Predicting nationwide obesity from food sales using machine learning.

The obesity epidemic progresses everywhere across the globe, and implementing frequent nationwide surveys to measure the percentage of obese population is costly. Conversely, country-level food sales information can be accessed inexpensively through different suppliers on a regular basis. This study applies a methodology to predict obesity prevalence at the country-level based on national sales of a small subset of food and beverage categories. Three machine learning algorithms for nonlinear regression were implemented using purchase and obesity prevalence data from 79 countries: support vector machines, random forests and extreme gradient boosting. The proposed method was validated in terms of both the absolute prediction error and the proportion of countries for which the obesity prevalence was predicted satisfactorily. We found that the most-relevant food category to predict obesity is baked goods and flours, followed by cheese and carbonated drinks.

Associations of multidimensional socioeconomic and built environment factors with body mass index trajectories among youth in geographically heterogeneous communities.

Understanding contextual influences on obesity requires comparison of heterogeneous communities and concurrent assessment of multiple contextual domains. We used a theoretically-based measurement model to assess multidimensional socioeconomic and built environment factors theorized to influence childhood obesity across a diverse geography ranging from rural to urban. Confirmatory factor analysis specified four factors-community socioeconomic deprivation (CSED), food outlet abundance (FOOD), fitness and recreational assets (FIT), and utilitarian physical activity favorability (UTIL)-which were assigned to communities (townships, boroughs, city census tracts) in 37 Pennsylvania counties. Using electronic health records from 2001 to 2012 from 163,820 youth aged 3-18 years from 1288 communities, we conducted multilevel linear regression analyses with factor quartiles and their cross products with age, age2, and age3 to test whether community factors impacted body mass index (BMI) growth trajectories. Models controlled for sex, age, race/ethnicity, and Medical Assistance. Factor scores were lowest in townships, indicating less deprivation, fewer food and physical activity outlets, and lower utilitarian physical activity favorability. BMI at average age was lower in townships versus boroughs (beta [SE]) (0.217 [0.027], P < 0.001) and cities (0.378 [0.036], P < 0.001), as was BMI growth over time. Factor distributions across community types lacked overlap, requiring stratified analyses to avoid extrapolation. In townships, FOOD, UTIL, and FIT were inversely associated with BMI trajectories. Across community types, youth in the lowest (versus higher) CSED quartiles had lower BMI at average age and slower BMI growth, signifying the importance of community deprivation to the obesogenicity of environments.

Neighborhood social and economic change and diabetes incidence: The HeartHealthyHoods study.

We studied the association between neighborhood social and economic change and type 2 diabetes incidence in the city of Madrid (Spain). We followed 199,621 individuals living in 393 census tracts for diabetes incidence for 6 years using electronic health records, starting in 2009. We measured neighborhood social and economic change from 2005 to 2009 using a finite mixture model with 16 indicators that resulted in four types of neighborhood change. Adjusted results showed an association between neighborhood change and diabetes incidence: compared to those living in Aging/Stable areas, people living in Declining SES, New Housing and Improving SES areas have an 8% (HR = 0.92, 95% CI 0.87 to 0.99), 9% (HR = 0.91, 95% CI 0.81 to 1.01) and 11% (HR = 0.89, 95% CI 0.81 to 0.98) decrease in diabetes incidence. This evidence can help guide policies for diabetes prevention by focusing efforts on specific urban areas.

Socioeconomic Status, Race/Ethnicity, and Diurnal Cortisol Trajectories in Middle-Aged and Older Adults.

Objectives: Slow afternoon cortisol decline may be a marker of aging. We hypothesize that lower socioeconomic status (SES) and African American race are associated with lower waking cortisol and slower afternoon decline. Method: Six salivary cortisol samples, collected within a 24-hr period from 566 cohort participants aged 56-78 years, were examined in random-effects models. SES measures included socioeconomic vulnerability (household income and assets <500% of poverty) and education (≥college, some college, and ≤high school). African Americans were compared with all others. Results: Adjusting for age and sex, intermediate, but not low, education was associated with approximately 17% lower average waking cortisol and 1% slower decline, compared with high education. Socioeconomic vulnerability was not associated with waking cortisol or linear decline. Accounting for African American race/ethnicity, socioeconomic vulnerability was associated with a 3% faster decline, and education was not associated with cortisol. African Americans had 26% lower average waking cortisol and 1% slower decline than others. Discussion: African American race/ethnicity, but not lower SES, was associated with lower waking cortisol and slower afternoon decline in middle-aged and older adults. This pattern is likely a marker of earlier biological aging in vulnerable groups. Race/ethnicity may compete with SES as a measure of cumulative vulnerability.

Associations of LEP, CRH, ICAM-1, and LINE-1 methylation, measured in saliva, with waist circumference, body mass index, and percent body fat in mid-childhood.

BACKGROUND: Genetics explains a small proportion of variance in body mass index at the population level. Epigenetics, commonly measured by gene methylation, holds promise for understanding obesity risk factors and mechanisms. METHODS: Participants were 431 adolescents aged 10-15 years. BMI z-score, waist circumference z-score, and percent body fat were measured. Saliva samples were collected and methylation of promoter regions of four candidate genes or sequences (LEP, ICAM-1, CRH, and LINE-1) were measured in 3-4 CpG sites each. Linear regression was used to identify associations of methylation with obesity-related outcomes. RESULTS: After adjusting for age, in sex-stratified analysis, the three obesity-related outcomes were negatively associated with LEP methylation in obese boys only. There were no associations of methylation of the other genes or sequences and the obesity-related outcomes. CONCLUSIONS: Our results are consistent with prior studies that reported sex differences in associations of obesity-related outcomes with LEP methylation, and also as would be expected in adipose tissue, the source of circulating leptin. The findings suggest that saliva might be an acceptable tissue for epigenetics studies in adolescents.

Economic growth and mortality: do social protection policies matter?

Background: In the 20th century, periods of macroeconomic growth have been associated with increases in population mortality. Factors that cause or mitigate this association are not well understood. Evidence suggests that social policy may buffer the deleterious impact of economic growth. We sought to explore associations between changing unemployment (as a proxy for economic change) and trends in mortality over 30 years in the context of varying social protection expenditures. Methods: We model change in all-cause mortality in 21 OECD (Organization for Economic Cooperation and Development) countries from 1980 to 2010. Data from the Comparative Welfare States Data Set and the WHO Mortality Database were used. A decrease in the unemployment rate was used as a proxy for economic growth and age-adjusted mortality rates as the outcome. Social protection expenditure was measured as percentage of gross domestic product expended. Results: A 1% decrease in unemployment (i.e. the proxy for economic growth) was associated with a 0.24% increase in the overall mortality rate (95% confidence interval: 0.07;0.42) in countries with no changes in social protection. Reductions in social protection expenditure strengthened this association between unemployment and mortality. The magnitude of the association was diminished over time. Conclusions: Our results are consistent with the hypothesis that social protection policies that accompany economic growth can mitigate its potential deleterious effects on health. Further research should identify specific policies that are most effective.

Community Audit of Social, Civil, and Activity Domains in Diverse Environments (CASCADDE).

There are currently no direct observation environmental audit tools that measure diverse aspects of the obesity-related environment efficiently and reliably in a variety of geographic settings. The goal was to develop a new instrument to reliably characterize the overall properties and features of rural, suburban, and urban settings along multiple dimensions. The Community Audit of Social, Civil, and Activity Domains in Diverse Environments (CASCADDE) is an iPad-based instrument that incorporates GPS coordinates and photography and comprises 214 items yielding seven summary indices. A comprehensive spatial sampling strategy, training manual, and supporting data analysis code were also developed. Random geospatial sampling using GIS was used to capture features of the community as a whole. A single auditor collected 510 observation points in 30 communities (2013-2015). This analysis was done in 2015-2016. Correlation coefficients were used to compare items and indices to each other and to standard measures. Multilevel unconditional means models were used to calculate intraclass correlation coefficients to determine if there was significant variation between communities. Results suggest that CASCADDE measures aspects of communities not previously captured by secondary data sources. Additionally, seven summary indices capture meaningful differences between communities based on 15 observations per community. Community audit tools such as CASCADDE complement secondary data sources and have the potential to offer new insights about the mechanisms through which communities affect obesity and other health outcomes.

Associations of prenatal and childhood antibiotic use with child body mass index at age 3 years.

OBJECTIVE: Early-life antibiotic exposure, whether through prenatal or childhood antibiotic use, may contribute to increased child body mass. Associations of prenatal and childhood antibiotic use with body mass index z-score (BMIz) were evaluated at age 3 years. METHODS: Electronic health records were utilized from 8,793 mothers and singleton children delivered at Geisinger Clinic in Pennsylvania between 2006 and 2012. Antibiotic orders were ascertained for mothers during pregnancy and for children through their age-3 BMI measurement. Linear mixed-effects regression models evaluated associations of prenatal and childhood antibiotic use with child BMIz. RESULTS: Prenatal antibiotic orders were not associated with child BMIz. Children in the three largest categories of lifetime antibiotic orders had higher BMIz compared with children with no orders; associations persisted when controlling for prenatal antibiotics (ß [95% confidence interval]) (4-5 child orders: 0.090 [0.011 to 0.170]; 6 to 8: 0.113 [0.029 to 0.197]; ≥9: 0.175 [0.088 to 0.263]; trend P value <0.001). Two or more first-year orders were also associated with BMIz (1: 0.021 [-0.038 to 0.081]; 2: 0.088 [0.017 to 0.160]; ≥3: 0.104 [0.038 to 0.170]; trend P value < 0.001). CONCLUSIONS: Associations of early-life and lifetime childhood antibiotic use with increased child BMI highlight antibiotic exposure as a modifiable factor for reducing population-level excess weight.

Home Food Rules in Relation to Youth Eating Behaviors, Body Mass Index, Waist Circumference, and Percent Body Fat.

PURPOSE: To investigate agreement and associations between parent and youth acknowledgment of home food rules, youth eating behaviors, and measures of body composition and excess weight. METHODS: Parent-youth dyads (N = 413) completed the "rules for eating at home" scale (Active Where Survey) and reported dietary intake. Trained research staff obtained anthropometric data. Linear regression analyses separately evaluated relationships between youth and parent acknowledgment of rules and youth-reported eating behaviors and anthropometric outcomes. Food rules were evaluated as a 12-item scale and individually. RESULTS: Score on the food rule scale was positively associated with fruit and vegetable servings by youth acknowledgment only (ß = .09, p = .006), and not with anthropometric outcomes. The rule "no desserts except fruit" was positively associated with fruit and vegetable servings by youth (ß = .72, p = .002) and parent (ß = .53, p = .03) acknowledgment. The rules "no second helpings at meals" and "limited fast food" were positively associated with body mass index z-score by youth (ß = .38, p = .002; ß = .32, p = .02, respectively) and parent (ß = .74, p < .001; ß = .41, p = .006, respectively) acknowledgment, with similar results for waist circumference z-score and percent body fat. CONCLUSIONS: Inverse associations between specific food rules and healthful eating behaviors but positive associations with anthropometric outcomes suggest potentially bidirectional relationships between food rule implementation and youth weight. Future studies should disentangle how food rules guide youth behavior in the context of youth weight status.

Comparing methods of targeting obesity interventions in populations: An agent-based simulation.

Social networks as well as neighborhood environments have been shown to effect obesity-related behaviors including energy intake and physical activity. Accordingly, harnessing social networks to improve targeting of obesity interventions may be promising to the extent this leads to social multiplier effects and wider diffusion of intervention impact on populations. However, the literature evaluating network-based interventions has been inconsistent. Computational methods like agent-based models (ABM) provide researchers with tools to experiment in a simulated environment. We develop an ABM to compare conventional targeting methods (random selection, based on individual obesity risk, and vulnerable areas) with network-based targeting methods. We adapt a previously published and validated model of network diffusion of obesity-related behavior. We then build social networks among agents using a more realistic approach. We calibrate our model first against national-level data. Our results show that network-based targeting may lead to greater population impact. We also present a new targeting method that outperforms other methods in terms of intervention effectiveness at the population level.

Depression, its comorbidities and treatment, and childhood body mass index trajectories.

OBJECTIVE: No prior studies have evaluated depression diagnoses and cumulative antidepressant use in relation to longitudinal body mass index (BMI) trajectories in a population-representative sample. METHODS: Electronic health record data from 105,163 children ages 8 to 18 years with 314,648 BMI values were used. Depression diagnoses were evaluated as ever versus never, cumulative number of encounters with diagnoses, and total duration of diagnoses. Antidepressants were evaluated as months of use. Associations were evaluated with diagnoses alone, antidepressants alone, and then together, adjusting for covariates. RESULTS: A total of 6,172 (5.9%) and 10,628 (10.1%) children had a diagnosis of depression or received antidepressant treatment, respectively. At all ages, children receiving Medical Assistance (30.9%) were more likely to be treated with antidepressants. Depression diagnosis and antidepressant use were each independently and positively associated with BMI trajectories; associations were stronger with longer durations of diagnosis and treatment. Among children who received 12 or more months of antidepressants (vs. none), the mean (95% CI) weight gain at 18 years associated with antidepressant use (all classes) was 2.10 (1.76-2.45) kg. CONCLUSIONS: Depression and antidepressant use were both independently associated with increasing BMIs over time, suggesting an important unintended consequence of healthcare to the obesity epidemic.

Are neighborhoods causal? Complications arising from the 'stickiness' of ZNA.

Are neighborhoods causal? The answer remains elusive. Armed with new multilevel methods, enthusiasm for neighborhoods research surged at the turn of the century. However, a wave of skepticism has arisen based on the difficulty of drawing causal inferences from observational studies in which selection to neighborhoods is non-random. Researchers have sought answers from experimental and quasi-experimental studies of movers vs. stayers. We develop two related concepts in this essay in the hopes of shedding light on this problem. First, the inceptive environment into which persons are born (which we term ZNA for Zip code Nativity Area) exerts a potentially powerful causal impact on health. Detecting that causal effect is challenging for reasons similar that obtain in other fields (including genetics). Second, we explicate the problem of neighborhood 'stickiness' in terms of the persistence of neighborhood treatment assignment, and argue that under-appreciation of stickiness has led to systematic bias in causal estimates of neighborhoods proportional to the degree of stickiness. In sticky contexts, failure to account for the lasting influences of ZNA by adjusting for intermediate individual socioeconomic and health variables on the causal pathway can result in neighborhood effects estimates that are biased toward the null. We follow with an example drawn from evidence of neighborhood 'stickiness' and obesity. The stickiness of ZNA cautions us that experimental evidence may be insufficient or misleading as a solution to causal inference problems in neighborhood research.

Exploring the forest instead of the trees: An innovative method for defining obesogenic and obesoprotective environments.

Past research has assessed the association of single community characteristics with obesity, ignoring the spatial co-occurrence of multiple community-level risk factors. We used conditional random forests (CRF), a non-parametric machine learning approach to identify the combination of community features that are most important for the prediction of obesogenic and obesoprotective environments for children. After examining 44 community characteristics, we identified 13 features of the social, food, and physical activity environment that in combination correctly classified 67% of communities as obesoprotective or obesogenic using mean BMI-z as a surrogate. Social environment characteristics emerged as most important classifiers and might provide leverage for intervention. CRF allows consideration of the neighborhood as a system of risk factors.

Household and neighborhood conditions partially account for associations between education and physical capacity in the National Health and Aging Trends Study.

Socioeconomic resources, such as education, prevent disability but are not readily modifiable. We tested the hypothesis that household and neighborhood conditions, which may be modifiable, partially account for associations between education and physical capacity in a population-based sample of older adults. The National Health and Aging Trends Study measured education (

Community socioeconomic deprivation and obesity trajectories in children using electronic health records.

OBJECTIVES: Longitudinal studies of the role of community context in childhood obesity are lacking. The objective of this study was to examine associations of community socio economic deprivation (CSD) with trajectories of change in body mass index (BMI) in childhood and adolescence. METHODS: Data came from electronic health records on 163,473 children aged 3-18 residing in 1,288 communities in Pennsylvania whose weight and height were measured longitudinally. CSD at the year of birth was measured using six US Census variables and modeled in quartiles. Trajectories of BMI within CSD quartiles were estimated using random effects growth-curve models accounting for differences by age, sex, and race/ethnicity as well as correcting for non-constant residual variance across age groups. RESULTS: CSD was associated with higher BMI at average age (10.7 years) and with more rapid growth of BMI over time. Children born in communities with greater CSD had steeper increases of BMI at younger ages. Those born into the poorest communities displayed sustained accelerated BMI growth. CSD remained associated with BMI trajectories after adjustment for a measure of household socio economic deprivation. CONCLUSIONS: Higher CSD may be associated with more obesogenic growth trajectories in early life. Findings suggest that individual-level interventions that ignore the effect of community context on obesity-related behaviors may be less efficient.

Modeling US adult obesity trends: a system dynamics model for estimating energy imbalance gap.

OBJECTIVES: We present a system dynamics model that quantifies the energy imbalance gap responsible for the US adult obesity epidemic among gender and racial subpopulations. METHODS: We divided the adult population into gender-race/ethnicity subpopulations and body mass index (BMI) classes. We defined transition rates between classes as a function of metabolic dynamics of individuals within each class. We estimated energy intake in each BMI class within the past 4 decades as a multiplication of the equilibrium energy intake of individuals in that class. Through calibration, we estimated the energy gap multiplier for each gender-race-BMI group by matching simulated BMI distributions for each subpopulation against national data with maximum likelihood estimation. RESULTS: No subpopulation showed a negative or zero energy gap, suggesting that the obesity epidemic continues to worsen, albeit at a slower rate. In the past decade the epidemic has slowed for non-Hispanic Whites, is starting to slow for non-Hispanic Blacks, but continues to accelerate among Mexican Americans. CONCLUSIONS: The differential energy balance gap across subpopulations and over time suggests that interventions should be tailored to subpopulations' needs.

Attention deficit disorder, stimulant use, and childhood body mass index trajectory.

BACKGROUND: Childhood attention-deficit/hyperactivity disorder (ADHD) has been associated with childhood and adult obesity, and stimulant use with delayed childhood growth, but the independent influences are unclear. No longitudinal studies have examined associations of ADHD diagnosis and stimulant use on BMI trajectories throughout childhood and adolescence. METHODS: We used longitudinal electronic health record data from the Geisinger Health System on 163,820 children ages 3 to 18 years in Pennsylvania. Random effects linear regression models were used to model BMI trajectories with increasing age in relation to ADHD diagnosis, age at first stimulant use, and stimulant use duration, while controlling for confounding variables. RESULTS: Mean (SD) age at first BMI was 8.9 (5.0) years, and children provided a mean (SD) of 3.2 (2.4) annual BMI measurements. On average, BMI trajectories showed a curvilinear relation with age. There were consistent associations of unmedicated ADHD with higher BMIs during childhood compared with those without ADHD or stimulants. Younger age at first stimulant use and longer duration of stimulant use were each associated with slower BMI growth earlier in childhood but a more rapid rebound to higher BMIs in late adolescence. CONCLUSIONS: The study provides the first longitudinal evidence that ADHD during childhood not treated with stimulants was associated with higher childhood BMIs. In contrast, ADHD treated with stimulants was associated with slower early BMI growth but a rebound later in adolescence to levels above children without a history of ADHD or stimulant use. The findings have important clinical and neurobiological implications.

Neighborhood disadvantage in context: the influence of urbanicity on the association between neighborhood disadvantage and adolescent emotional disorders.

PURPOSE: Inconsistent evidence of a relationship between neighborhood disadvantage and adolescent mental health may be, in part, attributable to heterogeneity based on urban or rural residence. Using the largest nationally representative survey of US adolescent mental health available, we estimated the association between neighborhood disadvantage and adolescent emotional disorders and the extent to which urbanicity modified this association. METHODS: The National Comorbidity Survey Replication Adolescent Supplement (NCS-A) sampled adolescents aged 13-17 years (N = 10,123). Households were geocoded to Census tracts. Using a propensity score approach that addresses bias from non-random selection of individuals into neighborhoods, logistic regression models were used to estimate the relative odds of having a DSM-IV emotional disorder (any past-year anxiety disorder, major depressive disorder or dysthymia) comparing similar adolescents living in disadvantaged versus non-disadvantaged neighborhoods in urban center, urban fringe, and non-urban areas. RESULTS: The association between neighborhood disadvantage and emotional disorder was more than twice as large for adolescents living in urban centers versus non-urban areas. In urban centers, living in a disadvantaged neighborhood was associated with 59% (95% confidence interval 25-103) increased adjusted odds of emotional disorder. CONCLUSIONS: Urbanicity modifies the relationship between neighborhood disadvantage and emotional disorder in adolescents. This effect modification may explain why evidence of a relationship between neighborhood disadvantage and adolescent mental health has been inconsistent. Recognizing the joint influence of neighborhood socioeconomic context and urbanicity may improve specificity in identifying relevant neighborhood processes.

Normalization and extraction of interpretable metrics from raw accelerometry data.

We introduce an explicit set of metrics for human activity based on high-density acceleration recordings from a hip-worn tri-axial accelerometer. These metrics are based on two concepts: (i) Time Active, a measure of the length of time when activity is distinguishable from rest and (ii) AI, a measure of the relative amplitude of activity relative to rest. All measurements are normalized (have the same interpretation across subjects and days), easy to explain and implement, and reproducible across platforms and software implementations. Metrics were validated by visual inspection of results and quantitative in-lab replication studies, and by an association study with health outcomes.

The association between cortisol and neighborhood disadvantage in a U.S. population-based sample of adolescents.

The association between neighborhood conditions and cortisol is rarely studied in children or adolescents and has been hampered by small sample size and racial/ethnic and geographic homogeneity. Our objective was to estimate the association between neighborhood disadvantage and salivary cortisol levels in a large, geographically and racially/ethnically diverse sample of adolescents from the National Comorbidity Survey Replication Adolescent Supplement. Salivary cortisol was collected before and after an interview administered in the adolescent's home. We used a propensity score approach to match adolescents living in disadvantaged neighborhoods with those in non-disadvantaged neighborhoods to create two similar groups based on the time and day of cortisol collection as well as demographic characteristics. Adolescents living in disadvantaged neighborhoods had higher pre-interview cortisol levels and steeper rates of decline in cortisol levels over the course of the interview than similar adolescents in non-disadvantaged neighborhoods. This bolsters the evidence base suggesting that place may influence the stress response system.