A novel framework for validating and applying standardized small area measurement strategies

Prevalence and regional distribution of obstructive sleep apnea in Canada: Analysis from the Canadian Longitudinal Study on Aging

OBJECTIVES

Obstructive sleep apnea (OSA) is a common chronic condition that is often undiagnosed or diagnosed after many years of symptoms and has an impact on quality of life and several health factors. We estimated the Canadian national prevalence of OSA using a validated questionnaire and physical measurements in participants in the Canadian Longitudinal Study on Aging (CLSA).

METHODS

The method used individual risk estimation based upon the validated STOP-BANG scale developed for OSA. This stratified population sample spans Canada to provide regional estimates.

RESULTS

In this sample of adults aged 45 to 85 years old, the overall prevalence in 2015 of combined moderate and severe OSA in the 51,337 participants was 28.1% (95% confidence intervals, 27.8‒28.4). The regional prevalence varied statistically between Atlantic Canada and Western Canada (p < 0.001), although clinically the variations were limited. The provincial prevalence for moderate and severe OSA ranged from 27.5% (New Brunswick and British Columbia) to 29.1% (Manitoba). Body mass index (BMI) was the dominant determinant of the variance between provinces (β = 0.33, p < 0.001). Only 1.2% of participants had a clinical diagnosis of OSA.

CONCLUSION

The great majority (92.9%) of the participants at high risk of OSA were unrecognized and had no clinical diagnosis of OSA.

Adverse Childhood Experiences Among U.S. Adults: National and State Estimates by Adversity Type, 2019-2020

INTRODUCTION

Although adverse childhood experiences (ACEs) are associated with lifelong health harms, current surveillance data on exposures to childhood adversity among adults are either unavailable or incomplete for many states. In this study, recent data from a nationally representative survey were used to obtain the current and complete estimates of ACEs at the national and state levels.

METHODS

Current, complete, by-state estimates of adverse childhood experiences were obtained by applying small area estimation technique to individual-level data on adults aged ≥18 years from 2019-2020 Behavioral Risk Factor Surveillance System survey. The standardized questions about childhood adversity included in the 2019-2020 survey allowed for obtaining estimates of ACE consistent across states. All missing responses to childhood adversity questions (states did not offer such questions or offered them to only some respondents; respondents skipped questions) were predicted through multilevel mixed-effects logistic small area estimation regressions. The analyses were conducted between October 2022 and May 2023.

RESULTS

An estimated 62.8% of U.S. adults had past exposure to ACEs (range: 54.9% in Connecticut; 72.5% in Maine). Emotional abuse (34.5%) was the most common; household member incarceration (10.6%) was the least common. Sexual abuse varied markedly between females (22.2%) and males (5.4%). Exposure to most types of adverse childhood experiences was lowest for adults who were non-Hispanic White, had the highest level of education (college degree) or income (annual income ≥$50,000), or had access to a personal healthcare provider.

CONCLUSIONS

Current complete estimates of ACEs demonstrate high countrywide exposures and stark sociodemographic inequalities in the burden, highlighting opportunities to prevent adverse childhood experiences by focusing social, educational, medical, and public health interventions on populations disproportionately impacted.

Impact of the use of small-area models on estimation of attributable mortality at a regional level

The objective of this study is to assess the impact of applying prevalences derived from a small-area model at a regional level on smoking-attributable mortality (SAM). A prevalence-dependent method was used to estimate SAM. Prevalences of tobacco use were derived from a small-area model. SAM and population attributable fraction (PAF) estimates were compared against those calculated by pooling data from three national health surveys conducted in Spain (2011-2014-2017). We calculated the relative changes between the two estimates and assessed the width of the 95% CI of the PAF. Applying surveys-based prevalences, tobacco use was estimated to cause 53 825 (95% CI: 53 182-54 342) deaths in Spain in 2017, a figure 3.8% lower obtained with the small-area model prevalences. The lowest relative change was observed in the Castile-La Mancha region (1.1%) and the highest in Navarre (14.1%). The median relative change between regions was higher for women (26.1%), population aged ≥65 years (6.6%), and cardiometabolic diseases (9.0%). The differences between PAF by cause of death were never greater than 2%. Overall, the differences between estimates of SAM, PAF, and confidence interval width are small when using prevalences from both sources. Having these data available by region will allow decision-makers to implement smoking control measures based on more accurate data.

Bioinformatics reveals the pathophysiological relationship between diabetic nephropathy and periodontitis in the context of aging

Diabetic nephropathy (DN) is one of the most common microvascular complications of diabetes mellitus. Periodontitis (PD) is a microbially-induced chronic inflammatory disease that is thought to have a bidirectional relationship with diabetes mellitus. DN and PD are recognized as models associated with accelerated aging. This study is divided into two parts, the first of which explores the bidirectional causal relationship through Mendelian randomization (MR). The second part aims to investigate the relationship between PD and DN in terms of potential crosstalk genes, aging-related genes, biological pathways, and processes using bioinformatic methods. MR analysis showed no evidence to support a causal relationship between DN and PD (P = 0.34) or PD and DN (P = 0.77). Using the GEO database, we screened 83 crosstalk genes overlapping in two diseases. Twelve paired genes identified by Pearson correlation and the four hub genes in the key cluster were jointly evaluated as key crosstalk-aging genes. Using support vector machine recursive feature elimination (SVM-RFE) and maximal clique centrality (MCC) algorithms, feature selection established five genes as the key crosstalk-aging genes. Based on five key genes, an ANN diagnostic model with reliable diagnosis of two diseases was developed. Gene enrichment analysis indicates that AGE-RAGE pathway signaling, the complement system, and multiple immune inflammatory pathways may be involved in common features of both diseases. Immune infiltration analysis reveals that most immune cells are differentially expressed in PD and DN, with dendritic cells and T cells assuming vital roles in both diseases. Overall, although there is no causal link, CSF1R, CXCL6, VCAM1, JUN and IL1B may be potential crosstalk-aging genes linking PD and DN. The common pathways and markers explored in this study could contribute to a deeper understanding of the common pathogenesis of both diseases in the context of aging and provide a theoretical basis for future research.

Evaluation of the long-term variability of macular OCT/OCTA and visual field parameters

BACKGROUND/AIMS

To assess the long-term variability of macular optical coherence tomography (OCT)/OCT angiography (OCTA) and visual field (VF) parameters.

METHODS

Healthy and glaucoma eyes with ≥1-year follow-up were included. 24-2 VF and macular OCT/OCTA parameters, including VF mean deviation (MD), whole-image vessel density (wiVD) and ganglion cell complex thickness (wiGCC) were analysed. Intraclass correlation coefficient (ICC), root mean squared error (RMSE), within-subject test-retest SD (Sw) and test-retest variability were calculated for stable eye cohort (max follow-up=1.5 years). Rates of change and RMSE were evaluated in the extended cohort including all eyes (unlimited follow-up).

RESULTS

From a total of 230 eyes (150 participants; age=67.7 years), 86 eyes (37%, 62 participants) were stable. In stable eyes, OCT parameters showed the highest mean (95%) ICC (wiGCC=0.99 (0.99, 0.99)), followed by VF (VF MD=0.91 (0.88, 0.93)) and OCTA (wiVD=0.82 (0.75, 0.87)). RMSE and Sw for VF MD were 0.92 dB and 0.81 dB, respectively, for wiVD were 1.64% and 1.48%, respectively, and for wiGCC, 0.91 µm and 0.78 µm, respectively. The long-term test-rest variability of VF MD, wiVD and wiGCC was 2.2 dB, 4.1% and 2.2 µm, respectively. In the extended cohort (mean follow-up=3.0 years), all parameters had significant rates of change (p<0.001), and compared with the stable cohort, only slightly higher RMSE (VF MD=1.07 dB; wiGCC=2.03 µm; wiVD=2.57%) were found.

CONCLUSIONS

VF and macular OCT/OCTA, particularly OCT parameters, showed small long-term variability in all eyes, including stable ones, supporting the use of these instruments in glaucoma follow-up. Changes in macular VD and GCC greater than 4%-5% and 2 µm, respectively, indicate possible progression.

TRIAL REGISTRATION NUMBER

NCT00221897.

Validation of a small-area model for estimation of smoking prevalence at a subnational level

INTRODUCTION

Small-area estimation methods are an alternative to direct survey-based estimates in cases where a survey's sample size does not suffice to ensure representativeness. Nevertheless, the information yielded by small-area estimation methods must be validated. The objective of this study was thus to validate a small-area model.

METHODS

The prevalence of smokers, ex-smokers, and never smokers by sex and age group (15-34, 35-54, 55-64, 65-74, ≥75 years) was calculated in two Spanish Autonomous Regions (ARs) by applying a weighted ratio estimator (direct estimator) to data from representative surveys. These estimates were compared against those obtained with a small-area model applied to another survey, specifically the Spanish National Health Survey, which did not guarantee representativeness for these two ARs by sex and age. To evaluate the concordance of the estimates, we calculated the intraclass correlation coefficient (ICC) and the 95% confidence intervals of the differences between estimates. To assess the precision of the estimates, the coefficients of variation were obtained.

RESULTS

In all cases, the ICC was ≥0.87, indicating good concordance between the direct and small-area model estimates. Slightly more than eight in ten 95% confidence intervals for the differences between estimates included zero. In all cases, the coefficient of variation of the small-area model was <30%, indicating a good degree of precision in the estimates.

CONCLUSIONS

The small-area model applied to national survey data yields valid estimates of smoking prevalence by sex and age group at the AR level. These models could thus be applied to a single year's data from a national survey, which does not guarantee regional representativeness, to characterize various risk factors in a population at a subnational level.

Small-area models to assess the geographical distribution of tobacco consumption by sex and age in Spain

INTRODUCTION

Complete and accurate data on smoking prevalence at a local level would enable health authorities to plan context-dependent smoking interventions. However, national health surveys do not generally provide direct estimates of smoking prevalence by sex and age groups at the subnational level. This study uses a small-area model-based methodology to obtain precise estimations of smoking prevalence by sex, age group and region, from a population-based survey.

METHODS

The areas targeted for analysis consisted of 180 groups based on a combination of sex, age group (15-34, 35-54, 55-64, 65-74, and ≥75 years), and Autonomous Region. Data on tobacco use came from the 2017 Spanish National Health Survey (2017 SNHS). In each of the 180 groups, we estimated the prevalence of smokers (S), ex-smokers (ExS) and never smokers (NS), as well as their coefficients of variation (CV), using a weighted ratio estimator (direct estimator) and a multinomial logistic model with random area effects.

RESULTS

When smoking prevalence was estimated using the small-area model, the precision of direct estimates improved; the CV of S and ExS decreased on average by 26%, and those of NS by 25%. The range of S prevalence was 11-46% in men and 4-37% in women, excluding the group aged ≥75 years.

CONCLUSIONS

This study proposes a methodology for obtaining reliable estimates of smoking prevalence in groups or areas not covered in the survey design. The model applied is a good alternative for enhancing the precision of estimates at a detailed level, at a much lower cost than that involved in conducting large-scale surveys. This method could be easily integrated into routine data processing of population health surveys. Having such estimates directly after completing a health survey would help characterize the tobacco epidemic and/or any other risk factor more precisely.

Generating data to facilitate more equitable distribution of health resources: an illustration of how local health surveys can identify probable need in mixed socio-economic regions

OBJECTIVES

This study aimed to (1) encourage allocation of governmental and grant funds to the administration of local area health surveys and (2) illustrate the predictive impact of socio-economic resources on adult health status at the local area level to provide an example of how health surveys can identify residents with the greatest health needs.

STUDY DESIGN

Randomly sampled and weight-adjusted regional household health survey (7501 respondents) analyzed with categorical bivariate and multivariate statistics, combined with Census data. Survey sample consists of the lowest, highest, and near highest ranked counties in the County Health Rankings and Roadmaps for Pennsylvania.

METHODS

Socio-economic status (SES) is measured regionally with Census data consisting of seven indicators and individually with Health Survey data consisting of five indicators based on poverty level, overall household income, and education. Both of these composite measures are examined jointly for their predictive effects on a validated health status measure using binary logistic regression.

RESULTS

Once county-level measures of SES and health status are broken down into smaller areas, better identification of pockets of health need is possible. This was most strongly revealed in an urban county, Philadelphia, which is ranked lowest of 67 counties on health measures in the state of Pennsylvania, yet when broken down into 'neighborhood clusters' contained both the highest- and lowest-ranked local area in a five-county region. Overall, regardless of the SES level of the County subdivision one lives in, a low-SES adult has close to six times greater odds of reporting 'fair or poor health status' than does a high-SES adult.

CONCLUSION

Local health survey analysis can lead to a more precise identification of health needs than surveys attempting to cover broad areas. Low-SES communities within counties, and low-SES individuals, regardless of the community they live in, are substantially more likely to experience fair to poor health. This adds urgency to the need to implement and investigate socio-economic interventions, which can hopefully improve health and save healthcare costs. Novel local area research can identify the impact of intervening variables such as race in addition to SES to add more specificity in identifying populations with the greatest health needs.

Model-based community health surveillance via multilevel small area estimation using state behavioral risk factor surveillance system (BRFSS): a case study in Connecticut

PURPOSE

The main objective of state behavioral risk factor surveillance system (BRFSS) is to produce reliable state-level estimates of various population health outcomes. A multilevel Regression and Post-stratification (MRP) methodology for small area estimation has been applied to the 500 Cities Project to provide population estimates at both city-level and census tract-level using national BRFSS data. To date, MRP has not been applied to any state BRFSS to produce health data at local geographic areas. In addition, the use of single year BRFSS might produce temporary inconsistency in small area estimates (SAEs). The predicted standard errors (SEs) and confidence intervals (CIs) of SAEs using Monte Carlo simulation could be substantially underestimated or overestimated.

METHODS

By extending the current MRP approach and applying a parametric bootstrapping approach to Connecticut BRFSS (CT BRFSS), we were able to produce SAEs as well as SEs and CIs of SAEs for Connecticut counties and towns. We also applied this model to 5-year CT BRFSS (2011-2015) with an aim to improve the temporary consistency of SAEs.

RESULTS

Both single-year and 5-year estimates with SEs and CIs were generated for six selected population health indicators at town, county and state levels. Model-based SAEs were internally evaluated by comparing to single-year and 5-year direct BRFSS survey (2011-2015). SAEs were also externally validated when external data were available.

CONCLUSIONS

Model-based SAEs are valid and could be used to characterize local geographic variations using single state BRFSS data.

Life expectancy by county, race, and ethnicity in the USA, 2000-19: a systematic analysis of health disparities

BACKGROUND

There are large and persistent disparities in life expectancy among racial-ethnic groups in the USA, but the extent to which these patterns vary geographically on a local scale is not well understood. This analysis estimated life expectancy for five racial-ethnic groups, in 3110 US counties over 20 years, to describe spatial-temporal variations in life expectancy and disparities between racial-ethnic groups.

METHODS

We applied novel small-area estimation models to death registration data from the US National Vital Statistics System and population data from the US National Center for Health Statistics to estimate annual sex-specific and age-specific mortality rates stratified by county and racial-ethnic group (non-Latino and non-Hispanic White [White], non-Latino and non-Hispanic Black [Black], non-Latino and non-Hispanic American Indian or Alaska Native [AIAN], non-Latino and non-Hispanic Asian or Pacific Islander [API], and Latino or Hispanic [Latino]) from 2000 to 2019. We adjusted these mortality rates to correct for misreporting of race and ethnicity on death certificates and then constructed abridged life tables to estimate life expectancy at birth.

FINDINGS

Between 2000 and 2019, trends in life expectancy differed among racial-ethnic groups and among counties. Nationally, there was an increase in life expectancy for people who were Black (change 3·9 years [95% uncertainty interval 3·8 to 4·0]; life expectancy in 2019 75·3 years [75·2 to 75·4]), API (2·9 years [2·7 to 3·0]; 85·7 years [85·3 to 86·0]), Latino (2·7 years [2·6 to 2·8]; 82·2 years [82·0 to 82·5]), and White (1·7 years [1·6 to 1·7]; 78·9 years [78·9 to 79·0]), but remained the same for the AIAN population (0·0 years [-0·3 to 0·4]; 73·1 years [71·5 to 74·8]). At the national level, the negative difference in life expectancy for the Black population compared with the White population decreased during this period, whereas the negative difference for the AIAN population compared with the White population increased; in both cases, these patterns were widespread among counties. The positive difference in life expectancy for the API and Latino populations compared with the White population increased at the national level from 2000 to 2019; however, this difference declined in a sizeable minority of counties (615 [42·0%] of 1465 counties) for the Latino population and in most counties (401 [60·2%] of 666 counties) for the API population. For all racial-ethnic groups, improvements in life expectancy were more widespread across counties and larger from 2000 to 2010 than from 2010 to 2019.

INTERPRETATION

Disparities in life expectancy among racial-ethnic groups are widespread and enduring. Local-level data are crucial to address the root causes of poor health and early death among disadvantaged groups in the USA, eliminate health disparities, and increase longevity for all.

FUNDING

National Institute on Minority Health and Health Disparities; National Heart, Lung, and Blood Institute; National Cancer Institute; National Institute on Aging; National Institute of Arthritis and Musculoskeletal and Skin Diseases; Office of Disease Prevention; and Office of Behavioral and Social Science Research, US National Institutes of Health.

Mapping HIV prevalence in Nigeria using small area estimates to develop a targeted HIV intervention strategy

Objective

Although geographically specific data can help target HIV prevention and treatment strategies, Nigeria relies on national- and state-level estimates for policymaking and intervention planning. We calculated sub-state estimates along the HIV continuum of care in Nigeria.

Design

Using data from the Nigeria HIV/AIDS Indicator and Impact Survey (NAIIS) (July–December 2018), we conducted a geospatial analysis estimating three key programmatic indicators: prevalence of HIV infection among adults (aged 15–64 years); antiretroviral therapy (ART) coverage among adults living with HIV; and viral load suppression (VLS) rate among adults living with HIV.

Methods

We used an ensemble modeling method called stacked generalization to analyze available covariates and a geostatistical model to incorporate the output from stacking as well as spatial autocorrelation in the modeled outcomes. Separate models were fitted for each indicator. Finally, we produced raster estimates of each indicator on an approximately 5×5-km grid and estimates at the sub-state/local government area (LGA) and state level.

Results

Estimates for all three indicators varied both within and between states. While state-level HIV prevalence ranged from 0.3% (95% uncertainty interval [UI]: 0.3%–0.5%]) to 4.3% (95% UI: 3.7%–4.9%), LGA prevalence ranged from 0.2% (95% UI: 0.1%–0.5%) to 8.5% (95% UI: 5.8%–12.2%). Although the range in ART coverage did not substantially differ at state level (25.6%–76.9%) and LGA level (21.9%–81.9%), the mean absolute difference in ART coverage between LGAs within states was 16.7 percentage points (range, 3.5–38.5 percentage points). States with large differences in ART coverage between LGAs also showed large differences in VLS—regardless of level of effective treatment coverage—indicating that state-level geographic targeting may be insufficient to address coverage gaps.

Conclusion

Geospatial analysis across the HIV continuum of care can effectively highlight sub-state variation and identify areas that require further attention in order to achieve epidemic control. By generating local estimates, governments, donors, and other implementing partners will be better positioned to conduct targeted interventions and prioritize resource distribution.

US county-level estimation for maternal and infant health-related behavior indicators using pregnancy risk assessment monitoring system data, 2016-2018

Background

There is a critical need for maternal and child health data at the local level (for example, county), yet most counties lack sustainable resources or capabilities to collect local-level data. In such case, model-based small area estimation (SAE) could be a feasible approach. SAE for maternal or infant health-related behaviors at small areas has never been conducted or evaluated.

Methods

We applied multilevel regression with post-stratification approach to produce county-level estimates using Pregnancy Risk Assessment Monitoring System (PRAMS) data, 2016–2018 (n = 65,803 from 23 states) for 2 key outcomes, breastfeeding at 8 weeks and infant non-supine sleeping position.

Results

Among the 1,471 counties, the median model estimate of breastfeeding at 8 weeks was 59.8% (ranged from 34.9 to 87.4%), and the median of infant non-supine sleeping position was 16.6% (ranged from 10.3 to 39.0%). Strong correlations were found between model estimates and direct estimates for both indicators at the state level. Model estimates for both indicators were close to direct estimates in magnitude for Philadelphia County, Pennsylvania.

Conclusion

Our findings support this approach being potentially applied to other maternal and infant health and behavioral indicators in PRAMS to facilitate public health decision-making at the local level.

Generating Subcounty Health Data Products: Methods and Recommendations From a Multistate Pilot Initiative

BACKGROUND

County Health Rankings & Roadmaps (CHR&R) makes data on health determinants and outcomes available at the county level, but health data at subcounty levels are needed. Three pilot projects in California, Missouri, and New York explored multiple approaches for defining measures and producing data at subcounty geographic and demographic levels based on the CHR&R model. This article summarizes the collective technical and implementation considerations from the projects, challenges inherent in analyzing subcounty health data, and lessons learned to inform future subcounty health data projects.

METHODS

The research teams used 12 data sources to produce 40 subcounty measures that replicate or approximate county-level measures from the CHR&R model. Using varying technical methods, the pilot projects followed similar stages: (1) conceptual development of data sources and measures; (2) analysis and presentation of small-area and subpopulation measures for public health, health care, and lay audiences; and (3) positioning the subcounty data initiatives for growth and sustainability. Unique technical considerations, such as degree of data suppression or data stability, arose during the project implementation. A compendium of technical resources, including samples of automated programs for analyzing and reporting subcounty data, was also developed.

RESULTS

The teams summarized the common themes shared by all projects as well as unique technical considerations arising during the project implementation. Furthermore, technical challenges and implementation challenges involved in subcounty data analyses are discussed. Lessons learned and proposed recommendations for prospective analysts of subcounty data are provided on the basis of project experiences, successes, and challenges.

CONCLUSIONS

This multistate pilot project offers 3 successful approaches for creating and disseminating subcounty data products to communities. Subcounty data often are more difficult to obtain than county-level data and require additional considerations such as estimate stability, validating accuracy, and protecting individual confidentiality. We encourage future projects to further refine techniques for addressing these critical considerations.

Environmental Factors Associated with Physical Activity in Rural U.S. Counties

Background: Rural U.S. adults’ prevalence of meeting physical activity (PA) guidelines is lower than urban adults, yet rural-urban differences in environmental influences of adults’ PA are largely unknown. The study’s objective was to identify rural-urban variations in environmental factors associated with the prevalence of adults meeting PA guidelines. Methods: County-level data for non-frontier counties (n = 2697) were used. A five-category rurality variable was created using the percentage of a county’s population living in a rural area. Factor scores from Factor Analyses (FA) were used in subsequent Multiple Linear Regression (MLR) analyses stratified by rurality to identify associations between environmental factor scores and the prevalence of males and females meeting PA guidelines. Results: FA revealed a 13-variable, four-factor structure of natural, social, recreation, and transportation environments. MLR revealed that natural, social, and recreation environments were associated with PA for males and females, with variation by sex for social environment. The natural environment was associated with PA in all but urban counties; the recreation environment was associated with PA in the urban counties and the two most rural counties. Conclusions: Variations across the rural-urban continuum in environmental factors associated with adults’ PA, highlight the uniqueness of rural PA and the need to further study what succeeds in creating active rural places.

Evaluating equality in prescribing Novel Oral Anticoagulants (NOACs) in England: The protocol of a Bayesian small area analysis

Background

Atrial fibrillation (AF) is the most common cardiac arrhythmia, affecting about 1.6% of the population in England. Novel oral anticoagulants (NOACs) are approved AF treatments that reduce stroke risk. In this study, we estimate the equality in individual NOAC prescriptions with high spatial resolution in Clinical Commissioning Groups (CCGs) across England from 2014 to 2019.

Methods

A Bayesian spatio-temporal model will be used to estimate and predict the individual NOAC prescription trend on ‘prescription data’ as an indicator of health services utilisation, using a small area analysis methodology. The main dataset in this study is the “Practice Level Prescribing in England,” which contains four individual NOACs prescribed by all registered GP practices in England. We will use the defined daily dose (DDD) equivalent methodology, as recommended by the World Health Organization (WHO), to compare across space and time. Four licensed NOACs datasets will be summed per 1,000 patients at the CCG-level over time. We will also adjust for CCG-level covariates, such as demographic data, Multiple Deprivation Index, and rural-urban classification. We aim to employ the extended BYM2 model (space-time model) using the RStan package.

Discussion

This study suggests a new statistical modelling approach to link prescription and socioeconomic data to model pharmacoepidemiologic data. Quantifying space and time differences will allow for the evaluation of inequalities in the prescription of NOACs. The methodology will help develop geographically targeted public health interventions, campaigns, audits, or guidelines to improve areas of low prescription. This approach can be used for other medications, especially those used for chronic diseases that must be monitored over time.

Creating local estimates from a population health survey: practical application of small area estimation methods

Regular health surveys can produce reliable estimates at higher geographic levels but not for small areas. Alternatives are to aggregate data over several years or use model-based methods. We created and evaluated model-based estimates for four health-related outcomes by gender, for 153 Local Government Areas using data from the New South Wales Population Health Survey. The evaluation examined evidence on bias and determined the covariates available and appropriate for each outcome variable. The evaluation considered the likely precision of the resulting estimates. The bias and precision of results for single years (2006–2008) for each outcome variable using six covariate specifications were compared with direct survey estimates based on a single year's data and those obtained by aggregating over seven years. A practical issue is how to choose covariates to include in the models as the best covariate specification varies between outcome variables. Model-based results had median root mean squared errors between 3.3% and 5.5% (max 5.2% and 11.3% respectively) and median relative root mean squared errors between 6.8% and 24.5% (max 11.7% and 41.5% respectively). The model-based estimates were unbiased compared with direct estimates based on one or seven years of data and when aggregated to a point where direct estimates were reliable. The bias and reliability assessment process provides a way for policymakers to have confidence in model-based estimates.

Use of Small-Area Estimates to Describe County-Level Geographic Variation in Prevalence of Extreme Obesity Among US Adults

Importance

The prevalence of extreme obesity continues to increase among adults in the US, yet there is an absence of subnational estimates and geographic description of extreme obesity. This shortcoming prevents a thorough understanding of the geographic distribution of extreme obesity, which in turn limits the ability of public health agencies and policy makers to target areas with a known higher prevalence.

Objectives

To use small-area estimation to create county-level estimates of extreme obesity in the US and apply spatial methods to identify clusters of high and low prevalence.

Design, Setting, and Participants

A cross-sectional analysis was conducted using multilevel regression and poststratification with data from the 2012 Behavioral Risk Factor Surveillance System and the US Census Bureau to create prevalence estimates of county-level extreme obesity (body mass index ≥40 [calculated as weight in kilograms divided by height in meters squared]). Data were included on adults (aged ≥18 years) living in the contiguous US. Analysis was performed from June 4 to December 28, 2018.

Main Outcomes and Measures

Multilevel logistic regression models estimated the probability of extreme obesity based on individual-level and area-level characteristics. Census counts were multiplied by these probabilities and summed by county to create county-level prevalence estimates. Moran index values were calculated to assess spatial autocorrelation and identify spatial clusters of hot and cold spots. Estimates of moderate obesity were obtained for comparison.

Results

Overall, the weighted prevalence of extreme obesity was 4.0% (95% CI, 3.9%-4.1%) and the prevalence of moderate obesity was 23.7% (95% CI, 23.4%-23.9%). County-level prevalence of extreme obesity ranged from 1.3% (95% CI, 1.3%-1.3%) to 15.7% (95% CI, 15.3%-16.0%). The Pearson correlation coefficient comparing model-predicted estimates with direct estimates was 0.81 (P < .001). The Moran index I score was 0.35 (P < .001), indicating spatial clustering. Significant clusters of high and low prevalence were identified. Hot spots indicating clustering of high prevalence of extreme obesity in several regions, including the Mississippi Delta region and the Southeast, were identified, as well as clusters of low prevalence in the Rocky Mountain region and the Northeast.

Conclusions and Relevance

Substantial geographic variation was identified in the prevalence of extreme obesity; there was considerable county-level variation even in states generally known as having high or low prevalence of obesity. The results suggest that extreme obesity prevalence demonstrates spatial dependence and clustering and may support the need for substate analysis and benefit of disaggregation of obesity by group. Findings from this study can inform local and national policies seeking to identify populations most at risk from very high body mass index.

Spatial Clusters of Breast Cancer Mortality and Incidence in the Contiguous USA: 2000-2014

BACKGROUND

Clusters of breast cancer with varied incidence or mortality are known to exist. No national scale of analysis of geographical variation in breast cancer incidence has been published before for the contiguous USA.

METHODS

This was a spatial cluster analysis of incidence and mortality data on breast cancer in the contiguous USA at the county resolution. Data for the years 2000-2014 were downloaded and analyzed with the software SaTScan with the goal to identify significant spatial clusters of breast cancer. Regression analysis was used to then adjust breast cancer incidence and mortality for several key risk factors such as age, smoking, particulate matter air pollution, physical inactivity, urban living, education level, and race.

RESULTS

Spatial clusters of counties for higher than expected breast cancer incidence and also for breast cancer mortality were identified. All identified clusters have p < 0.05. The mortality clusters show the mean breast cancer rates inside the cluster, while the incidence clusters show the relative risk inside each cluster. This is the first study of the contiguous USA for breast cancer mortality and incidence together. The clustering for mortality is quite different from the clustering for incidence. Using the software JOINPOINT, it is shown that the annual US downward trend for breast cancer mortality slowed down in recent years.

CONCLUSIONS

There exist several significant clusters in the contiguous USA, both for breast cancer incidence and for breast cancer mortality. Some of the clusters persisted even after adjusting for several key risk factors. These geographic areas warrant further investigation to potentially identify additional local concerns or needs to further address female breast cancer in those specific sites.

Missing information in statewide and national cancer databases: Correlation with health risk factors, geographic disparities, and outcomes

OBJECTIVE

The objectives of this study were to analyze factors associated with outcomes and missing data in women with epithelial ovarian cancer using institutional, state and national databases.

METHODS

Data were abstracted from the University of Virginia cancer registry, Virginia Department of Health (VDH) database, and Surveillance, Epidemiology, and End Results (SEER) Program and analyzed for correlations with demographics, cancer characteristics, and outcomes. Statewide spatial associations between health risk factors such as smoking, obesity, and missing grade/stage were evaluated using bivariate LiSA in Geoda.

RESULTS

There were 524 institutional, 3544 VDH, and 44,464 SEER cases of epithelial ovarian cancer. Institutional cases were younger, most often of white race, had increased grade 1, and decreased unknown grade and stage (all p < 0.001). Significant predictors of unknown grade were non-white race, older age, no surgery, unknown stage/stage IV, and unknown histology/adenocarcinoma. Unknown grade correlated with a significant survival disadvantage. Missing stage and grade correlated with county-level obesity and smoking, as rural regions in Southwest and Southside Virginia had high rates of health risk factors and missing stage/grade compared to urban, affluent regions in Northern Virginia.

CONCLUSIONS

Over a third of nationally reported cases have an unknown grade and 10-20% have an unknown stage which correlates with the worst survival. Predictors of unknown grade include insurance, age, race, smoking status, obesity, and rural setting. Missing data may represent geographical differences or disparities in cancer care available as significantly fewer cases had an unknown grade/stage at a tertiary academic medical center compared to VDH and SEER.

Progress toward eliminating TB and HIV deaths in Brazil, 2001-2015: a spatial assessment

BACKGROUND

Brazil has high burdens of tuberculosis (TB) and HIV, as previously estimated for the 26 states and the Federal District, as well as high levels of inequality in social and health indicators. We improved the geographic detail of burden estimation by modelling deaths due to TB and HIV and TB case fatality ratios for the more than 5400 municipalities in Brazil.

METHODS

This ecological study used vital registration data from the national mortality information system and TB case notifications from the national communicable disease notification system from 2001 to 2015. Mortality due to TB and HIV was modelled separately by cause and sex using a Bayesian spatially explicit mixed effects regression model. TB incidence was modelled using the same approach. Results were calibrated to the Global Burden of Disease Study 2016. Case fatality ratios were calculated for TB.

RESULTS

There was substantial inequality in TB and HIV mortality rates within the nation and within states. National-level TB mortality in people without HIV infection declined by nearly 50% during 2001 to 2015, but HIV mortality declined by just over 20% for males and 10% for females. TB and HIV mortality rates for municipalities in the 90th percentile nationally were more than three times rates in the 10th percentile, with nearly 70% of the worst-performing municipalities for male TB mortality and more than 75% for female mortality in 2001 also in the worst decile in 2015. The same municipality ranking metric for HIV was observed to be between 55% and 61%. Within states, the TB mortality rate ratios by sex for municipalities in the worst decile versus the best decile varied from 1.4 to 2.9, and HIV varied from 1.4 to 4.2. The World Health Organization target case fatality rate for TB of less than 10% was achieved in 9.6% of municipalities for males versus 38.4% for females in 2001 and improved to 38.4% and 56.6% of municipalities for males versus females, respectively, by 2014.

CONCLUSIONS

Mortality rates in municipalities within the same state exhibited nearly as much relative variation as within the nation as a whole. Monitoring the mortality burden at this level of geographic detail is critical for guiding precision public health responses.

Towards a Routine External Evaluation Protocol for Small Area Estimation

Statistical criteria are needed by which to evaluate the potential success or failure of applications of small area estimation. A necessary step to achieve this is a protocol-a series of steps-by which to assess whether an instance of small area estimation has given satisfactory results or not. Most customary attempts at evaluation of small area techniques have deficiencies. Often, evaluation is not attempted. Every small area study requires an external evaluation. With proper planning, this can be routinely achieved, although at some cost, amounting to some sacrifice of efficiency of global estimates. We propose a Routine External Evaluation Protocol to allow us to judge whether, in a given survey, small area estimation has led to accurate results and sound inference.

Estimating missing values in China's official socioeconomic statistics using progressive spatiotemporal Bayesian hierarchical modeling

Due to a large number of missing values, both spatially and temporally, China has not published a complete official socioeconomic statistics dataset at the county level, which is the country’s basic scale of official statistics data collection. We developed a procedure to impute the missing values under the Bayesian hierarchical modeling framework. The procedure incorporates two novelties. First, it takes into account spatial autocorrelations and temporal trends for those easier-to-impute variables with small missing percentages. Second, it further uses the first-step complete variables as covariate information to improve the modeling of more-difficult-to-impute variables with large missing percentages. We applied this progressive spatiotemporal (PST) method to China’s official socioeconomic statistics during 2002–2011 and compared it with four other widely used imputation methods, including k-nearest neighbors (kNN), expectation maximum (EM), singular value decomposition (SVD) and random forest (RF). The results show that the PST method outperforms these methods, thus proving the effects of sophisticatedly incorporating the additional spatial and temporal information and progressively utilizing the covariate information. This study has an outcome that allows China to construct a complete socioeconomic dataset and establishes a methodology that can be generally useful for estimating missing values in large spatiotemporal datasets.

A Flexible Bayesian Model for Estimating Subnational Mortality

Reliable subnational mortality estimates are essential in the study of health inequalities within a country. One of the difficulties in producing such estimates is the presence of small populations among which the stochastic variation in death counts is relatively high, and thus the underlying mortality levels are unclear. We present a Bayesian hierarchical model to estimate mortality at the subnational level. The model builds on characteristic age patterns in mortality curves, which are constructed using principal components from a set of reference mortality curves. Information on mortality rates are pooled across geographic space and are smoothed over time. Testing of the model shows reasonable estimates and uncertainty levels when it is applied both to simulated data that mimic U.S. counties and to real data for French départements. The model estimates have direct applications to the study of subregional health patterns and disparities.

Calculating census tract-based life expectancy in New York state: a generalizable approach

BACKGROUND

Life expectancy at birth (LE) has been calculated for states and counties. LE estimates at these levels mask health disparities in local communities. There are no nationwide estimates at the sub-county level. We present a stepwise approach for calculating LE using census tracts in New York state to identify health disparities.

METHODS

Our study included 2751 census tracts in New York state, but excluded New York City. We used population data from the 2010 United States Census and 2008-2010 mortality data from the state health department. Tracts were assigned to 99.97% of the deaths. We removed tracts which had a majority of people living in group quarters. Deaths in these tracts are often recorded elsewhere. Of the remaining 2679 tracts, 6.6% of the tracts had standard errors ≥ 2 years. A geographic aggregation tool was used to aggregate tracts with fewer than 60 deaths, and then aggregate areas that had standard errors of ≥ 2 years.

RESULTS

Aggregation resulted in a 9.9% reduction in the number of areas. Tracts with < 2% of population living below the poverty level had a LE of 82.8 years, while tracts with a poverty level ≥ 25% had a LE of 75.5. We observed differences in LE in border areas, of up to 10.4 years, when excluding or including deaths of study area residents that occurred outside the study area. The range and standard deviation at the county level (77.5-82.8, SD = 1.2 years) were smaller than our final sub-county areas (64.7-92.0, SD = 3.3 years). The correlation between LE and poverty were similar and statistically significant (p < 0.0001) at the county (r = - 0.58) and sub-county level (r = - 0.58). The correlations between LE and percent African-American at the county level were (r = 0.11, p = 0.43) and at the sub-county level (r = - 0.25, p < 0.0001).

CONCLUSION

The proposed approach for geocoding and aggregation of mortality and population data provides a solution for health departments to produce stable empirically-derived LE estimates using data coded to the tract. Reliable estimates within sub-county areas are needed to aid public health officials in focusing preventive health programs in areas where health disparities would be masked by county level estimates.

US County-Level Trends in Mortality Rates for Major Causes of Death, 1980-2014

IMPORTANCE

County-level patterns in mortality rates by cause have not been systematically described but are potentially useful for public health officials, clinicians, and researchers seeking to improve health and reduce geographic disparities.

OBJECTIVES

To demonstrate the use of a novel method for county-level estimation and to estimate annual mortality rates by US county for 21 mutually exclusive causes of death from 1980 through 2014.

DESIGN, SETTING, AND PARTICIPANTS

Redistribution methods for garbage codes (implausible or insufficiently specific cause of death codes) and small area estimation methods (statistical methods for estimating rates in small subpopulations) were applied to death registration data from the National Vital Statistics System to estimate annual county-level mortality rates for 21 causes of death. These estimates were raked (scaled along multiple dimensions) to ensure consistency between causes and with existing national-level estimates. Geographic patterns in the age-standardized mortality rates in 2014 and in the change in the age-standardized mortality rates between 1980 and 2014 for the 10 highest-burden causes were determined.

EXPOSURE

County of residence.

MAIN OUTCOMES AND MEASURES

Cause-specific age-standardized mortality rates.

RESULTS

A total of 80 412 524 deaths were recorded from January 1, 1980, through December 31, 2014, in the United States. Of these, 19.4 million deaths were assigned garbage codes. Mortality rates were analyzed for 3110 counties or groups of counties. Large between-county disparities were evident for every cause, with the gap in age-standardized mortality rates between counties in the 90th and 10th percentiles varying from 14.0 deaths per 100 000 population (cirrhosis and chronic liver diseases) to 147.0 deaths per 100 000 population (cardiovascular diseases). Geographic regions with elevated mortality rates differed among causes: for example, cardiovascular disease mortality tended to be highest along the southern half of the Mississippi River, while mortality rates from self-harm and interpersonal violence were elevated in southwestern counties, and mortality rates from chronic respiratory disease were highest in counties in eastern Kentucky and western West Virginia. Counties also varied widely in terms of the change in cause-specific mortality rates between 1980 and 2014. For most causes (eg, neoplasms, neurological disorders, and self-harm and interpersonal violence), both increases and decreases in county-level mortality rates were observed.

CONCLUSIONS AND RELEVANCE

In this analysis of US cause-specific county-level mortality rates from 1980 through 2014, there were large between-county differences for every cause of death, although geographic patterns varied substantially by cause of death. The approach to county-level analyses with small area models used in this study has the potential to provide novel insights into US disease-specific mortality time trends and their differences across geographic regions.

Geography of Adolescent Obesity in the U.S., 2007-2011

INTRODUCTION

Obesity remains a significant threat to the current and long-term health of U.S. adolescents. The authors developed county-level estimates of adolescent obesity for the contiguous U.S., and then explored the association between 23 conceptually derived area-based correlates of adolescent obesity and ecologic obesity prevalence.

METHODS

Multilevel small area regression methods applied to the 2007 and 2011-2012 National Survey of Children's Health produced county-level obesity prevalence estimates for children aged 10-17 years. Exploratory multivariable Bayesian regression estimated the cross-sectional association between nutrition, activity, and macrosocial characteristics of counties and states, and county-level obesity prevalence. All analyses were conducted in 2015.

RESULTS

Adolescent obesity varies geographically with clusters of high prevalence in the Deep South and Southern Appalachian regions. Geographic disparities and clustering in observed data are largely explained by hypothesized area-based variables. In adjusted models, activity environment, but not nutrition environment variables were associated with county-level obesity prevalence. County violent crime was associated with higher obesity, whereas recreational facility density was associated with lower obesity. Measures of the macrosocial and relational domain, including community SES, community health, and social marginalization, were the strongest correlates of county-level obesity.

CONCLUSIONS

County-level estimates of adolescent obesity demonstrate notable geographic disparities, which are largely explained by conceptually derived area-based contextual measures. This ecologic exploratory study highlights the importance of taking a multidimensional approach to understanding the social and community context in which adolescents make obesity-relevant behavioral choices.

Diagnosed and Undiagnosed Diabetes Prevalence by County in the U.S., 1999-2012

OBJECTIVE

Previous analyses of diabetes prevalence in the U.S. have considered either only large geographic regions or only individuals in whom diabetes had been diagnosed. We estimated county-level trends in the prevalence of diagnosed, undiagnosed, and total diabetes as well as rates of diagnosis and effective treatment from 1999 to 2012.

RESEARCH DESIGN AND METHODS

We used a two-stage modeling procedure. In the first stage, self-reported and biomarker data from the National Health and Nutrition Examination Survey (NHANES) were used to build models for predicting true diabetes status, which were applied to impute true diabetes status for respondents in the Behavioral Risk Factor Surveillance System (BRFSS). In the second stage, small area models were fit to imputed BRFSS data to derive county-level estimates of diagnosed, undiagnosed, and total diabetes prevalence, as well as rates of diabetes diagnosis and effective treatment.

RESULTS

In 2012, total diabetes prevalence ranged from 8.8% to 26.4% among counties, whereas the proportion of the total number of cases that had been diagnosed ranged from 59.1% to 79.8%, and the proportion of successfully treated individuals ranged from 19.4% to 31.0%. Total diabetes prevalence increased in all counties between 1999 and 2012; however, the rate of increase varied widely. Over the same period, rates of diagnosis increased in all counties, while rates of effective treatment stagnated.

CONCLUSIONS

Our findings demonstrate substantial disparities in diabetes prevalence, rates of diagnosis, and rates of effective treatment within the U.S. These findings should be used to target high-burden areas and select the right mix of public health strategies.

Surveillance of HIV assisted partner services using routine health information systems in Kenya

BACKGROUND

The utilization of routine health information systems (HIS) for surveillance of assisted partner services (aPS) for HIV in sub-Saharan is sub-optimal, in part due to poor data quality and limited use of information technology. Consequently, little is known about coverage, scope and quality of HIV aPS. Yet, affordable electronic data tools, software and data transmission infrastructure are now widely accessible in sub-Saharan Africa.

METHODS

We designed and implemented a cased-based surveillance system using the HIV testing platform in 18 health facilities in Kenya. The components of this system included an electronic HIV Testing and Counseling (HTC) intake form, data transmission on the Global Systems for Mobile Communication (GSM), and data collection using the Open Data Kit (ODK) platform. We defined rates of new HIV diagnoses, and characterized HIV-infected cases. We also determined the proportion of clients who reported testing for HIV because a) they were notified by a sexual partner b) they were notified by a health provider, or c) they were informed of exposure by another other source. Data collection times were evaluated.

RESULTS

Among 4351 clients, HIV prevalence was 14.2 %, ranging from 4.4-25.4 % across facilities. Regardless of other reasons for testing, only 107 (2.5 %) of all participants reported testing after being notified by a health provider or sexual partner. A similar proportion, 1.8 % (79 of 4351), reported partner notification as the only reason for seeking an HIV test. Among 79 clients who reported HIV partner services as the reason for testing, the majority (78.5 %), were notified by their sexual partners. The majority (52.8 %) of HIV-infected patients initiated their HIV testing, and 57.2 % tested in a Voluntary Counseling and Testing (VCT) site co-located in a health facility. Median time for data capture was 4 min (IQR: 3-15), with a longer duration for HIV-infected participants, and there was no reported data loss.

CONCLUSION

aPS surveillance using new technologies is feasible, and could be readily expanded into HIV registries in Kenya and other sub-Saharan countries. Partner services are under-utilized in Kenya but further documentation of coverage and implementation gaps for HIV and aPS services is required.

Regional Differences in Intervention Coverage and Health System Strength in Tanzania

Background

Assessments of subnational progress and performance coverage within countries should be an integral part of health sector reviews, using recent data from multiple sources on health system strength and coverage.

Method

As part of the midterm review of the national health sector strategic plan of Tanzania mainland, summary measures of health system strength and coverage of interventions were developed for all 21 regions, focusing on the priority indicators of the national plan. Household surveys, health facility data and administrative databases were used to compute the regional scores.

Findings

Regional Millennium Development Goal (MDG) intervention coverage, based on 19 indicators, ranged from 47% in Shinyanga in the northwest to 71% in Dar es Salaam region. Regions in the eastern half of the country have higher coverage than in the western half of mainland. The MDG coverage score is strongly positively correlated with health systems strength (r = 0.84). Controlling for socioeconomic status in a multivariate analysis has no impact on the association between the MDG coverage score and health system strength. During 1991–2010 intervention coverage improved considerably in all regions, but the absolute gap between the regions did not change during the past two decades, with a gap of 22% between the top and bottom three regions.

Interpretation

The assessment of regional progress and performance in 21 regions of mainland Tanzania showed considerable inequalities in coverage and health system strength and allowed the identification of high and low-performing regions. Using summary measures derived from administrative, health facility and survey data, a subnational picture of progress and performance can be obtained for use in regular health sector reviews.

Regions of High Out-Of-Hospital Cardiac Arrest Incidence and Low Bystander CPR Rates in Victoria, Australia

Background

Out-of-hospital cardiac arrest (OHCA) remains a major public health issue and research has shown that large regional variation in outcomes exists. Of the interventions associated with survival, the provision of bystander CPR is one of the most important modifiable factors. The aim of this study is to identify census areas with high incidence of OHCA and low rates of bystander CPR in Victoria, Australia

Methods

We conducted an observational study using prospectively collected population-based OHCA data from the state of Victoria in Australia. Using ArcGIS (ArcMap 10.0), we linked the location of the arrest using the dispatch coordinates (longitude and latitude) to Victorian Local Government Areas (LGAs). We used Bayesian hierarchical models with random effects on each LGA to provide shrunken estimates of the rates of bystander CPR and the incidence rates.

Results

Over the study period there were 31,019 adult OHCA attended, of which 21,436 (69.1%) cases were of presumed cardiac etiology. Significant variation in the incidence of OHCA among LGAs was observed. There was a 3 fold difference in the incidence rate between the lowest and highest LGAs, ranging from 38.5 to 115.1 cases per 100,000 person-years. The overall rate of bystander CPR for bystander witnessed OHCAs was 62.4%, with the rate increasing from 56.4% in 2008–2010 to 68.6% in 2010–2013. There was a 25.1% absolute difference in bystander CPR rates between the highest and lowest LGAs.

Conclusion

Significant regional variation in OHCA incidence and bystander CPR rates exists throughout Victoria. Regions with high incidence and low bystander CPR participation can be identified and would make suitable targets for interventions to improve CPR participation rates.

Validation of multilevel regression and poststratification methodology for small area estimation of health indicators from the behavioral risk factor surveillance system

Small area estimation is a statistical technique used to produce reliable estimates for smaller geographic areas than those for which the original surveys were designed. Such small area estimates (SAEs) often lack rigorous external validation. In this study, we validated our multilevel regression and poststratification SAEs from 2011 Behavioral Risk Factor Surveillance System data using direct estimates from 2011 Missouri County-Level Study and American Community Survey data at both the state and county levels. Coefficients for correlation between model-based SAEs and Missouri County-Level Study direct estimates for 115 counties in Missouri were all significantly positive (0.28 for obesity and no health-care coverage, 0.40 for current smoking, 0.51 for diabetes, and 0.69 for chronic obstructive pulmonary disease). Coefficients for correlation between model-based SAEs and American Community Survey direct estimates of no health-care coverage were 0.85 at the county level (811 counties) and 0.95 at the state level. Unweighted and weighted model-based SAEs were compared with direct estimates; unweighted models performed better. External validation results suggest that multilevel regression and poststratification model-based SAEs using single-year Behavioral Risk Factor Surveillance System data are valid and could be used to characterize geographic variations in health indictors at local levels (such as counties) when high-quality local survey data are not available.

An assessment of the impact of the JSY cash transfer program on maternal mortality reduction in Madhya Pradesh, India

BACKGROUND

The Indian Janani Suraksha Yojana (JSY) program is a demand-side program in which the state pays women a cash incentive to deliver in an institution, with the aim of reducing maternal mortality. The JSY has had 54 million beneficiaries since inception 7 years ago. Although a number of studies have demonstrated the effect of JSY on coverage, few have examined the direct impact of the program on maternal mortality.

OBJECTIVE

To study the impact of JSY on maternal mortality in Madhya Pradesh (MP), one of India's largest provinces.

DESIGN

By synthesizing data from various sources, district-level maternal mortality ratios (MMR) from 2005 to 2010 were estimated using a Bayesian spatio-temporal model. Based on these, a mixed effects multilevel regression model was applied to assess the impact of JSY. Specifically, the association between JSY intensity, as reflected by 1) proportion of JSY-supported institutional deliveries, 2) total annual JSY expenditure, and 3) MMR, was examined.

RESULTS

The proportion of all institutional deliveries increased from 23.9% in 2005 to 55.9% in 2010 province-wide. The proportion of JSY-supported institutional deliveries rose from 14% (2005) to 80% (2010). MMR declines in the districts varied from 2 to 35% over this period. Despite the marked increase in JSY-supported delivery, our multilevel models did not detect a significant association between JSY-supported delivery proportions and changes in MMR in the districts. The results from the analysis examining the association between MMR and JSY expenditure are similar.

CONCLUSIONS

Our analysis was unable to detect an association between maternal mortality reduction and the JSY in MP. The high proportion of institutional delivery under the program does not seem to have converted to lower mortality outcomes. The lack of significant impact could be related to supply-side constraints. Demand-side programs like JSY will have a limited effect if the supply side is unable to deliver care of adequate quality.

Scalable combinatorial tools for health disparities research

Despite staggering investments made in unraveling the human genome, current estimates suggest that as much as 90% of the variance in cancer and chronic diseases can be attributed to factors outside an individual’s genetic endowment, particularly to environmental exposures experienced across his or her life course. New analytical approaches are clearly required as investigators turn to complicated systems theory and ecological, place-based and life-history perspectives in order to understand more clearly the relationships between social determinants, environmental exposures and health disparities. While traditional data analysis techniques remain foundational to health disparities research, they are easily overwhelmed by the ever-increasing size and heterogeneity of available data needed to illuminate latent gene x environment interactions. This has prompted the adaptation and application of scalable combinatorial methods, many from genome science research, to the study of population health. Most of these powerful tools are algorithmically sophisticated, highly automated and mathematically abstract. Their utility motivates the main theme of this paper, which is to describe real applications of innovative transdisciplinary models and analyses in an effort to help move the research community closer toward identifying the causal mechanisms and associated environmental contexts underlying health disparities. The public health exposome is used as a contemporary focus for addressing the complex nature of this subject.

Effective coverage: a metric for monitoring Universal Health Coverage

A major challenge in monitoring universal health coverage (UHC) is identifying an indicator that can adequately capture the multiple components underlying the UHC initiative. Effective coverage, which unites individual and intervention characteristics into a single metric, offers a direct and flexible means to measure health system performance at different levels. We view effective coverage as a relevant and actionable metric for tracking progress towards achieving UHC. In this paper, we review the concept of effective coverage and delineate the three components of the metric - need, use, and quality - using several examples. Further, we explain how the metric can be used for monitoring interventions at both local and global levels. We also discuss the ways that current health information systems can support generating estimates of effective coverage. We conclude by recognizing some of the challenges associated with producing estimates of effective coverage. Despite these challenges, effective coverage is a powerful metric that can provide a more nuanced understanding of whether, and how well, a health system is delivering services to its populations.

The use of continuous surveys to generate and continuously report high quality timely maternal and newborn health data at the district level in Tanzania and Uganda

BACKGROUND

The lack of high quality timely data for evidence-informed decision making at the district level presents a challenge to improving maternal and newborn survival in low income settings. To address this problem, the EQUIP project (Expanded Quality Management using Information Power) implemented a continuous household and health facility survey for continuous feedback of data in two districts each in Tanzania and Uganda as part of a quality improvement innovation for mothers and newborns.

METHODS

Within EQUIP, continuous survey data were used for quality improvement (intervention districts) and for effect evaluation (intervention and comparison districts). Over 30 months of intervention (November 2011 to April 2014), EQUIP conducted continuous cross-sectional household and health facility surveys using 24 independent probability samples of household clusters to represent each district each month, and repeat censuses of all government health facilities. Using repeat samples in this way allowed data to be aggregated at six four-monthly intervals to track progress over time for evaluation, and for continuous feedback to quality improvement teams in intervention districts.In both countries, one continuous survey team of eight people was employed to complete approximately 7,200 household and 200 facility interviews in year one. Data were collected using personal digital assistants. After every four months, routine tabulations of indicators were produced and synthesized to report cards for use by the quality improvement teams.

RESULTS

The first 12 months were implemented as planned. Completion of household interviews was 96% in Tanzania and 91% in Uganda. Indicators across the continuum of care were tabulated every four months, results discussed by quality improvement teams, and report cards generated to support their work.

CONCLUSIONS

The EQUIP continuous surveys were feasible to implement as a method to continuously generate and report on demand and supply side indicators for maternal and newborn health; they have potential to be expanded to include other health topics. Documenting the design and implementation of a continuous data collection and feedback mechanism for prospective description, quality improvement, and evaluation in a low-income setting potentially represents a new paradigm that places equal weight on data systems for course correction, as well as evaluation.

The use of sampling weights in Bayesian hierarchical models for small area estimation

Hierarchical modeling has been used extensively for small area estimation. However, design weights that are required to reflect complex surveys are rarely considered in these models. We develop computationally efficient, Bayesian spatial smoothing models that acknowledge the design weights. Computation is carried out using the integrated nested Laplace approximation, which is fast. An extensive simulation study is presented that considers the effects of non-response and non-random selection of individuals, allowing examination of the impact of ignoring the design weights and the benefits of spatial smoothing. The results show that, when compared with standard approaches, mean squared error can be greatly reduced with the proposed methods. Bias reduction occurs through the inclusion of the design weights, with variance reduction being achieved through hierarchical smoothing. We analyze data from the Washington State 2006 Behavioral Risk Factor Surveillance System. The models are easily and quickly fitted within the R environment, using existing packages.

Multilevel regression and poststratification for small-area estimation of population health outcomes: a case study of chronic obstructive pulmonary disease prevalence using the behavioral risk factor surveillance system

A variety of small-area statistical models have been developed for health surveys, but none are sufficiently flexible to generate small-area estimates (SAEs) to meet data needs at different geographic levels. We developed a multilevel logistic model with both state- and nested county-level random effects for chronic obstructive pulmonary disease (COPD) using 2011 data from the Behavioral Risk Factor Surveillance System. We applied poststratification with the (decennial) US Census 2010 counts of census-block population to generate census-block-level SAEs of COPD prevalence which could be conveniently aggregated to all other census geographic units, such as census tracts, counties, and congressional districts. The model-based SAEs and direct survey estimates of COPD prevalence were quite consistent at both the county and state levels. The Pearson correlation coefficient was 0.99 at the state level and ranged from 0.88 to 0.95 at the county level. Our extended multilevel regression modeling and poststratification approach could be adapted for other geocoded national health surveys to generate reliable SAEs for population health outcomes at all administrative and legislative geographic levels of interest in a scalable framework.

Cigarette smoking prevalence in US counties: 1996-2012

BACKGROUND

Cigarette smoking is a leading risk factor for morbidity and premature mortality in the United States, yet information about smoking prevalence and trends is not routinely available below the state level, impeding local-level action.

METHODS

We used data on 4.7 million adults age 18 and older from the Behavioral Risk Factor Surveillance System (BRFSS) from 1996 to 2012. We derived cigarette smoking status from self-reported data in the BRFSS and applied validated small area estimation methods to generate estimates of current total cigarette smoking prevalence and current daily cigarette smoking prevalence for 3,127 counties and county equivalents annually from 1996 to 2012. We applied a novel method to correct for bias resulting from the exclusion of the wireless-only population in the BRFSS prior to 2011.

RESULTS

Total cigarette smoking prevalence varies dramatically between counties, even within states, ranging from 9.9% to 41.5% for males and from 5.8% to 40.8% for females in 2012. Counties in the South, particularly in Kentucky, Tennessee, and West Virginia, as well as those with large Native American populations, have the highest rates of total cigarette smoking, while counties in Utah and other Western states have the lowest. Overall, total cigarette smoking prevalence declined between 1996 and 2012 with a median decline across counties of 0.9% per year for males and 0.6% per year for females, and rates of decline for males and females in some counties exceeded 3% per year. Statistically significant declines were concentrated in a relatively small number of counties, however, and more counties saw statistically significant declines in male cigarette smoking prevalence (39.8% of counties) than in female cigarette smoking prevalence (16.2%). Rates of decline varied by income level: counties in the top quintile in terms of income experienced noticeably faster declines than those in the bottom quintile.

CONCLUSIONS

County-level estimates of cigarette smoking prevalence provide a unique opportunity to assess where prevalence remains high and where progress has been slow. These estimates provide the data needed to better develop and implement strategies at a local and at a state level to further reduce the burden imposed by cigarette smoking.

Self-rated health: small area large area comparisons amongst older adults at the state, district and sub-district level in India

We compared prevalence estimates of self-rated health (SRH) derived indirectly using four different small area estimation methods for the Vadu (small) area from the national Study on Global AGEing (SAGE) survey with estimates derived directly from the Vadu SAGE survey. The indirect synthetic estimate for Vadu was 24% whereas the model based estimates were 45.6% and 45.7% with smaller prediction errors and comparable to the direct survey estimate of 50%. The model based techniques were better suited to estimate the prevalence of SRH than the indirect synthetic method. We conclude that a simplified mixed effects regression model can produce valid small area estimates of SRH.

The geographic distribution of obesity by census tract among 59 767 insured adults in King County, WA

OBJECTIVE

To evaluate the geographic concentration of adult obesity prevalence by census tract (CT) in King County, WA, in relation to social and economic factors.

METHODS AND DESIGN

Measured heights and weights from 59 767 adult men and women enrolled in the Group Health (GH) healthcare system were used to estimate obesity prevalence at the CT level. CT-level measures of socioeconomic status (SES) were median home values of owner-occupied housing units, percent of residents with a college degree and median household incomes, all drawn from the 2000 Census. Spatial regression models were used to assess the relation between CT-level obesity prevalence and socioeconomic variables.

RESULTS

Smoothed CT obesity prevalence, obtained using an Empirical Bayes tool, ranged from 16.2-43.7% (a 2.7-fold difference). The spatial pattern of obesity was non-random, showing a concentration in south and southeast King County. In spatial regression models, CT-level home values and college education were more strongly associated with obesity than household incomes. For each additional $100 000 in median home values, CT obesity prevalence was 2.3% lower. The three SES factors together explained 70% of the variance in CT obesity prevalence after accounting for population density, race/ethnicity, age and spatial dependence.

CONCLUSIONS

To our knowledge, this is the first report to show major social disparities in adult obesity prevalence at the CT scale that is based, moreover, on measured heights and weights. Analyses of data at sufficiently fine geographic scale are needed to guide targeted local interventions to stem the obesity epidemic.

Using empirical Bayes methods to rank counties on population health measures

The University of Wisconsin Population Health Institute has published County Health Rankings (The Rankings) since 2010. These rankings use population-based data to highlight variation in health and encourage health assessment for all US counties. However, the uncertainty of estimates remains a limitation. We sought to quantify the precision of TheRankings for selected measures. We developed hierarchical models for 5 health outcome measures and applied empirical Bayes methods to obtain county rank estimates for a composite health outcome measure. We compared results using models with and without demographic fixed effects to determine whether covariates improved rank precision. Counties whose rank had wide confidence intervals had smaller populations or ranked in the middle of all counties for health outcomes. Incorporating covariates in the models produced narrower intervals, but rank estimates remained imprecise for many counties. Local health officials, especially in smaller population and mid-performing communities, should consider these limitations when interpreting the results of TheRankings.

Prevalence of physical activity and obesity in US counties, 2001-2011: a road map for action

Background

Obesity and physical inactivity are associated with several chronic conditions, increased medical care costs, and premature death.

Methods

We used the Behavioral Risk Factor Surveillance System (BRFSS), a state-based random-digit telephone survey that covers the majority of United States counties, and the National Health and Nutrition Examination Survey (NHANES), a nationally representative sample of the US civilian noninstitutionalized population. About 3.7 million adults aged 20 years or older participated in the BRFSS from 2000 to 2011, and 30,000 adults aged 20 or older participated in NHANES from 1999 to 2010. We calculated body mass index (BMI) from self-reported weight and height in the BRFSS and adjusted for self-reporting bias using NHANES. We calculated self-reported physical activity—both any physical activity and physical activity meeting recommended levels—from self-reported data in the BRFSS. We used validated small area estimation methods to generate estimates of obesity and physical activity prevalence for each county annually for 2001 to 2011.

Results

Our results showed an increase in the prevalence of sufficient physical activity from 2001 to 2009. Levels were generally higher in men than in women, but increases were greater in women than men. Counties in Kentucky, Florida, Georgia, and California reported the largest gains. This increase in level of activity was matched by an increase in obesity in almost all counties during the same time period. There was a low correlation between level of physical activity and obesity in US counties. From 2001 to 2009, controlling for changes in poverty, unemployment, number of doctors per 100,000 population, percent rural, and baseline levels of obesity, for every 1 percentage point increase in physical activity prevalence, obesity prevalence was 0.11 percentage points lower.

Conclusions

Our study showed that increased physical activity alone has a small impact on obesity prevalence at the county level in the US. Indeed, the rise in physical activity levels will have a positive independent impact on the health of Americans as it will reduce the burden of cardiovascular diseases and diabetes. Other changes such as reduction in caloric intake are likely needed to curb the obesity epidemic and its burden.

Left behind: widening disparities for males and females in US county life expectancy, 1985-2010

BACKGROUND

The United States spends more than any other country on health care. The poor relative performance of the US compared to other high-income countries has attracted attention and raised questions about the performance of the US health system. An important dimension to poor national performance is the large disparities in life expectancy.

METHODS

We applied a mixed effects Poisson statistical model and Gaussian Process Regression to estimate age-specific mortality rates for US counties from 1985 to 2010. We generated uncertainty distributions for life expectancy at each age using standard simulation methods.

RESULTS

Female life expectancy in the United States increased from 78.0 years in 1985 to 80.9 years in 2010, while male life expectancy increased from 71.0 years in 1985 to 76.3 years in 2010. The gap between female and male life expectancy in the United States was 7.0 years in 1985, narrowing to 4.6 years in 2010. For males at the county level, the highest life expectancy steadily increased from 75.5 in 1985 to 81.7 in 2010, while the lowest life expectancy remained under 65. For females at the county level, the highest life expectancy increased from 81.1 to 85.0, and the lowest life expectancy remained around 73. For male life expectancy at the county level, there have been three phases in the evolution of inequality: a period of rising inequality from 1985 to 1993, a period of stable inequality from 1993 to 2002, and rising inequality from 2002 to 2010. For females, in contrast, inequality has steadily increased during the 25-year period. Compared to only 154 counties where male life expectancy remained stagnant or declined, 1,405 out of 3,143 counties (45%) have seen no significant change or a significant decline in female life expectancy from 1985 to 2010. In all time periods, the lowest county-level life expectancies are seen in the South, the Mississippi basin, West Virginia, Kentucky, and selected counties with large Native American populations.

CONCLUSIONS

The reduction in the number of counties where female life expectancy at birth is declining in the most recent period is welcome news. However, the widening disparities between counties and the slow rate of increase compared to other countries should be viewed as a call for action. An increased focus on factors affecting health outcomes, morbidity, and mortality such as socioeconomic factors, difficulty of access to and poor quality of health care, and behavioral, environmental, and metabolic risk factors is urgently required.

Small area variation in diabetes prevalence in Puerto Rico

OBJECTIVE

To estimate the 2009 prevalence of diagnosed diabetes in Puerto Rico among adults ≥ 20 years of age in order to gain a better understanding of its geographic distribution so that policymakers can more efficiently target prevention and control programs.

METHODS

A Bayesian multilevel model was fitted to the combined 2008-2010 Behavioral Risk Factor Surveillance System and 2009 United States Census data to estimate diabetes prevalence for each of the 78 municipios (counties) in Puerto Rico.

RESULTS

The mean unadjusted estimate for all counties was 14.3% (range by county, 9.9%-18.0%). The average width of the confidence intervals was 6.2%. Adjusted and unadjusted estimates differed little.

CONCLUSIONS

These 78 county estimates are higher on average and showed less variability (i.e., had a smaller range) than the previously published estimates of the 2008 diabetes prevalence for all United States counties (mean, 9.9%; range, 3.0%-18.2%).

A multilevel approach to estimating small area childhood obesity prevalence at the census block-group level

Introduction

Traditional survey methods for obtaining nationwide small-area estimates (SAEs) of childhood obesity are costly. This study applied a geocoded national health survey in a multilevel modeling framework to estimate prevalence of childhood obesity at the census block-group level.

Methods

We constructed a multilevel logistic regression model to evaluate the influence of individual demographic characteristics, zip code, county, and state on the childhood obesity measures from the 2007 National Survey of Children’s Health. The obesity risk for a child in each census block group was then estimated on the basis of this multilevel model. We compared direct survey and model-based SAEs to evaluate the model specification.

Results

Multilevel models in this study explained about 60% of state-level variances associated with childhood obesity, 82.8% to 86.5% of county-level, and 93.1% of zip code-level. The 95% confidence intervals of block- group level SAEs have a wide range (0.795-20.0), a low median of 2.02, and a mean of 2.12. The model-based SAEs of childhood obesity prevalence ranged from 2.3% to 54.7% with a median of 16.0% at the block-group level.

Conclusion

The geographic variances among census block groups, counties, and states demonstrate that locale may be as significant as individual characteristics such as race/ethnicity in the development of the childhood obesity epidemic. Our estimates provide data to identify priority areas for local health programs and to establish feasible local intervention goals. Model-based SAEs of population health outcomes could be a tool of public health assessment and surveillance.

Prevalence, awareness, treatment, and control of hypertension in United States counties, 2001-2009

Hypertension is an important and modifiable risk factor for cardiovascular disease and mortality. Over the last decade, national-levels of controlled hypertension have increased, but little information on hypertension prevalence and trends in hypertension treatment and control exists at the county-level. We estimate trends in prevalence, awareness, treatment, and control of hypertension in US counties using data from the National Health and Nutrition Examination Survey (NHANES) in five two-year waves from 1999-2008 including 26,349 adults aged 30 years and older and from the Behavioral Risk Factor Surveillance System (BRFSS) from 1997-2009 including 1,283,722 adults aged 30 years and older. Hypertension was defined as systolic blood pressure (BP) of at least 140 mm Hg, self-reported use of antihypertensive treatment, or both. Hypertension control was defined as systolic BP less than 140 mm Hg. The median prevalence of total hypertension in 2009 was estimated at 37.6% (range: 26.5 to 54.4%) in men and 40.1% (range: 28.5 to 57.9%) in women. Within-state differences in the county prevalence of uncontrolled hypertension were as high as 7.8 percentage points in 2009. Awareness, treatment, and control was highest in the southeastern US, and increased between 2001 and 2009 on average. The median county-level control in men was 57.7% (range: 43.4 to 65.9%) and in women was 57.1% (range: 43.0 to 65.46%) in 2009, with highest rates in white men and black women. While control of hypertension is on the rise, prevalence of total hypertension continues to increase in the US. Concurrent increases in treatment and control of hypertension are promising, but efforts to decrease the prevalence of hypertension are needed.

Bayesian Small Area Estimates of Diabetes Incidence by United States County, 2009

In the United States, diabetes is common and costly. Programs to prevent new cases of diabetes are often carried out at the level of the county, a unit of local government. Thus, efficient targeting of such programs requires county-level estimates of diabetes incidence-the fraction of the non-diabetic population who received their diagnosis of diabetes during the past 12 months. Previously, only estimates of prevalence-the overall fraction of population who have the disease-have been available at the county level. Counties with high prevalence might or might not be the same as counties with high incidence, due to spatial variation in mortality and relocation of persons with incident diabetes to another county. Existing methods cannot be used to estimate county-level diabetes incidence, because the fraction of the population who receive a diabetes diagnosis in any year is too small. Here, we extend previously developed methods of Bayesian small-area estimation of prevalence, using diffuse priors, to estimate diabetes incidence for all U.S. counties based on data from a survey designed to yield state-level estimates. We found high incidence in the southeastern United States, the Appalachian region, and in scattered counties throughout the western U.S. Our methods might be applicable in other circumstances in which all cases of a rare condition also must be cases of a more common condition (in this analysis, "newly diagnosed cases of diabetes" and "cases of diabetes"). If appropriate data are available, our methods can be used to estimate proportion of the population with the rare condition at greater geographic specificity than the data source was designed to provide.

Disentangling the effects of risk factors and clinical care on subnational variation in early neonatal mortality in the United States

OBJECTIVE

Between 1990 and 2010, the U.S ranking in neonatal mortality slipped from 29(th) to 45(th) among countries globally. Substantial subnational variation in newborn mortality also exists. Our objective is to measure the extent to which trends and subnational variation in early neonatal mortality reflect differences in the prevalence of risk factors (gestational age and birth weight) compared to differences in clinical care.

METHODS

Observational study using linked birth and death data for all births in the United States between 1996 and 2006. We examined health service area (HSA) level variation in the expected early neonatal mortality rate, based on gestational age (GA) and birth-weight (BW), and GA-BW adjusted mortality as a proxy for clinical care. We analyzed the relationship between selected health system indicators and GA-BW-adjusted mortality.

RESULTS

The early neonatal death (ENND) rate declined 12% between 1996 and 2006 (2.39 to 2.10 per 1000 live births). This occurred despite increases in risk factor prevalence. There was significant HSA-level variation in the expected ENND rate (Rate Ratio: 0.73-1.47) and the GA-BW adjusted rate (Rate ratio: 0.63-1.68). Accounting for preterm volume (defined as <34 weeks), the number of neonatologist and NICU beds, 25.2% and 58.7% of the HSA-level variance in outcomes was explained among all births and very low birth weight babies, respectively.

CONCLUSION

Improvements in mortality could be realized through the expansion or reallocation of clinical neonatal resources, particularly in HSAs with a high volume of preterm births; however, prevention of preterm births and low-birth weight babies has a greater potential to improve newborn survival in the United States.

Roundtable on Urban Living Environment Research (RULER)

For 18 months in 2009-2010, the Rockefeller Foundation provided support to establish the Roundtable on Urban Living Environment Research (RULER). Composed of leading experts in population health measurement from a variety of disciplines, sectors, and continents, RULER met for the purpose of reviewing existing methods of measurement for urban health in the context of recent reports from UN agencies on health inequities in urban settings. The audience for this report was identified as international, national, and local governing bodies; civil society; and donor agencies. The goal of the report was to identify gaps in measurement that must be filled in order to assess and evaluate population health in urban settings, especially in informal settlements (or slums) in low- and middle-income countries. Care must be taken to integrate recommendations with existing platforms (e.g., Health Metrics Network, the Institute for Health Metrics and Evaluation) that could incorporate, mature, and sustain efforts to address these gaps and promote effective data for healthy urban management. RULER noted that these existing platforms focus primarily on health outcomes and systems, mainly at the national level. Although substantial reviews of health outcomes and health service measures had been conducted elsewhere, such reviews covered these in an aggregate and perhaps misleading way. For example, some spatial aspects of health inequities, such as those pointed to in the 2008 report from the WHO's Commission on the Social Determinants of Health, received limited attention. If RULER were to focus on health inequities in the urban environment, access to disaggregated data was a priority. RULER observed that some urban health metrics were already available, if not always appreciated and utilized in ongoing efforts (e.g., census data with granular data on households, water, and sanitation but with little attention paid to the spatial dimensions of these data). Other less obvious elements had not exploited the gains realized in spatial measurement technology and techniques (e.g., defining geographic and social urban informal settlement boundaries, classification of population-based amenities and hazards, and innovative spatial measurement of local governance for health). In summary, the RULER team identified three major areas for enhancing measurement to motivate action for urban health-namely, disaggregation of geographic areas for intra-urban risk assessment and action, measures for both social environment and governance, and measures for a better understanding of the implications of the physical (e.g., climate) and built environment for health. The challenge of addressing these elements in resource-poor settings was acknowledged, as was the intensely political nature of urban health metrics. The RULER team went further to identify existing global health metrics structures that could serve as platforms for more granular metrics specific for urban settings.