Background: There are limited efforts to incorporate different predisposing factors into prediction models that account for racial/ethnic plurality in exploring the spatial distribution of COVID-19 Severe Health Risk Index (SHRI). Therefore, this study examined the factors associated with COVID-19 SHRI in the United States.

Objective: This ecological study used publicly available datasets for 3107 United States counties. Racial/ethnic plurality was included as the primary predictor in this study. Analyses were carried out by constructing spatial and non-spatial regression models that adjusted for rurality, socio-demographic factors, perceived poor physical health, smoking, insufficient sleep, health insurance, primary care providers, intensive care units, preventable hospitalizations, and environmental risk factors.

Results: Findings indicate that spatial models can explain the geospatial disparities of COVID-19 SHRI. White, Hispanic, and other racial/ethnic plural counties showed a lower burden of COVID-19 SHRI when compared to plural black counties. Counties with older population, low income, high smoking, high reported insufficient sleep, and a high percentage of preventable hospitalizations had a high burden of COVID-19 SHRI. Counties with better health access and internet coverage had a low burden of COVID-19 SHRI.

Conclusion: This study helped identify county-level characteristics of risk populations to help guide resource allocation efforts. Racial/ethnic inequalities drive disparities concerning the higher burden of COVID-19 SHRI. Therefore, addressing these factors is essential to decrease inequalities in health outcomes. This work provided the baseline typology to further explore many social, health, economic, and political factors that contribute to different health outcomes.

Background: Mapping space-time patterns of COVID-19 is an important application of infectious disease surveillance, and few studies have assessed spatiotemporal patterns of COVID-19 at the census tract level.

Objective: The purpose of this study was to conduct a spatiotemporal analysis of COVID-19 in Clark County, Nevada, which contains the Las Vegas metropolitan area, between January 2020 and January 2023.

Methods: Incident COVID-19 cases from the Southern Nevada Health District were geocoded and aggregated by census tract and by month. Three-dimensional space-time cubes were constructed, where the calculated values in each bin (n=19,795) represented COVID-19 case counts at each location (i.e., census tract) across 37 time-steps (i.e., months). The Emerging Hotspot Analysis tool in ArcGIS Pro was used to analyze spatiotemporal COVID-19 trends by calculating the Getis-Ord Gi* statistic.

Results: Of the 535 census tracts in Clark County, 183 (34.2%) demonstrated statistically significant hot or cold spot trends in COVID-19 cases. Several statistically significant spatial clusters were identified, including a cluster of census tracts with intensifying hot spots in North Las Vegas, a cluster of persistent cold spots around the Las Vegas Strip, and a cluster of consecutive and sporadic hot spots in southwest Las Vegas.

Conclusion: Identifying census tracts that are consistently or increasingly susceptible to a higher COVID-19 burden over time can provide insight to local policymakers, health departments, and health care providers for prevention and control measures, resource allocation, and surveillance. Further analysis to determine factors contributing to hot and cold spots is warranted.

Background: The U.S. Census Bureau (“the Census”) provides standard geographies for data collection and analysis in public health research and practice. The Census periodically updates these geographies to reflect changes in underlying populations and political boundaries, potentially making data incomparable across time.

Methods: The City and Congressional Health Dashboards are public data websites that incorporate over 10 national data sources from 2012–2021, each utilizing multiple Census geographies. Through careful analyses, we have compiled our best practices for addressing Census changes across time.

Results: Census geography changes can include geographic identifier updates, boundary shifts, combining or splitting existing geographic units, and creating new geographic units. Frequency, magnitude, and type of change varies by the Census geography of interest.

When addressing geography change, we aim to harmonize the geographies to ensure a consistent underlying population across time. Some changes, such as geographic identifier updates, frequently do not impact a geography’s population. Boundary shifts can introduce population changes, but we often do not expect these changes to bias estimates. In some situations, however, such as large boundary shifts or combining/splitting geographic units, population comparability no longer exists in the longitudinal data. In these instances, we employ more complex geospatial aggregation methods to standardize data into a shared geography.

Because data sources incorporate Census geography changes on different timelines, we carefully review each data source to identify potential comparability issues and select harmonization strategies.

Conclusion: Careful accounting of Census geography changes can increase accuracy and comparability of longitudinal data.

The Developmental Origins of Health and Disease (DOHaD) theory postulates that exposure to adverse environmental factors in the early stages of life may have a significant impact short- and long-term health outcomes ^1–3. Indeed, evidence from the current research indicates that early-life environmental influences may act as a protective factor or increase susceptibility to the risk of chronic diseases, including cardiovascular disease, metabolic diseases, psychiatric diseases, and linear growth in children ^2,4,5. While the studies have enhanced our understanding of developmental plasticity – the ability of an organism to develop in various ways, depending on the particular environment or setting 2 – there is limited knowledge of spatial variations in the risk burden associated with exposure to environmental factors in early life. Our study explores possible variations in the associations between environmental exposures and childhood health outcomes, specifically acute respiratory infection and anemia. The study adopts a big data approach using individual-level demographic and health data from the Demographic and Health Survey program and environmental data from NASA’s Goddard Earth Sciences Data and Information Services Center (GES DISC). We employ Bayesian spatially varying coefficient modeling with the Stochastic Partial Differential Equations (SPDE) approach to investigate the variability of effects of ambient air pollution (specifically PM2.5, nitrogen dioxide, sulfur dioxide, ozone, carbon monoxide, organic carbon, and black carbon) on acute respiratory infection and anemia.

Background: Wildfires devastated California (CA) during 2020–2021, disrupting millions of residents’ lives and access to care. Did the wildfires impact provider appointment availability for patients? Or did other factors (e. g., social determinants of health [SDoH] and rurality) have a greater influence on access? This study explores whether these factors are associated with providers not meeting appointment availability standards.

Methods: All CA HMO (Health Maintenance Organization; group managed-care plans) providers assigned to participate in the CA regulator’s Provider Appointment Availability Survey (PAAS) were included. We created (i) county-level bubble charts of rurality vs. appointment availability, stratified by SDoH and wildfire occurrence; (ii) maps of 4-mile wildfire buffer, Hot Spots (Getis Ord Gi* statistic) of SDoH, rurality, and providers (stratified by telehealth, specialty); and (iii) temporal visualization stratified by provider group. Analyses and visualizations were created with R, python, Figma, and ArcGIS.

Results: SDoH and rurality of CA HMO providers appear to have had a greater influence on whether providers met appointment access standards relative to wildfires, and these variables also influenced patient use of telehealth. Wildfires, while varied, had less impact on provider appointment availability. Additionally, we identified high-performing provider groups meeting PAAS standards despite adverse wildfire, SDoH, and rurality impact.

Conclusion: Study results highlight the multi-factor impacts on patient access. The outcomes help guide appropriate interventions, including educating providers and groups on best practices, benchmarking and evaluating provider performance to other providers in their county, and reinforcing provider discussions with patients about virtual care options.

The health of the Hispanic population is an important topic research that will continue to be relevant as the population grows and disperses across the country.

This paper aims to explore the community context and its implications for obesity prevalence in places that contain a sizeable concentration of the Hispanic population. Also, to examine obesity in areas with Hispanic population growth that are population hubs compared to new destinations. As such, the following questions are asked: How do differences in the proportion of Hispanics in US counties impact rates of obesity over space? How do the rates of obesity vary between areas that are new destinations for Hispanics compared to those that have been longstanding Hispanic population hubs?

We hypothesize that counties with the highest proportion of Hispanics will have a higher concentration of obesity compared to those where Hispanics are a smaller section of the population. We also predict that areas experiencing new waves of Hispanic migration will have lower rates of obesity compared to historic Hispanic population destinations.

Data from CDC PLACES, the American Community Survey (ACS), and the USDA will be used. Relevant sociodemographic characteristics will be included. The outcome of interest is the county-level prevalence of obesity. The indicators of interest will be the county-level proportion of the population made up of Hispanics and differentiation of counties as a historic population hub or a new destination. An exploratory spatial data analysis will be conducted, followed by a spatial regression model.

Background

Studies examining shifts in treatment utilization over time for proximal humerus fracture (PHF) have documented 570-1,841% increases in reverse total shoulder arthroplasty (RTSA) use over the past decade with associated decreases in hemiarthroplasty (HA). However, diffusion of medical innovations is influenced by numerous patient, provider, and health system factors leading to geographic variation in adoption rates. The objective of this study is to assess local area characteristics associated with high adoption of initial RTSA utilization for Medicare beneficiaries with PHF.

Methods

This study will utilize complete Medicare datasets capturing all fee-for-service encounters with a diagnosis of PHF in 2011 and 2017. I will calculate risk-adjusted area treatment ratios (ATR) for all patients receiving RTSA at the hospital referral region (HRR) level in each year. I will then assess regional factors independently associated with ATR for RTSA in both 2011 and 2017 using linear regression models, including hospital competition, provider supply, and academic presence as predictors.

Results

Analysis for this project is not yet complete; however, it is expected to be completed prior to the annual meeting date. The expected sample size is approximately 75,000 patients in each year based on prior literature utilizing the project data.

Conclusion

This study will increase our understanding of regional factors associated with increased RTSA utilization for PHF and may support the need for clinical decision support tools and policy initiatives to increase equity in treatment access for PHF.

Individual-level health and well-being data permits analysis of community health and well-being and health risk evaluations across several dimensions. It also enables rankings of reported health and well-being for large geographical areas such as states, metropolitan areas and counties. However, there is large variation in reported well-being within such large areas. In this paper, we address this limitation by modelling well-being data to generate zip-code level rankings through spatially-informed statistical modeling. We use a graph Laplacian matrix that enables us to estimate the zip-code level effect on well-being using individual level data as well as by borrowing information from neighboring zip-codes. We conduct simulation studies to show that if spatial patterns exist at the zip-code level, our model is able to successfully capture these spatial patterns. We deploy our model on well-being data for the state of Massachusetts, where an overall well-being index (WBI) and scores from five subscales (Physical, Financial, Social, Community, Purpose) are used as outcome variables. We model WBI and the subscales using individual-level demographic characteristics as predictors while including a zip-code-level spatial effect. We find that our model can capture the effects of demographic features, while also offering spatial effect estimates for all zip-codes, even when there is low (or zero) response rate in certain zip-codes. These spatial effect estimates provide community health and well-being rankings of zip-codes and our method can be deployed more generally.

Hypertension is a major risk factor for cardiovascular diseases, stroke, and death. Our aim was to predict sustained, uncontrolled HTN (SUH) (blood pressure (BP) ≥ 140/90 mm Hg on multiple encounters) and hypertensive crisis (HC) (BP ≥ 180/120 mm Hg) within 1 year of an elevated BP reading (≥ 140/90 mm Hg). The cohort consisted of 142,897 patients with an elevated BP reading at an outpatient facility of a major Charlotte-based health system during 2018. The dataset included 53 EHR-based 1-year-look-back predictors and was randomly split into training (80%) and validation (20%) sets. Penalized logistic regression (LR), gradient boosting, multilayer perceptron, and random forest models were hyperparameter-tuned using 5-fold cross-validation on the training set. For both outcomes, the cross-validation c-statistics of the LR model and other models were comparable. Thus, given the LR model’s relative simplicity, it was selected for further validation. On the validation set, point estimates and 95% confidence intervals for the c-statistic, calibration intercept, and calibration slope are 0.708 (0.702 – 0.715), 0.024 (-0.004 – 0.052), 1.001 (0.966 – 1.035) for predicting SUH, and 0.798 (0.784 – 0.813), 0.228 (0.070 – 0.387), 1.086 (1.029 – 1.142) for predicting HC. Calibration plots show the models are properly calibrated. We demonstrate the models’ clinical utility by plotting net benefit against plausible decision thresholds. In summary, the models achieved satisfactory/good discrimination, accurate risk estimates, and clinical usefulness in internal validation. Once externally validated, these models can be integrated within the EHR to support clinical decision-making and population health outreach interventions.

Background

Clinical trials often have limited participation of pregnant populations. Consequently, post-marketing safety surveillance is crucial to evaluate the safety of medical products in pregnancy. The FDA BEST Initiative conducts post-marketing safety surveillance of biologics, including vaccines, using administrative claims and electronic health record databases. To assess safety outcomes following vaccine use in pregnancy, records of mothers and infants must be linked in these databases.

Methods

Mother-infant linkage was established in three commercial claims databases and time periods: CVS Health/Aetna, 2018–2022; Optum, 2020–2023; and Carelon Research 2015–2022. Pregnancies were identified in females 12–55 years using a validated algorithm. Both liveborn deliveries and infants were identified one year after the start date through the most recent available data of each database. Mothers and infants were linked using insurance subscriber ID among those with delivery and birth dates within three days of each other. We determined the proportion of deliveries that linked to an infant and infants with a known subscriber ID that linked to a delivery.

Results

The data contained 2,225,992 deliveries and 2,293,323 infants with a known subscriber ID. Among identified deliveries and infants, 73.4% of deliveries linked to at least one infant (linkage range by database: 70.3%–81.0%) and 72.8% of infants linked to a delivery (linkage range by database: 71.4%–73.7%).

Conclusions

Most mother-infant records were linked successfully within three BEST commercial claims databases using only insurance subscriber ID, among those with delivery and infant birth dates within three days.