weil LF, Tique JA, Moon TD & Ciampa Tanser F, Gijsbertsen B & Herbst K (2006) Modelling and PJ (2012) Transitioning HIV care and treatment programs in understanding primary health care accessibility and utilization southern Africa to full local management. AIDS (London, in rural South Africa: an exploration using a geographical England), 26, 1303–1310. information system. Social Science & Medicine 63, 691–705. Yao J, Murray AT, Agadjanian V & Hayford SR (2012) Geo- Thornton RL (2008) The Demand for, and impact of, learning graphic inﬂuences on sexual and reproductive health service HIV status. The American Economic Review 98, 1829–1863. utilization in rural Mozambique. Applied Geography 32, 601– Torbick N, Hession S, Stommel E & Caller T (2014) Mapping 607. amyotrophic lateral sclerosis lake risk factors across northern Yao J, Murray AT & Agadjanian V (2013) A geographical New England. International Journal of Health Geographics perspective on access to sexual and reproductive health care 13,1. for women in rural Africa. Social Science & Medicine 96, Turi KN, Christoph MJ & Grigsby-Toussaint DS (2013) Spatial 60–68. distribution of underweight, overweight and obesity among Yeshiwondim AK, Gopal S, Hailemariam AT, Dengela DO women and children: results from the 2011 Uganda Demo- & Patel HP (2009) Spatial analysis of malaria incidence graphic and Health Survey. International Journal of Environ- at the village level in areas with unstable transmission mental Research and Public Health 10, 4967–4981. in Ethiopia. International Journal of Health Geographics UNAIDS Joint United Nations Programme on HIV/AIDS (2007) 8,5. UNAIDS Annual Report: Knowing your epidemic. Retrieved

Corresponding Author Charlotte P.Buehler, Vanderbilt Institute for Global Health, 2525 West End Avenue, Suite 750, Nashville, Tennessee, 37203, USA. Tel.: +1-615-343-8264; Fax +1-615-343-7797; E-mail: [email protected]