Deep Learning - Methods to Amplify Epidemiological Data Collection and Analyses.

Deep learning is a subfield of artificial intelligence and machine learning based mostly on neural networks and often combined with attention algorithms that has been used to detect and identify objects in text, audio, images, and video. Serghiou and Rough (Am J Epidemiol. 0000;000(00):0000-0000) present a primer for epidemiologists on deep learning models. These models provide substantial opportunities for epidemiologists to expand and amplify their research in both data collection and analyses by increasing the geographic reach of studies, including more research subjects, and working with large or high dimensional data. The tools for implementing deep learning methods are not quite yet as straightforward or ubiquitous for epidemiologists as traditional regression methods found in standard statistical software, but there are exciting opportunities for interdisciplinary collaboration with deep learning experts, just as epidemiologists have with statisticians, healthcare providers, urban planners, and other professionals. Despite the novelty of these methods, epidemiological principles of assessing bias, study design, interpretation and others still apply when implementing deep learning methods or assessing the findings of studies that have used them.

Statistical approaches for the integration of external controls in a cystic fibrosis clinical trial: a simulation and an application.

Development of new therapeutics for a rare disease such as cystic fibrosis (CF) is hindered by challenges in accruing enough patients for clinical trials. Using external controls from well-matched historical trials can reduce prospective trial sizes, and this approach has supported regulatory approval of new interventions for other rare diseases. We consider three statistical methods that incorporate external controls into a hypothetical clinical trial of a new treatment to reduce pulmonary exacerbations in CF patients: 1) inverse probability weighting, 2) Bayesian modeling with propensity score-based power priors, and 3) hierarchical Bayesian modeling with commensurate priors. We compare the methods via simulation study and in a real clinical trial data setting. Simulations showed that bias in the treatment effect was <4% using any of the methods, with type 1 error (or in the Bayesian cases, posterior probability of the null hypothesis) usually <5%. Inverse probability weighting was sensitive to similarity in prevalence of the covariates between historical and prospective trial populations. The commensurate prior method performed best with real clinical trial data. Using external controls to reduce trial size in future clinical trials holds promise and can advance the therapeutic pipeline for rare diseases.

Design of a Location-Based Case-Control Study of Built Environment Risk Factors for Pedestrian Fatalities in the U.S.

The location-based case-control design is a useful approach for studies where the exposures of interest are aspects of the environment around the location of a health event such as a pedestrian fatality. In this design locations are the unit of analysis and an enumerated cohort of locations are followed through time for the health events of interest and a case-control study of locations is nested within the cohort. Locations where events occurred (case-locations) are compared to matched locations where these events did not occur (control-locations). We describe the application of this design to the issue of pedestrian fatalities using a cohort of 9,612,698 intersections, 17,737,728 road segments, and 222,318 entrance/exit ramp segments that existed in 2017 across all 384 U.S. Metropolitan Statistical Areas. This cohort of locations was followed up from Jan 1, 2017 to Dec 31, 2018 for pedestrian fatalities using the National Highway Traffic Safety Administration Fatality Analysis Reporting System. In total, 10,587 fatalities were identified as having occurred on cohort locations and 21,174 matched control locations were selected using incidence density sampling. Geographic information systems, spatially linked administrative data sets and virtual neighborhood audits via Google Street View are underway to characterize study locations.

The Burden of Injuries Associated With E-Bikes, Powered Scooters, Hoverboards, and Bicycles in the United States: 2019‒2022.

Objectives. To describe the national burden of injuries associated with e-bikes, bicycles, hoverboards, and powered scooters (micromobility devices) in the United States. Methods. We compared patterns and trends for 1 933 296 estimated injuries associated with micromobility devices from 2019 to 2022 using National Electronic Injury Surveillance System data. Results. The population-based rates of e-bike and powered scooter injuries increased by 293.0% and 88.0%, respectively. When reported, powered scooter injuries had the highest proportion for alcohol use (9.0%) compared with other modes, whereas e-bike injuries had the highest proportion for motor vehicle involvement (35.4%). Internal injuries were more likely among e-bike diagnoses than hoverboard and bicycle (P < .05), but fractures and concussions were more likely among hoverboard diagnoses compared with all other devices (P < .05). When helmet use was identified in clinical notes (20.3%), helmet usage was higher among e-bike injuries (43.8%) compared with powered scooter (34.8%) and hoverboard (30.3%) injuries but lower compared with bicycle injuries (48.7%). Conclusions. The incidence of severe e-bike and powered scooter injuries increased over the 4-year period. Public health stakeholders should focus on improved surveillance and prevention of injuries associated with electric micromobility devices. (Am J Public Health. Published online ahead of print September 12, 2024:e1-e10. https://doi.org/10.2105/AJPH.2024.307820).

Road Traffic Injuries and the Built Environment in Bogotá, Colombia, 2015-2019: A Cross-Sectional Analysis.

Nine in 10 road traffic deaths occur in low- and middle-income countries (LMICs). Despite this disproportionate burden, few studies have examined built environment correlates of road traffic injury in these settings, including in Latin America. We examined road traffic collisions in Bogotá, Colombia, occurring between 2015 and 2019, and assessed the association between neighborhood-level built environment features and pedestrian injury and death. We used descriptive statistics to characterize all police-reported road traffic collisions that occurred in Bogotá between 2015 and 2019. Cluster detection was used to identify spatial clustering of pedestrian collisions. Adjusted multivariate Poisson regression models were fit to examine associations between several neighborhood-built environment features and rate of pedestrian road traffic injury and death. A total of 173,443 police-reported traffic collisions occurred in Bogotá between 2015 and 2019. Pedestrians made up about 25% of road traffic injuries and 50% of road traffic deaths in Bogotá between 2015 and 2019. Pedestrian collisions were spatially clustered in the southwestern region of Bogotá. Neighborhoods with more street trees (RR, 0.90; 95% CI, 0.82-0.98), traffic signals (0.89, 0.81-0.99), and bus stops (0.89, 0.82-0.97) were associated with lower pedestrian road traffic deaths. Neighborhoods with greater density of large roads were associated with higher pedestrian injury. Our findings highlight the potential for pedestrian-friendly infrastructure to promote safer interactions between pedestrians and motorists in Bogotá and in similar urban contexts globally.

Geospatial analysis of injury severity on major roads in Ghana (2017-2020): implications for targeted injury prevention and control initiatives.

BACKGROUND: Road safety authorities in high-income countries use geospatial motor vehicle collision data for planning hazard reduction and intervention targeting. However, low-income and middle-income countries (LMICs) rarely conduct such geospatial analyses due to a lack of data. Since 1991, Ghana has maintained a database of all collisions and is uniquely positioned to lead data-informed road injury prevention and control initiatives. METHODS: We identified and mapped geospatial patterns of hotspots of collisions, injuries, severe injuries and deaths using a well-known injury severity index with geographic information systems statistical methods (Getis-Ord Gi*). RESULTS: We identified specific areas (4.66% of major roads in urban areas and 6.16% of major roads in rural areas) to target injury control. Key roads, including National Road 1 (from the border of Cote D'Ivoire to the border of Togo) and National Road 6 (from Accra to Kumasi), have a significant concentration of high-risk roads. CONCLUSIONS: A few key road sections are critical to target for injury prevention. We conduct a collaborative geospatial study to demonstrate the importance of addressing data and research gaps in LMICs and call for similar future research on targeting injury control and prevention efforts.

Identifying emerging hot spots of road traffic injury severity using spatiotemporal methods: longitudinal analyses on major roads in Ghana from 2005 to 2020.

BACKGROUND: Although road traffic injuries and deaths have decreased globally, there is substantial national and sub-national heterogeneity, particularly in low- and middle-income countries (LMICs). Ghana is one of few countries in Africa collecting comprehensive, spatially detailed data on motor vehicle collisions (MVCs). This data is a critical step towards improving roadway safety, as accurate and reliable information is essential for devising targeted countermeasures. METHODS: Here, we analyze 16 years of police-report data using emerging hot spot analysis in ArcGIS to identify hot spots with trends of increasing injury severity (a weighted composite measure of MVCs, minor injuries, severe injuries, and deaths), and counts of injuries, severe injuries, and deaths along major roads in urban and rural areas of Ghana. RESULTS: We find injury severity index sums and minor injury counts are significantly decreasing over time in Ghana while severe injury and death counts are not, indicating the latter should be the focus for road safety efforts. We identify new, consecutive, intensifying, and persistent hot spots on 2.65% of urban roads and 4.37% of rural roads. Hot spots are intensifying in terms of severity and frequency on major roads in rural areas. CONCLUSIONS: A few key road sections, particularly in rural areas, show elevated levels of road traffic injury severity, warranting targeted interventions. Our method for evaluating spatiotemporal trends in MVC, road traffic injuries, and deaths in a LMIC includes sufficient detail for replication and adaptation in other countries, which is useful for targeting countermeasures and tracking progress.

Association between any underlying health condition and COVID-19-associated hospitalization by age group, Washington State, 2020-2021: a retrospective cohort study.

BACKGROUND: Presence of at least one underlying health condition (UHC) is positively associated with severe COVID-19, but there is limited research examining this association by age group, particularly among young adults. METHODS: We examined age-stratified associations between any UHC and COVID-19-associated hospitalization using a retrospective cohort study of electronic health record data from the University of Washington Medicine healthcare system for adult patients with a positive SARS-CoV-2 test from February 29, 2020, to March 13, 2021. Any UHC was defined as documented diagnosis of at least one UHC identified by the CDC as a potential risk factor for severe COVID-19. Adjusting for sex, age, race and ethnicity, and health insurance, we estimated risk ratios (aRRs) and risk differences (aRDs), overall and by age group (18-39, 40-64, and 65 + years). RESULTS: Among patients aged 18-39 (N = 3,249), 40-64 (N = 2,840), 65 + years (N = 1,363), and overall (N = 7,452), 57.5%, 79.4%, 89.4%, and 71.7% had at least one UHC, respectively. Overall, 4.4% of patients experienced COVID-19-associated hospitalization. For all age groups, the risk of COVID-19-associated hospitalization was greater for patients with any UHC vs. those without (18-39: 2.2% vs. 0.4%; 40-64: 5.6% vs. 0.3%; 65 + : 12.2% vs. 2.8%; overall: 5.9% vs. 0.6%). The aRR comparing patients with vs. those without UHCs was notably higher for patients aged 40-64 years (aRR [95% CI] for 18-39: 4.3 [1.8, 10.0]; 40-64: 12.9 [3.2, 52.5]; 65 + : 3.1 [1.2, 8.2]; overall: 5.3 [3.0, 9.6]). The aRDs increased across age groups (aRD [95% CI] per 1,000 SARS-CoV-2-positive persons for 18-39: 10 [2, 18]; 40-64: 43 [33, 54]; 65 + : 84 [51, 116]; overall: 28 [21, 35]). CONCLUSIONS: Individuals with UHCs are at significantly increased risk of COVID-19-associated hospitalization regardless of age. Our findings support the prevention of severe COVID-19 in adults with UHCs in all age groups and in older adults aged 65 + years as ongoing local public health priorities.

G-Computation and Agent-Based Modeling for Social Epidemiology: Can Population Interventions Prevent Posttraumatic Stress Disorder?

Agent-based modeling and g-computation can both be used to estimate impacts of intervening on complex systems. We explored each modeling approach within an applied example: interventions to reduce posttraumatic stress disorder (PTSD). We used data from a cohort of 2,282 adults representative of the adult population of the New York City metropolitan area from 2002-2006, of whom 16.3% developed PTSD over their lifetimes. We built 4 models: g-computation, an agent-based model (ABM) with no between-agent interactions, an ABM with violent-interaction dynamics, and an ABM with neighborhood dynamics. Three interventions were tested: 1) reducing violent victimization by 37.2% (real-world reduction); 2) reducing violent victimization by100%; and 3) supplementing the income of 20% of lower-income participants. The g-computation model estimated population-level PTSD risk reductions of 0.12% (95% confidence interval (CI): -0.16, 0.29), 0.28% (95% CI: -0.30, 0.70), and 1.55% (95% CI: 0.40, 2.12), respectively. The ABM with no interactions replicated the findings from g-computation. Introduction of interaction dynamics modestly decreased estimated intervention effects (income-supplement risk reduction dropped to 1.47%), whereas introduction of neighborhood dynamics modestly increased effectiveness (income-supplement risk reduction increased to 1.58%). Compared with g-computation, agent-based modeling permitted deeper exploration of complex systems dynamics at the cost of further assumptions.

Associations between neighborhood disinvestment and breast cancer outcomes within a populous state registry.

BACKGROUND: Breast cancer (BrCa) outcomes vary by social environmental factors, but the role of built-environment factors is understudied. The authors investigated associations between environmental physical disorder-indicators of residential disrepair and disinvestment-and BrCa tumor prognostic factors (stage at diagnosis, tumor grade, triple-negative [negative for estrogen receptor, progesterone receptor, and HER2 receptor] BrCa) and survival within a large state cancer registry linkage. METHODS: Data on sociodemographic, tumor, and vital status were derived from adult women who had invasive BrCa diagnosed from 2008 to 2017 ascertained from the New Jersey State Cancer Registry. Physical disorder was assessed through virtual neighborhood audits of 23,276 locations across New Jersey, and a personalized measure for the residential address of each woman with BrCa was estimated using universal kriging. Continuous covariates were z scored (mean ± standard deviation [SD], 0 ± 1) to reduce collinearity. Logistic regression models of tumor factors and accelerated failure time models of survival time to BrCa-specific death were built to investigate associations with physical disorder adjusted for covariates (with follow-up through 2019). RESULTS: There were 3637 BrCa-specific deaths among 40,963 women with a median follow-up of 5.3 years. In adjusted models, a 1-SD increase in physical disorder was associated with higher odds of late-stage BrCa (odds ratio, 1.09; 95% confidence interval, 1.02-1.15). Physical disorder was not associated with tumor grade or triple-negative tumors. A 1-SD increase in physical disorder was associated with a 10.5% shorter survival time (95% confidence interval, 6.1%-14.6%) only among women who had early stage BrCa. CONCLUSIONS: Physical disorder is associated with worse tumor prognostic factors and survival among women who have BrCa diagnosed at an early stage.

Differences in Weight Gain Following Residential Relocation in the Moving to Health (M2H) Study.

BACKGROUND: Neighborhoods may play an important role in shaping long-term weight trajectory and obesity risk. Studying the impact of moving to another neighborhood may be the most efficient way to determine the impact of the built environment on health. We explored whether residential moves were associated with changes in body weight. METHODS: Kaiser Permanente Washington electronic health records were used to identify 21,502 members aged 18-64 who moved within King County, WA between 2005 and 2017. We linked body weight measures to environment measures, including population, residential, and street intersection densities (800 m and 1,600 m Euclidian buffers) and access to supermarkets and fast foods (1,600 m and 5,000 m network distances). We used linear mixed models to estimate associations between postmove changes in environment and changes in body weight. RESULTS: In general, moving from high-density to moderate- or low-density neighborhoods was associated with greater weight gain postmove. For example, those moving from high to low residential density neighborhoods (within 1,600 m) gained an average of 4.5 (95% confidence interval [CI] = 3.0, 5.9) lbs 3 years after moving, whereas those moving from low to high-density neighborhoods gained an average of 1.3 (95% CI = -0.2, 2.9) lbs. Also, those moving from neighborhoods without fast-food access (within 1600m) to other neighborhoods without fast-food access gained less weight (average 1.6 lbs [95% CI = 0.9, 2.4]) than those moving from and to neighborhoods with fast-food access (average 2.8 lbs [95% CI = 2.5, 3.2]). CONCLUSIONS: Moving to higher-density neighborhoods may be associated with reductions in adult weight gain.

A Clinician's Guide to Conducting Research on Causal Effects.

Surgeons are uniquely poised to conduct research to improve patient care, yet a gap often exists between the clinician's desire to guide patient care with causal evidence and having adequate training necessary to produce causal evidence. This guide aims to address this gap by providing clinically relevant examples to illustrate necessary assumptions required for clinical research to produce causal estimates.

Wage theft and life expectancy inequities in the United States: A simulation study.

Wage theft - employers not paying workers their legally entitled wages and benefits - costs workers billions of dollars annually. We tested whether preventing wage theft could increase U.S. life expectancy and decrease inequities therein. We obtained nationally representative estimates of the 2001-2014 association between income and expected age at death for 40-year-olds (40 plus life expectancy at age 40) compiled from tax and Social Security Administration records, and estimates of the burden of wage theft from several sources, including estimates regarding minimum-wage violations (not paying workers the minimum wage) developed from Current Population Survey data. After modeling the relationship between income and expected age at death, we simulated the effects of scenarios preventing wage theft on mean expected age at death, assuming a causal effect of income on expected age at death. We simulated several scenarios, including one using data suggesting minimum-wage violations constituted 38% of all wage theft and caused 58% of affected workers' losses. Among women in the lowest income decile, mean expected age at death was 0.17 years longer in the counterfactual scenario than observed (95% confidence interval [CI]: 0.11-0.22), corresponding to 528,685 (95% CI: 346,018-711,353) years extended in the total 2001-2014 age-40 population. Among men in the lowest decile, the estimates were 0.12 (95% CI: 0.07-0.17) and 380,502 (95% CI: 229,630-531,374). Moreover, among women, mean expected age at death in the counterfactual scenario increased 0.16 (95% CI: 0.06-0.27) years more among the lowest decile than among the highest decile; among men, the estimate was 0.12 (95% CI: 0.03-0.21).

The Disclosure of Personally Identifiable Information in Studies of Neighborhood Contexts and Patient Outcomes.

Clinical epidemiology and patient-oriented health care research that incorporates neighborhood-level data is becoming increasingly common. A key step in conducting this research is converting patient address data to longitude and latitude data, a process known as geocoding. Several commonly used approaches to geocoding (eg, ggmap or the tidygeocoder R package) send patient addresses over the internet to web-based third-party geocoding services. Here, we describe how these approaches to geocoding disclose patients' personally identifiable information (PII) and how the subsequent publication of the research findings discloses the same patients' protected health information (PHI). We explain how these disclosures can occur and recommend strategies to maintain patient privacy when studying neighborhood effects on patient outcomes.

Earned Income Tax Credit and Youth Violence: Findings from the Youth Risk Behavior Surveillance System.

Family- and neighborhood-level poverty are associated with youth violence. Economic policies may address this risk factor by reducing parental stress and increasing opportunities. The federal Earned Income Tax Credit (EITC) is the largest cash transfer program in the US providing support to low-income working families. Many states have additional EITCs that vary in structure and generosity. To estimate the association between state EITC and youth violence, we conducted a repeated cross-sectional analysis using the variation in state EITC generosity over time by state and self-reported data in the Youth Risk Behavior Surveillance System (YRBSS) from 2005 to 2019. We estimated the association for all youth and then stratified by sex and race and ethnicity. A 10-percentage point greater state EITC was significantly associated with 3.8% lower prevalence of physical fighting among youth, overall (PR: 0.96; 95% CI 0.94-0.99), and for male students, 149 fewer (95% CI: -243, -55) students per 10,000 experiencing physical fighting. A 10-percentage point greater state EITC was significantly associated with 118 fewer (95% CI: -184, -52) White students per 10,000 experiencing physical fighting in the past 12 months while reductions among Black students (75 fewer; 95% CI: -176, 26) and Hispanic/Latino students (14 fewer; 95% CI: -93, 65) were not statistically significant. State EITC generosity was not significantly associated with measures of violence at school. Economic policies that increase financial security and provide financial resources may reduce the burden of youth violence; further attention to their differential benefits among specific population subgroups is warranted.

Predictive Value of 3 Clinical Criteria for Sepsis (Quick Sequential Organ Failure Assessment, Systemic Inflammatory Response Syndrome, and National Early Warning Score) With Respect to Short-term Mortality in Allogeneic Hematopoietic Cell Transplant Recipients With Suspected Infections.

BACKGROUND: Sepsis disproportionately affects allogeneic hematopoietic cell transplant (HCT) recipients and is challenging to define. Clinical criteria that predict mortality and intensive care unit end-points in patients with suspected infections (SIs) are used in sepsis definitions, but their predictive value among immunocompromised populations is largely unknown. Here, we evaluate 3 criteria among allogeneic HCT recipients with SIs. METHODS: We evaluated Systemic Inflammatory Response Syndrome (SIRS), quick Sequential Organ Failure Assessment (qSOFA), and National Early Warning Score (NEWS) in relation to short-term mortality among recipients transplanted between September 2010 and July 2017. We used cut-points of ≥ 2 for qSOFA/SIRS and ≥ 7 for NEWS and restricted to first SI per hospital encounter during patients' first 100 days posttransplant. RESULTS: Of the 880 recipients who experienced ≥ 1 SI, 58 (6.6%) died within 28 days and 22 (2.5%) within 10 days of an SI. In relation to 10-day mortality, SIRS was the most sensitive (91.3% [95% confidence interval {CI}, 72.0%-98.9%]) but least specific (35.0% [95% CI, 32.6%-37.5%]), whereas qSOFA was the most specific (90.5% [95% CI, 88.9%-91.9%]) but least sensitive (47.8% [95% CI, 26.8%-69.4%]). NEWS was moderately sensitive (78.3% [95% CI, 56.3%-92.5%]) and specific (70.2% [95% CI, 67.8%-72.4%]). CONCLUSIONS: NEWS outperformed qSOFA and SIRS, but each criterion had low to moderate predictive accuracy, and the magnitude of the known limitations of qSOFA and SIRS was at least as large as in the general population. Our data suggest that population-specific criteria are needed for immunocompromised patients.

Thirteen Questions About Using Machine Learning in Causal Research (You Won't Believe the Answer to Number 10!).

Machine learning is gaining prominence in the health sciences, where much of its use has focused on data-driven prediction. However, machine learning can also be embedded within causal analyses, potentially reducing biases arising from model misspecification. Using a question-and-answer format, we provide an introduction and orientation for epidemiologists interested in using machine learning but concerned about potential bias or loss of rigor due to use of "black box" models. We conclude with sample software code that may lower the barrier to entry to using these techniques.

Does the Union Make Us Strong? Labor-Union Membership, Self-Rated Health, and Mental Illness: A Parametric G-Formula Approach.

Union members enjoy better wages and benefits and greater power than nonmembers, which can improve health. However, the longitudinal union-health relationship remains uncertain, partially because of healthy-worker bias, which cannot be addressed without high-quality data and methods that account for exposure-confounder feedback and structural nonpositivity. Applying one such method, the parametric g-formula, to US-based Panel Study of Income Dynamics data, we analyzed the longitudinal relationships between union membership, poor/fair self-rated health (SRH), and moderate mental illness (Kessler 6-item score of ≥5). The SRH analyses included 16,719 respondents followed from 1985-2017, while the mental-illness analyses included 5,813 respondents followed from 2001-2017. Using the parametric g-formula, we contrasted cumulative incidence of the outcomes under 2 scenarios, one in which we set all employed-person-years to union-member employed-person-years (union scenario), and one in which we set no employed-person-years to union-member employed-person-years (nonunion scenario). We also examined whether the contrast varied by sex, sex and race, and sex and education. Overall, the union scenario was not associated with reduced incidence of poor/fair SRH (relative risk = 1.01, 95% confidence interval (CI): 0.95, 1.09; risk difference = 0.01, 95% CI: -0.03, 0.04) or moderate mental illness (relative risk = 1.02, 95% CI: 0.92, 1.12; risk difference = 0.01, 95% CI: -0.04, 0.06) relative to the nonunion scenario. These associations largely did not vary by subgroup.

Does the built environment have independent obesogenic power? Urban form and trajectories of weight gain.

OBJECTIVE: To determine whether selected features of the built environment can predict weight gain in a large longitudinal cohort of adults. METHODS: Weight trajectories over a 5-year period were obtained from electronic health records for 115,260 insured patients aged 18-64 years in the Kaiser Permanente Washington health care system. Home addresses were geocoded using ArcGIS. Built environment variables were population, residential unit, and road intersection densities captured using Euclidean-based SmartMaps at 800-m buffers. Counts of area supermarkets and fast food restaurants were obtained using network-based SmartMaps at 1600, and 5000-m buffers. Property values were a measure of socioeconomic status. Linear mixed effects models tested whether built environment variables at baseline were associated with long-term weight gain, adjusting for sex, age, race/ethnicity, Medicaid insurance, body weight, and residential property values. RESULTS: Built environment variables at baseline were associated with differences in baseline obesity prevalence and body mass index but had limited impact on weight trajectories. Mean weight gain for the full cohort was 0.06 kg at 1 year (95% CI: 0.03, 0.10); 0.64 kg at 3 years (95% CI: 0.59, 0.68), and 0.95 kg at 5 years (95% CI: 0.90, 1.00). In adjusted regression models, the top tertile of density metrics and frequency counts were associated with lower weight gain at 5-years follow-up compared to the bottom tertiles, though the mean differences in weight change for each follow-up year (1, 3, and 5) did not exceed 0.5 kg. CONCLUSIONS: Built environment variables that were associated with higher obesity prevalence at baseline had limited independent obesogenic power with respect to weight gain over time. Residential unit density had the strongest negative association with weight gain. Future work on the influence of built environment variables on health should also examine social context, including residential segregation and residential mobility.

Differential associations of the built environment on weight gain by sex and race/ethnicity but not age.

OBJECTIVE: To explore the built environment (BE) and weight change relationship by age, sex, and racial/ethnic subgroups in adults. METHODS: Weight trajectories were estimated using electronic health records for 115,260 insured Kaiser Permanente Washington members age 18-64 years. Member home addresses were geocoded using ArcGIS. Population, residential, and road intersection densities and counts of area supermarkets and fast food restaurants were measured with SmartMaps (800 and 5000-meter buffers) and categorized into tertiles. Linear mixed-effect models tested whether associations between BE features and weight gain at 1, 3, and 5 years differed by age, sex, and race/ethnicity, adjusting for demographics, baseline weight, and residential property values. RESULTS: Denser urban form and greater availability of supermarkets and fast food restaurants were associated with differential weight change across sex and race/ethnicity. At 5 years, the mean difference in weight change comparing the 3rd versus 1st tertile of residential density was significantly different between males (-0.49 kg, 95% CI: -0.68, -0.30) and females (-0.17 kg, 95% CI: -0.33, -0.01) (P-value for interaction = 0.011). Across race/ethnicity, the mean difference in weight change at 5 years for residential density was significantly different among non-Hispanic (NH) Whites (-0.47 kg, 95% CI: -0.61, -0.32), NH Blacks (-0.86 kg, 95% CI: -1.37, -0.36), Hispanics (0.10 kg, 95% CI: -0.46, 0.65), and NH Asians (0.44 kg, 95% CI: 0.10, 0.78) (P-value for interaction <0.001). These findings were consistent for other BE measures. CONCLUSION: The relationship between the built environment and weight change differs across demographic groups. Careful consideration of demographic differences in associations of BE and weight trajectories is warranted for investigating etiological mechanisms and guiding intervention development.