A comparison of spatial smoothing methods for small area estimation with sampling weights

Bayesian modeling of spatial ordinal data from health surveys

Health surveys allow exploring health indicators that are of great value from a public health point of view and that cannot normally be studied from regular health registries. These indicators are usually coded as ordinal variables and may depend on covariates associated with individuals. In this article, we propose a Bayesian individual-level model for small-area estimation of survey-based health indicators. A categorical likelihood is used at the first level of the model hierarchy to describe the ordinal data, and spatial dependence among small areas is taken into account by using a conditional autoregressive distribution. Post-stratification of the results of the proposed individual-level model allows extrapolating the results to any administrative areal division, even for small areas. We apply this methodology to describe the geographical distribution of a self-perceived health indicator from the Health Survey of the Region of Valencia (Spain) for the year 2016.

Bayesian spatio-temporal analysis of malaria prevalence in children between 2 and 10 years of age in Gabon

BACKGROUND

Gabon still bears significant malaria burden despite numerous efforts. To reduce this burden, policy-makers need strategies to design effective interventions. Besides, malaria distribution is well known to be related to the meteorological conditions. In Gabon, there is limited knowledge of the spatio-temporal effect or the environmental factors on this distribution. This study aimed to investigate on the spatio-temporal effects and environmental factors on the distribution of malaria prevalence among children 2-10 years of age in Gabon.

METHODS

The study used cross-sectional data from the Demographic Health Survey (DHS) carried out in 2000, 2005, 2010, and 2015. The malaria prevalence was obtained by considering the weighting scheme and using the space-time smoothing model. Spatial autocorrelation was inferred using the Moran's I index, and hotspots were identified with the local statistic Getis-Ord General Gi. For the effect of covariates on the prevalence, several spatial methods implemented in the Integrated Nested Laplace Approximation (INLA) approach using Stochastic Partial Differential Equations (SPDE) were compared.

RESULTS

The study considered 336 clusters, with 153 (46%) in rural and 183 (54%) in urban areas. The prevalence was highest in the Estuaire province in 2000, reaching 46%. It decreased until 2010, exhibiting strong spatial correlation (P < 0.001), decreasing slowly with distance. Hotspots were identified in north-western and western Gabon. Using the Spatial Durbin Error Model (SDEM), the relationship between the prevalence and insecticide-treated bed nets (ITNs) coverage was decreasing after 20% of coverage. The prevalence in a cluster decreased significantly with the increase per percentage of ITNs coverage in the nearby clusters, and per degree Celsius of day land surface temperature in the same cluster. It slightly increased with the number of wet days and mean temperature per month in neighbouring clusters.

CONCLUSIONS

In summary, this study showed evidence of strong spatial effect influencing malaria prevalence in household clusters. Increasing ITN coverage by 20% and prioritizing hotspots are essential policy recommendations. The effects of environmental factors should be considered, and collaboration with the national meteorological department (DGM) for early warning systems is needed.

Mapping the prevalence of cancer risk factors at the small area level in Australia

BACKGROUND

Cancer is a significant health issue globally and it is well known that cancer risk varies geographically. However in many countries there are no small area-level data on cancer risk factors with high resolution and complete reach, which hinders the development of targeted prevention strategies.

METHODS

Using Australia as a case study, the 2017-2018 National Health Survey was used to generate prevalence estimates for 2221 small areas across Australia for eight cancer risk factor measures covering smoking, alcohol, physical activity, diet and weight. Utilising a recently developed Bayesian two-stage small area estimation methodology, the model incorporated survey-only covariates, spatial smoothing and hierarchical modelling techniques, along with a vast array of small area-level auxiliary data, including census, remoteness, and socioeconomic data. The models borrowed strength from previously published cancer risk estimates provided by the Social Health Atlases of Australia. Estimates were internally and externally validated.

RESULTS

We illustrated that in 2017-2018 health behaviours across Australia exhibited more spatial disparities than previously realised by improving the reach and resolution of formerly published cancer risk factors. The derived estimates revealed higher prevalence of unhealthy behaviours in more remote areas, and areas of lower socioeconomic status; a trend that aligned well with previous work.

CONCLUSIONS

Our study addresses the gaps in small area level cancer risk factor estimates in Australia. The new estimates provide improved spatial resolution and reach and will enable more targeted cancer prevention strategies at the small area level. Furthermore, by including the results in the next release of the Australian Cancer Atlas, which currently provides small area level estimates of cancer incidence and relative survival, this work will help to provide a more comprehensive picture of cancer in Australia by supporting policy makers, researchers, and the general public in understanding the spatial distribution of cancer risk factors. The methodology applied in this work is generalisable to other small area estimation applications and has been shown to perform well when the survey data are sparse.

A small area model to assess temporal trends and sub-national disparities in healthcare quality

Monitoring subnational healthcare quality is important for identifying and addressing geographic inequities. Yet, health facility surveys are rarely powered to support the generation of estimates at more local levels. With this study, we propose an analytical approach for estimating both temporal and subnational patterns of healthcare quality indicators from health facility survey data. This method uses random effects to account for differences between survey instruments; space-time processes to leverage correlations in space and time; and covariates to incorporate auxiliary information. We applied this method for three countries in which at least four health facility surveys had been conducted since 1999 - Kenya, Senegal, and Tanzania - and estimated measures of sick-child care quality per WHO Service Availability and Readiness Assessment (SARA) guidelines at programmatic subnational level, between 1999 and 2020. Model performance metrics indicated good out-of-sample predictive validity, illustrating the potential utility of geospatial statistical models for health facility data. This method offers a way to jointly estimate indicators of healthcare quality over space and time, which could then provide insights to decision-makers and health service program managers.

A Novel Bayesian Spatial-Temporal Approach to Quantify SARS-CoV-2 Testing Disparities for Small Area Estimation

Objectives. To propose a novel Bayesian spatial-temporal approach to identify and quantify severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) testing disparities for small area estimation. Methods. In step 1, we used a Bayesian inseparable space-time model framework to estimate the testing positivity rate (TPR) at geographically granular areas of the census block groups (CBGs). In step 2, we adopted a rank-based approach to compare the estimated TPR and the testing rate to identify areas with testing deficiency and quantify the number of needed tests. We used weekly SARS-CoV-2 infection and testing surveillance data from Cameron County, Texas, between March 2020 and February 2022 to demonstrate the usefulness of our proposed approach. Results. We identified the CBGs that had experienced substantial testing deficiency, quantified the number of tests that should have been conducted in these areas, and evaluated the short- and long-term testing disparities. Conclusions. Our proposed analytical framework offers policymakers and public health practitioners a tool for understanding SARS-CoV-2 testing disparities in geographically small communities. It could also aid COVID-19 response planning and inform intervention programs to improve goal setting and strategy implementation in SARS-CoV-2 testing uptake. (Am J Public Health. 2023;113(1):40-48. https://doi.org/10.2105/AJPH.2022.307127).

Neighbourhood income inequalities in mental health in Barcelona 2001-2016: a Bayesian smoothed estimate

OBJECTIVE

Obtaining reliable health estimates at the small area level (such as neighbourhoods) using survey data usually poses the problem of small sample sizes. To overcome this limitation, we explored smoothing techniques in order to estimate poor mental health prevalence at the neighbourhood level and analyse its profile by income in Barcelona city (Spain).

METHOD

A Bayesian smoothing model with a logit-normal transformation was applied to four repeated cross-sectional waves of the Barcelona health survey for 2001, 2006, 2011 and 2016. Mental health status was identified from the 12-item General Health Questionnaire. Income inequalities were analysed with neighbourhood income in quantiles for each year and trends in the pooled analysis.

RESULTS

The prevalence of poor mental health ranged from 14.6% in 2001 to 18.9% in 2016. The yearly difference between neighbourhoods was 12.4% in 2001, 16.7% in 2006, 14.2% in 2011, and 20.0% in 2016. The odds ratio and 95% credible interval (95%CI) of experiencing poor mental health was 1.40 times higher (95%CI: 1.02-1.91) in less advantaged neighbourhoods than in more advantaged neighbourhoods in 2001, 1.61 times higher (95%CI: 1.01-2.59) in 2006 and 2.31 times higher (95%CI: 1.57-3.40) in 2016.

CONCLUSIONS

This study shows that the Bayesian smoothed techniques allows detection of inequalities in health in neighbourhoods and monitoring of interventions against them. In Barcelona, mental health problems are more prevalent in low-income neighbourhoods and raised in 2016.

Variation in smoking attributable all-cause mortality across municipalities in Belgium, 2018: application of a Bayesian approach for small area estimations

BACKGROUND

Smoking is one of the leading causes of preventable mortality and morbidity worldwide, with the European Region having the highest prevalence of tobacco smoking among adults compared to other WHO regions. The Belgian Health Interview Survey (BHIS) provides a reliable source of national and regional estimates of smoking prevalence; however, currently there are no estimates at a smaller geographical resolution such as the municipality scale in Belgium. This hinders the estimation of the spatial distribution of smoking attributable mortality at small geographical scale (i.e., number of deaths that can be attributed to tobacco). The objective of this study was to obtain estimates of smoking prevalence in each Belgian municipality using BHIS and calculate smoking attributable mortality at municipality level.

METHODS

Data of participants aged 15 + on smoking behavior, age, gender, educational level and municipality of residence were obtained from the BHIS 2018. A Bayesian hierarchical Besag-York-Mollie (BYM) model was used to model the logit transformation of the design-based Horvitz-Thompson direct prevalence estimates. Municipality-level variables obtained from Statbel, the Belgian statistical office, were used as auxiliary variables in the model. Model parameters were estimated using Integrated Nested Laplace Approximation (INLA). Deviance Information Criterion (DIC) and Conditional Predictive Ordinate (CPO) were computed to assess model fit. Population attributable fractions (PAF) were computed using the estimated prevalence of smoking in each of the 589 Belgian municipalities and relative risks obtained from published meta-analyses. Smoking attributable mortality was calculated by multiplying PAF with age-gender standardized and stratified number of deaths in each municipality.

RESULTS

BHIS 2018 data included 7,829 respondents from 154 municipalities. Smoothed estimates for current smoking ranged between 11% [Credible Interval 3;23] and 27% [21;34] per municipality, and for former smoking between 4% [0;14] and 34% [21;47]. Estimates of smoking attributable mortality constituted between 10% [7;15] and 47% [34;59] of total number of deaths per municipality.

CONCLUSIONS

Within-country variation in smoking and smoking attributable mortality was observed. Computed estimates should inform local public health prevention campaigns as well as contribute to explaining the regional differences in mortality.

Spatiotemporal Variation and Predictors of Unsuppressed Viral Load among HIV-Positive Men and Women in Rural and Peri-Urban KwaZulu-Natal, South Africa

Unsuppressed HIV viral load is an important marker of sustained HIV transmission. We investigated the prevalence, predictors, and high-risk areas of unsuppressed HIV viral load among HIV-positive men and women. Unsuppressed HIV viral load was defined as viral load of ≥400 copies/mL. Data from the HIV Incidence District Surveillance System (HIPSS), a longitudinal study undertaken between June 2014 to June 2016 among men and women aged 15−49 years in rural and peri-urban KwaZulu-Natal, South Africa, were analysed. A Bayesian geoadditive regression model which includes a spatial effect for a small enumeration area was applied using an integrated nested Laplace approximation (INLA) function while accounting for unobserved factors, non-linear effects of selected continuous variables, and spatial autocorrelation. The prevalence of unsuppressed HIV viral load was 46.1% [95% CI: 44.3−47.8]. Predictors of unsuppressed HIV viral load were incomplete high school education, being away from home for more than a month, alcohol consumption, no prior knowledge of HIV status, not ever tested for HIV, not on antiretroviral therapy (ART), on tuberculosis (TB) medication, having two or more sexual partners in the last 12 months, and having a CD4 cell count of <350 cells/μL. A positive non-linear effect of age, household size, and the number of lifetime HIV tests was identified. The higher-risk pattern of unsuppressed HIV viral load occurred in the northwest and northeast of the study area. Identifying predictors of unsuppressed viral load in a localized geographic area and information from spatial risk maps are important for targeted prevention and treatment programs to reduce the transmission of HIV.

A time-series approach to mapping livestock density using household survey data

More than one billion people rely on livestock for income, nutrition, and social cohesion, however livestock keeping can facilitate disease transmission and contribute to climate change. While data on the distribution of livestock have broad utility across a range of applications, efforts to map the distribution of livestock on a large scale are limited to the Gridded Livestock of the World (GLW) project. We present a complimentary effort to map the distribution of cattle and pigs in Malawi, Uganda, Democratic Republic of Congo, and South Sudan. In contrast to GLW, which uses dasymmetric modeling applied to census data to produce time-stratified estimates of livestock counts and spatial density, our work uses complex survey data and distinct modeling methods to generate a time-series of livestock distribution, defining livestock density as the ratio of animals to humans. In addition to favorable cross-validation results and general agreement with national density estimates derived from external data on national human and livestock populations, our results demonstrate extremely good agreement with GLW-3 estimates, supporting the validity of both efforts. Our results furthermore offer a high-resolution time series result and employ a definition of density which is particularly well-suited to the study of livestock-origin zoonoses.

Accounting for Sampling Weights in the Analysis of Spatial Distributions of Disease Using Health Survey Data, with an Application to Mapping Child Health in Malawi and Mozambique

Most analyses of spatial patterns of disease risk using health survey data fail to adequately account for the complex survey designs. Particularly, the survey sampling weights are often ignored in the analyses. Thus, the estimated spatial distribution of disease risk could be biased and may lead to erroneous policy decisions. This paper aimed to present recent statistical advances in disease-mapping methods that incorporate survey sampling in the estimation of the spatial distribution of disease risk. The methods were then applied to the estimation of the geographical distribution of child malnutrition in Malawi, and child fever and diarrhoea in Mozambique. The estimation of the spatial distributions of the child disease risk was done by Bayesian methods. Accounting for sampling weights resulted in smaller standard errors for the estimated spatial disease risk, which increased the confidence in the conclusions from the findings. The estimated geographical distributions of the child disease risk were similar between the methods. However, the fits of the models to the data, as measured by the deviance information criteria (DIC), were different.

Geographic Variation in HIV Testing Among Transgender and Nonbinary Adults in the United States

BACKGROUND

Transgender and nonbinary (TNB) populations are disproportionately affected by HIV and few local health departments or HIV surveillance systems collect/report data on TNB identities. Our objective was to estimate the prevalence of HIV testing among TNB adults by US county and state, with a focus on the Ending the HIV Epidemic (EHE) geographies.

METHODS

We applied a Bayesian hierarchical spatial small area estimation model to data from the 2015 US Transgender Survey, a large national cross-sectional Internet-based survey. We estimated the county- and state-level proportion of TNB adults who ever tested or tested for HIV in the last year by gender identity, race/ethnicity, and age.

RESULTS

Our analysis included 26,100 TNB participants with valid zip codes who resided in 1688 counties (54% of all 3141 counties that cover 92% of the US population). The median county-level proportion of TNB adults who ever tested for HIV was 44% (range 10%-80%) and who tested in the last year was 17% (range 4%-44%). Within most counties, testing was highest among transgender women, black respondents, and people aged ≥25 years. HIV testing was lowest among nonbinary people and young adults aged <25 years. The proportion of TNB adults who tested within the last year was very low in most EHE counties and in all 7 rural states.

CONCLUSIONS

HIV testing among TNB adults is likely below national recommendations in the majority of EHE geographies. Geographic variation in HIV testing patterns among TNB adults indicates that testing strategies need to be tailored to local settings.

Finding gaps in routine TB surveillance activities in Bangladesh

BACKGROUND :

TB was the leading cause of death from a single infectious pathogen globally between 2014 and 2019. Fine-scale estimates of TB prevalence and case notifications can be combined to guide priority-setting for strengthening routine surveillance activities in high-burden countries. We produce policy-relevant estimates of the TB epidemic at the second administrative unit in Bangladesh.

METHODS :

We used a Bayesian spatial framework and the cross-sectional National TB Prevalence Survey from 2015–2016 in Bangladesh to estimate prevalence by district. We used case notifications to calculate prevalence-to-notification ratio, a key metric of under-diagnosis and under-reporting.

RESULTS :

TB prevalence rates were highest in the north-eastern districts and ranged from 160 cases per 100,000 (95% uncertainty interval [UI] 80–310) in Jashore to 840 (UI 690–1020) in Sunamganj. Despite moderate prevalence rates, the Rajshahi and Dhaka Divisions presented the highest prevalence-to-notification ratios due to low case notifications. Resolving subnational disparities in case detection could lead to 26,500 additional TB cases (UI 8,500–79,400) notified every year.

CONCLUSION :

This study is the first to produce and map subnational estimates of TB prevalence and prevalence-to-notification ratios, which are essential to target prevention and treatment efforts in high-burden settings. Reaching TB cases currently missing from care will be key to ending the TB epidemic.

Data Smoothing, Extrapolation, and Triangulation

While population size data at the national and subnational levels are critical for planning and resource allocation, most PSE studies are conducted in a few selected cities and towns due to resource and time constraints. Extrapolation methods to estimate the population size for locations not included in the PSE exercise are critical to providing data for all subnational jurisdictions and the entire country as a whole. The decision on how to select cities or towns that reflect best the national picture is a critical first step. You may need to include cities or towns from areas with low, medium, or high prevalence of your target population(s). Having some data from all these (three) areas increase the power of your study and helps to better extrapolate the results to unobserved areas, and to the national level. Through this process, data smoothing of subnational data is often needed due to relatively small samples size problem. In this chapter, we describe methods that are required for smoothing subnational data, model subnational data to extrapolate estimates to regional and national levels, and triangulate findings from different size estimation methods by a Bayesian framework to produce credible estimates.

Model building and assessment of the impact of covariates for disease prevalence mapping in low-resource settings: to explain and to predict

This paper provides statistical guidance on the development and application of model-based geostatistical methods for disease prevalence mapping. We illustrate the different stages of the analysis, from exploratory analysis to spatial prediction of prevalence, through a case study on malaria mapping in Tanzania. Throughout the paper, we distinguish between predictive modelling, whose main focus is on maximizing the predictive accuracy of the model, and explanatory modelling, where greater emphasis is placed on understanding the relationships between the health outcome and risk factors. We demonstrate that these two paradigms can result in different modelling choices. We also propose a simple approach for detecting over-fitting based on inspection of the correlation matrix of the estimators of the regression coefficients. To enhance the interpretability of geostatistical models, we introduce the concept of domain effects in order to assist variable selection and model validation. The statistical ideas and principles illustrated here in the specific context of disease prevalence mapping are more widely applicable to any regression model for the analysis of epidemiological outcomes but are particularly relevant to geostatistical models, for which the separation between fixed and random effects can be ambiguous.

Mapping changes in district level prevalence of childhood stunting in India 1998-2016: An application of small area estimation techniques

The four rounds of National Family Health Survey (NFHS) conducted during 1992-93, 1998-99, 2005-06 and 2015-16 is main source to track the health and development related indicators including nutritional status of children at national and state level in India. Except NFHS-4, first three rounds of NFHS were unable to provides district-level estimates of childhood stunting due to the insufficient sample sizes. The small area estimation (SAE) techniques offer a viable solution to overcome the problem of small sample size. Therefore, this study uses SAE techniques to derive district level prevalence of childhood stunting corresponding to NFHS-2 (1998-99). Study further estimated GIS maps, univariate Local indicator of spatial autocorrelation (LISA) and Moran's I to understand the trend in district level childhood stunting between NFHS-2 and NFHS-4. Estimates obtained by SAE techniques suggest that prevalence of childhood stunting ranges from 20.7% (95% CI: 18.8-22.7) in South Goa district of Goa to 64.4% (95%CI: 63.1-65.7) in Dhaulpur district of Rajasthan during 1998-99. The diagnostic measures used to validate the reliability of estimates obtained by SAE techniques indicate that the model-based estimates are reliable and representative at district level. Results of geospatial analysis indicates substantial reduction in childhood stunting between 1998 and 2016. Out of 640 district,about 81 district experience reduction of more than 50%. At the same time 60 district experience less than 10% of reduction between 1998 and 2016. Spatial clustering of childhood stunting remains same over the study period except few additional cluster in Maharashtra, Andhra and Meghalaya in 2016. The district level estimates obtained from this study might be helpful in framing decentralized policies and implementation of vertical programs to enhance the efficacy of various nutrition interventions in priority districts of the country.

District-level estimation of vaccination coverage: Discrete vs continuous spatial models

Health and development indicators (HDIs) such as vaccination coverage are regularly measured in many low- and middle-income countries using household surveys, often due to the unreliability or incompleteness of routine data collection systems. Recently, the development of model-based approaches for producing subnational estimates of HDIs using survey data, particularly cluster-level data, has been an active area of research. This is mostly driven by the increasing demand for estimates at certain administrative levels, for example, districts, at which many development goals are set and evaluated. In this study, we explore spatial modeling approaches for producing district-level estimates of vaccination coverage. Specifically, we compare discrete spatial smoothing models which directly model district-level data with continuous Gaussian process (GP) models that utilize geolocated cluster-level data. We adopt a fully Bayesian framework, implemented using the INLA and SPDE approaches. We compare the predictive performance of the models by analyzing vaccination coverage using data from two Demographic and Health Surveys (DHS), namely the 2014 Kenya DHS and the 2015-16 Malawi DHS. We find that the continuous GP models performed well, offering a credible alternative to traditional discrete spatial smoothing models. Our analysis also revealed that accounting for between-cluster variation in the continuous GP models did not have any real effect on the district-level estimates. Our results provide guidance to practitioners on the reliability of these model-based approaches for producing estimates of vaccination coverage and other HDIs.

Individual and Community Social Determinants of Health Associated With Diabetes Management in a Mexican American Population

Background: Diabetes is a major health burden in Mexican American populations, especially among those in the Lower Rio Grande Valley (LRGV) in the border region of Texas. Understanding the roles that social determinants of health (SDOH) play in diabetes management programs, both at the individual and community level, may inform future intervention strategies.

Methods: This study performed a secondary data analysis on 1,568 individuals who participated in Salud y Vida (SyV), a local diabetes and chronic disease management program, between October 2013 and September 2018 recruited from a local clinic. The primary outcome was the reduction of hemoglobin A1C (HbA1C) at the last follow-up visit compared to the baseline. In addition to age, gender, insurance status, education level and marital status, we also investigated 15 community (census tract) SDOH using the American Community Survey. Because of the high correlation in the community SDOH, we developed the community-level indices representing different domains. Using Bayesian multilevel spatial models that account for the geographic dependency, we were able to simultaneously investigate the individual- and community-level SDOH that may impact HbA1C reduction.

Results: After accounting for the diabetes self-management education classes taken by the participants and their length of stay in the program, we found that older age at baseline, being married (compared to being widowed or divorced) and English speaking (compared to Spanish) were significantly associated with greater HbA1C reduction. Moreover, we found that the community level SDOH were also highly associated with HbA1C reduction. With every percentile rank decrease in the socioeconomic advantage index, we estimated an additional 0.018% reduction in HbA1C [95% CI (−0.028, −0.007)]. Besides the socioeconomic advantage index, urban core opportunity and immigrant's cohesion and accessibility indices were also statistically associated with HbA1C reduction.

Conclusion: To our knowledge, our study is the first to utilize Bayesian multilevel spatial models and simultaneously investigate both individual- and community-level SDOH in the context of diabetes management. Our findings suggest that community SDOH play an important role in diabetes control and management, and the need to consider community and neighborhood context in future interventions programs to maximize their overall effectiveness.

A flexible hierarchical framework for improving inference in area-referenced environmental health studies

Study designs where data have been aggregated by geographical areas are popular in environmental epidemiology. These studies are commonly based on administrative databases and, providing a complete spatial coverage, are particularly appealing to make inference on the entire population. However, the resulting estimates are often biased and difficult to interpret due to unmeasured confounders, which typically are not available from routinely collected data. We propose a framework to improve inference drawn from such studies exploiting information derived from individual-level survey data. The latter are summarized in an area-level scalar score by mimicking at ecological level the well-known propensity score methodology. The literature on propensity score for confounding adjustment is mainly based on individual-level studies and assumes a binary exposure variable. Here, we generalize its use to cope with area-referenced studies characterized by a continuous exposure. Our approach is based upon Bayesian hierarchical structures specified into a two-stage design: (i) geolocated individual-level data from survey samples are up-scaled at ecological level, then the latter are used to estimate a generalized ecological propensity score (EPS) in the in-sample areas; (ii) the generalized EPS is imputed in the out-of-sample areas under different assumptions about the missingness mechanisms, then it is included into the ecological regression, linking the exposure of interest to the health outcome. This delivers area-level risk estimates, which allow a fuller adjustment for confounding than traditional areal studies. The methodology is illustrated by using simulations and a case study investigating the risk of lung cancer mortality associated with nitrogen dioxide in England (UK).

A Scoping Review of Spatial Analysis Approaches Using Health Survey Data in Sub-Saharan Africa

Spatial analysis has become an increasingly used analytic approach to describe and analyze spatial characteristics of disease burden, but the depth and coverage of its usage for health surveys data in Sub-Saharan Africa are not well known. The objective of this scoping review was to conduct an evaluation of studies using spatial statistics approaches for national health survey data in the SSA region. An organized literature search for studies related to spatial statistics and national health surveys was conducted through PMC, PubMed/Medline, Scopus, NLM Catalog, and Science Direct electronic databases. Of the 4,193 unique articles identified, 153 were included in the final review. Spatial smoothing and prediction methods were predominant (n = 108), followed by spatial description aggregation (n = 25), and spatial autocorrelation and clustering (n = 19). Bayesian statistics methods and lattice data modelling were predominant (n = 108). Most studies focused on malaria and fever (n = 47) followed by health services coverage (n = 38). Only fifteen studies employed nonstandard spatial analyses (e.g., spatial model assessment, joint spatial modelling, accounting for survey design). We recommend that for future spatial analysis using health survey data in the SSA region, there must be an improve recognition and awareness of the potential dangers of a naïve application of spatial statistical methods. We also recommend a wide range of applications using big health data and the future of data science for health systems to monitor and evaluate impacts that are not well understood at local levels.

Multi-Scale Multivariate Models for Small Area Health Survey Data: A Chilean Example

Background: We propose a general approach to the analysis of multivariate health outcome data where geo-coding at different spatial scales is available. We propose multiscale joint models which address the links between individual outcomes and also allow for correlation between areas. The models are highly novel in that they exploit survey data to provide multiscale estimates of the prevalences in small areas for a range of disease outcomes. Results The models incorporate both disease specific, and common disease spatially structured components. The multiple scales envisaged is where individual survey data is used to model regional prevalences or risks at an aggregate scale. This approach involves the use of survey weights as predictors within our Bayesian multivariate models. Missingness has to be addressed within these models and we use predictive inference which exploits the correlation between diseases to provide estimates of missing prevalances. The Case study we examine is from the National Health Survey of Chile where geocoding to Province level is available. In that survey, diabetes, Hypertension, obesity and elevated low-density cholesterol (LDL) are available but differential missingness requires that aggregation of estimates and also the assumption of smoothed sampling weights at the aggregate level. Conclusions: The methodology proposed is highly novel and flexibly handles multiple disease outcomes at individual and aggregated levels (i.e., multiscale joint models). The missingness mechanism adopted provides realistic estimates for inclusion in the aggregate model at Provincia level. The spatial structure of four diseases within Provincias has marked spatial differentiation, with diabetes and hypertension strongly clustered in central Provincias and obesity and LDL more clustered in the southern areas.

Spatial Modelling to Inform Public Health Based on Health Surveys: Impact of Unsampled Areas at Lower Geographical Scale

Small area estimation is an important tool to provide area-specific estimates of population characteristics for governmental organizations in the context of education, public health and care. However, many demographic and health surveys are unrepresentative at a small geographical level, as often areas at a lower level are not included in the sample due to financial or logistical reasons. In this paper, we investigated (1) the effect of these unsampled areas on a variety of design-based and hierarchical model-based estimates and (2) the benefits of using auxiliary information in the estimation process by means of an extensive simulation study. The results showed the benefits of hierarchical spatial smoothing models towards obtaining more reliable estimates for areas at the lowest geographical level in case a spatial trend is present in the data. Furthermore, the importance of auxiliary information was highlighted, especially for geographical areas that were not included in the sample. Methods are illustrated on the 2008 Mozambique Poverty and Social Impact Analysis survey, with interest in the district-specific prevalence of school attendance.

Sub-national levels and trends in contraceptive prevalence, unmet need, and demand for family planning in Nigeria with survey uncertainty

Background

Ambitious global goals have been established to provide universal access to affordable modern contraceptive methods. To measure progress toward such goals in populous countries like Nigeria, it’s essential to characterize the current levels and trends of family planning (FP) indicators such as unmet need and modern contraceptive prevalence rates (mCPR). Moreover, the substantial heterogeneity across Nigeria and scale of programmatic implementation requires a sub-national resolution of these FP indicators. The aim of this study is to estimate the levels and trends of FP indicators at a subnational scale in Nigeria utilizing all available data and accounting for survey design and uncertainty.

Methods

We utilized all available cross-sectional survey data from Nigeria including the Demographic and Health Surveys, Multiple Indicator Cluster Surveys, National Nutrition and Health Surveys, and Performance, Monitoring, and Accountability 2020. We developed a hierarchical Bayesian model that incorporates all of the individual level data from each survey instrument, accounts for survey uncertainty, leverages spatio-temporal smoothing, and produces probabilistic estimates with uncertainty intervals.

Results

We estimate that overall rates and trends of mCPR and unmet need have remained low in Nigeria: the average annual rate of change for mCPR by state is 0.5% (0.4%,0.6%) from 2012-2017. Unmet need by age-parity demographic groups varied significantly across Nigeria; parous women express much higher rates of unmet need than nulliparous women.

Conclusions

Understanding the estimates and trends of FP indicators at a subnational resolution in Nigeria is integral to inform programmatic decision-making. We identify age-parity-state subgroups with large rates of unmet need. We also find conflicting trends by survey instrument across a number of states. Our model-based estimates highlight these inconsistencies, attempt to reconcile the direct survey estimates, and provide uncertainty intervals to enable interpretation of model and survey estimates for decision-making.

Spatial smoothing models to deal with the complex sampling design and nonresponse in the Florida BRFSS survey

Public health and governmental organizations have acknowledged the importance of obtaining information of various characteristics for small areas, such as counties. Spatial smoothing models have been developed to gain reliable information on the geographical distribution of the outcome of interest. When the geographical analysis is based on survey data, two issues pose challenges: (1) the complex design of the survey and (2) the presence of missing data due to non-response. We investigate the influence of missing data and the adjustment thereof in the context of the 2013 Florida Behavioral Risk Factor Surveillance System (BRFSS) health survey. We focus on the application and comparison of the Hajek ratio estimator and two model-based approaches for estimation of the spatial trend of the prevalence of having no health insurance coverage. The model-based methods are compared using the Deviance Information Criterion which show the benefits of modeling the weights as flexibly as possible. Methods are extended towards subgroup analyses and the estimation of area-specific standardized rates, where household incomes was identified as an important factor to include in the analysis.

Building a Data Platform for Cross-Country Urban Health Studies: the SALURBAL Study

Studies examining urban health and the environment must ensure comparability of measures across cities and countries. We describe a data platform and process that integrates health outcomes together with physical and social environment data to examine multilevel aspects of health across cities in 11 Latin American countries. We used two complementary sources to identify cities with ≥ 100,000 inhabitants as of 2010 in Argentina, Brazil, Chile, Colombia, Costa Rica, El Salvador, Guatemala, Mexico, Nicaragua, Panama, and Peru. We defined cities in three ways: administratively, quantitatively from satellite imagery, and based on country-defined metropolitan areas. In addition to "cities," we identified sub-city units and smaller neighborhoods within them using census hierarchies. Selected physical environment (e.g., urban form, air pollution and transport) and social environment (e.g., income, education, safety) data were compiled for cities, sub-city units, and neighborhoods whenever possible using a range of sources. Harmonized mortality and health survey data were linked to city and sub-city units. Finer georeferencing is underway. We identified 371 cities and 1436 sub-city units in the 11 countries. The median city population was 234,553 inhabitants (IQR 141,942; 500,398). The systematic organization of cities, the initial task of this platform, was accomplished and further ongoing developments include the harmonization of mortality and survey measures using available sources for between country comparisons. A range of physical and social environment indicators can be created using available data. The flexible multilevel data structure accommodates heterogeneity in the data available and allows for varied multilevel research questions related to the associations of physical and social environment variables with variability in health outcomes within and across cities. The creation of such data platforms holds great promise to support researching with greater granularity the field of urban health in Latin America as well as serving as a resource for the evaluation of policies oriented to improve the health and environmental sustainability of cities.

Spatial small area smoothing models for handling survey data with nonresponse

Spatial smoothing models play an important role in the field of small area estimation. In the context of complex survey designs, the use of design weights is indispensable in the estimation process. Recently, efforts have been made in these spatial smoothing models, in order to obtain reliable estimates of the spatial trend. However, the concept of missing data remains a prevalent problem in the context of spatial trend estimation as estimates are potentially subject to bias. In this paper, we focus on spatial health surveys where the available information consists of a binary response and its associated design weight. Furthermore, we investigate the impact of nonresponse as missing data on a range of spatial models for different missingness mechanisms and different degrees of missingness by means of an extensive simulation study. The computations were performed in R, using INLA and other existing packages. The results show that weight adjustment to correct for missingness has a beneficial effect on the bias in the missing at random setting for all models. Furthermore, we estimate the geographical distribution of perceived health at the district level based on the Belgian Health Interview Survey (2001). Copyright © 2017 John Wiley & Sons, Ltd.

Model-based inference for small area estimation with sampling weights

Obtaining reliable estimates about health outcomes for areas or domains where only few to no samples are available is the goal of small area estimation (SAE). Often, we rely on health surveys to obtain information about health outcomes. Such surveys are often characterised by a complex design, stratification, and unequal sampling weights as common features. Hierarchical Bayesian models are well recognised in SAE as a spatial smoothing method, but often ignore the sampling weights that reflect the complex sampling design. In this paper, we focus on data obtained from a health survey where the sampling weights of the sampled individuals are the only information available about the design. We develop a predictive model-based approach to estimate the prevalence of a binary outcome for both the sampled and non-sampled individuals, using hierarchical Bayesian models that take into account the sampling weights. A simulation study is carried out to compare the performance of our proposed method with other established methods. The results indicate that our proposed method achieves great reductions in mean squared error when compared with standard approaches. It performs equally well or better when compared with more elaborate methods when there is a relationship between the responses and the sampling weights. The proposed method is applied to estimate asthma prevalence across districts.

Using Small-Area Estimation to Calculate the Prevalence of Smoking by Subcounty Geographic Areas in King County, Washington, Behavioral Risk Factor Surveillance System, 2009-2013

Introduction

King County, Washington, fares well overall in many health indicators. However, county-level data mask disparities among subcounty areas. For disparity-focused assessment, a demand exists for examining health data at subcounty levels such as census tracts and King County health reporting areas (HRAs).

Methods

We added a “nearest intersection” question to the Behavioral Risk Factor Surveillance System (BRFSS) and geocoded the data for subcounty geographic areas, including census tracts. To overcome small sample size at the census tract level, we used hierarchical Bayesian models to obtain smoothed estimates in cigarette smoking rates at the census tract and HRA levels. We also used multiple imputation to adjust for missing values in census tracts.

Results

Direct estimation of adult smoking rates at the census tract level ranged from 0% to 56% with a median of 10%. The 90% confidence interval (CI) half-width for census tract with nonzero rates ranged from 1 percentage point to 37 percentage points with a median of 13 percentage points. The smoothed-multiple–imputation rates ranged from 5% to 28% with a median of 12%. The 90% CI half-width ranged from 4 percentage points to 13 percentage points with a median of 8 percentage points.

Conclusion

The nearest intersection question in the BRFSS provided geocoded data at subcounty levels. The Bayesian model provided estimation with improved precision at the census tract and HRA levels. Multiple imputation can be used to account for missing geographic data. Small-area estimation, which has been used for King County public health programs, has increasingly become a useful tool to meet the demand of presenting data at more granular levels.

Space-Time Smoothing of Complex Survey Data: Small Area Estimation for Child Mortality

Many people living in low and middle-income countries are not covered by civil registration and vital statistics systems. Consequently, a wide variety of other types of data including many household sample surveys are used to estimate health and population indicators. In this paper we combine data from sample surveys and demographic surveillance systems to produce small area estimates of child mortality through time. Small area estimates are necessary to understand geographical heterogeneity in health indicators when full-coverage vital statistics are not available. For this endeavor spatio-temporal smoothing is beneficial to alleviate problems of data sparsity. The use of conventional hierarchical models requires careful thought since the survey weights may need to be considered to alleviate bias due to non-random sampling and non-response. The application that motivated this work is estimation of child mortality rates in five-year time intervals in regions of Tanzania. Data come from Demographic and Health Surveys conducted over the period 1991-2010 and two demographic surveillance system sites. We derive a variance estimator of under five years child mortality that accounts for the complex survey weighting. For our application, the hierarchical models we consider include random effects for area, time and survey and we compare models using a variety of measures including the conditional predictive ordinate (CPO). The method we propose is implemented via the fast and accurate integrated nested Laplace approximation (INLA).

Spatial and socio-economic effects on malaria morbidity in children under 5 years in Malawi in 2012

BACKGROUND

Malaria is a major health challenge in sub-Saharan Africa with children under 5 being most vulnerable. Identifying regions of greater malarial burden is vital in targeting interventions.

METHODS

This study analysed malaria morbidity using data from the Malawi 2012 Malaria Indicator Survey that were obtained from Demographic and Health Survey (DHS) program website. These data captured malaria related information on children under 5. Poisson regression was done to determine associations between outcome (number of children under 5 with malaria in household) and explanatory variables. A Bayesian smoothing approach was employed to adjust for spatial random effects on child related variables.

RESULTS

There were 1878 households in 140 clusters. The number of children under five was 1900. Spatially structured effects accounted for more than 90% of random effects as these had a mean of 1.32 (95% Credible Interval (CI)=0.37, 2.50) whilst spatially unstructured had a mean of 0.10 (CI=9.0 × 10(-4), 0.38). Spatially adjusted significant variables were; type of place of residence (urban or rural) [posterior odds ratio (POR)=2.06; CI= 1.27, 3.34], not owning land [POR=1.77; CI=1.19, 2.64], not staying in a slum [POR=0.52; CI=0.33, 0.83] and enhanced vegetation index [POR=0.02; CI=0.00, 1.08]. A trend was observed on usage of insecticide treated mosquito nets [POR=0.80; CI=0.63, 1.03].

CONCLUSION

This study showed that malaria is a disease of poverty. Enhanced vegetation index was an important factor in malaria morbidity. The central region was identified as the area with greatest disease burden.

Estimating areas of common risk in low birth weight and infant mortality in Namibia: a joint spatial analysis at sub-regional level

There is lots of literature documenting a positive association between low birth weight (LBW) and infant mortality (IM), however, little is known how the risk of LBW and IM are geographically co-distributed. We fitted joint spatial models of LBW and IM, and used data from Namibia, to examine their geographical variability. We used a Bayesian approach to measure and rank areas according to specific and shared risk of LBW and IM. Our findings show some degree of similarities in the spatial pattern of LBW and IM, with high risk in the central and north-eastern parts of the country. Results suggest a need for comprehensive programming of maternal and newborn interventions that reach areas of spatially concentrated risk of LBW and IM. It further presents an opportunity for generating hypotheses for further research aimed at improving child health, especially in higher risk constituencies thus identified.