Combining Food Frequency and Survey Data to Quantify Long-Term Dietary Exposure: A Methyl Mercury Case Study

Twenty-four-hour recall data from the Continuing Survey of Food Intake by Individuals (CSFII) are frequently used to estimate dietary exposure for risk assessment. Food frequency questionnaires are traditional instruments of epidemiological research; however, their application in dietary exposure and risk assessment has been limited. This article presents a probabilistic method of bridging the National Health and Nutrition Examination Survey (NHANES) food frequency and the CSFII data to estimate longitudinal (usual) intake, using a case study of seafood mercury exposures for two population subgroups (females 16 to 49 years and children 1 to 5 years). Two hundred forty-nine CSFII food codes were mapped into 28 NHANES fish/shellfish categories. FDA and state/local seafood mercury data were used. A uniform distribution with minimum and maximum blood-diet ratios of 0.66 to 1.07 was assumed. A probabilistic assessment was conducted to estimate distributions of individual 30-day average daily fish/shellfish intakes, methyl mercury exposure, and blood levels. The upper percentile estimates of fish and shellfish intakes based on the 30-day daily averages were lower than those based on two- and three-day daily averages. These results support previous findings that distributions of "usual" intakes based on a small number of consumption days provide overestimates in the upper percentiles. About 10% of the females (16 to 49 years) and children (1 to 5 years) may be exposed to mercury levels above the EPA's RfD. The predicted 75th and 90th percentile blood mercury levels for the females in the 16-to-49-year group were similar to those reported by NHANES. The predicted 90th percentile blood mercury levels for children in the 1-to-5-year subgroup was similar to NHANES and the 75th percentile estimates were slightly above the NHANES.     

Estimating prediction error for complex samples

With a growing interest in using non-representative samples to train prediction models for numerous outcomes it is necessary to account for the sampling design that gives rise to the data in order to assess the generalized predictive utility of a proposed prediction rule. After learning a prediction rule based on a non-uniform sample, it is of interest to estimate the rule's error rate when applied to unobserved members of the population. Efron (1986) proposed a general class of covariance penalty inflated prediction error estimators that assume the available training data are representative of the target population for which the prediction rule is to be applied. We extend Efron's estimator to the complex sample context by incorporating Horvitz–Thompson sampling weights and show that it is consistent for the true generalization error rate when applied to the underlying superpopulation. The resulting Horvitz–Thompson–Efron estimator is equivalent to dAIC, a recent extension of Akaike's information criteria to survey sampling data, but is more widely applicable. The proposed methodology is assessed with simulations and is applied to models predicting renal function obtained from the large-scale National Health and Nutrition Examination Study survey. The Canadian Journal of Statistics 48: 204–221; 2020 © 2019 Statistical Society of Canada     

Tract- and county-level income inequality and individual risk of obesity in the United States

ObjectivesWe tested three alternative hypotheses regarding the relationship between income inequality and individual risk of obesity at two geographical scales: U.S. Census tract and county.MethodsIncome inequality was measured by Gini coefficients, created from the 2000 U.S. Census. Obesity was clinically measured in the 2003–2008 National Health and Nutrition Examination Survey (NHANES). The individual measures and area measures were geo-linked to estimate three sets of multi-level models: tract only, county only, and tract and county simultaneously. Gender was tested as a moderator.ResultsAt both the tract and county levels, higher income inequality was associated with lower individual risk of obesity. The size of the coefficient was larger for county-level Gini than for tract-level Gini; and controlling income inequality at one level did not reduce the impact of income inequality at the other level. Gender was not a significant moderator for the obesity-income inequality association.ConclusionsHigher tract and county income inequality was associated with lower individual risk of obesity, indicating that at least at the tract and county levels and in the context of cross-sectional data, the public health goal of reducing the rate of obesity is in line with anti-poverty policies of addressing poverty through mixed-income development where neighborhood income inequality is likely higher than homogeneous neighborhoods.     

A pooled Bayes test of independence for sparse contingency tables from small areas

We study the association between bone mineral density (BMD) and body mass index (BMI) when contingency tables are constructed from the several U.S. counties, where BMD has three levels (normal, osteopenia and osteoporosis) and BMI has four levels (underweight, normal, overweight and obese). We use the Bayes factor (posterior odds divided by prior odds or equivalently the ratio of the marginal likelihoods) to construct the new test. Like the chi-squared test and Fisher's exact test, we have a direct Bayes test which is a standard test using data from each county. In our main contribution, for each county techniques of small area estimation are used to borrow strength across counties and a pooled test of independence of BMD and BMI is obtained using a hierarchical Bayesian model. Our pooled Bayes test is computed by performing a Monte Carlo integration using random samples rather than Gibbs samples. We have seen important differences among the pooled Bayes test, direct Bayes test and the Cressie-Read test that allows for some degree of sparseness, when the degree of evidence against independence is studied. As expected, we also found that the direct Bayes test is sensitive to the prior specifications but the pooled Bayes test is not so sensitive. Moreover, the pooled Bayes test has competitive power properties, and it is superior when the cell counts are small to moderate.     

Body Weight Distributions for Risk Assessment

Precise age-specific average body weight estimates are necessary for deterministic risk assessments, and an accurate body weight distribution is equally important in probabilistic risk assessments. Age-specific body weight distributions for U.S. residents are estimated using NHANES (National Health and Nutrition Examination Survey) data collected in four surveys over the last 24 years. The weighted mean and standard deviation of natural log-transformed body weights are computed for single-year age groups and population age-specific weight patterns further described using piece-wise polynomial spline functions and nonparametric age-smoothed trend lines. These functions are used to compare distributional changes in age-specific body weight in the United States from the first NHANES survey in 1976-1980 to the most recent in 1999-2002. Analysis demonstrates that age- and sex-specific average body weight changes over this time period are not uniform. Use of these functions to compute body weight distributions for selected child-age categories is demonstrated.     

Modeling multiple‐response categorical data from complex surveys

Although “choose all that apply” questions are common in modern surveys, methods for analyzing associations among responses to such questions have only recently been developed. These methods are generally valid only for simple random sampling, but these types of questions often appear in surveys conducted under more complex sampling plans. The purpose of this article is to provide statistical analysis methods that can be applied to “choose all that apply” questions in complex survey sampling situations. Loglinear models are developed to incorporate the multiple responses inherent in these types of questions. Statistics to compare models and to measure association are proposed and their asymptotic distributions are derived. Monte Carlo simulations show that tests based on adjusted Pearson statistics generally hold their correct size when comparing models. These simulations also show that confidence intervals for odds ratios estimated from loglinear models have good coverage properties, while being shorter than those constructed using empirical estimates. Furthermore, the methods are shown to be applicable to more general problems of modeling associations between elements of two or more binary vectors. The proposed analysis methods are applied to data from the National Health and Nutrition Examination Survey. The Canadian Journal of Statistics © 2009 Statistical Society of Canada     

Analysis of cardiovascular risk factors associated with serum testosterone levels according to the US 2011–2012 National Health and Nutrition Examination Survey

Objective: To investigate associations between cardiovascular disease risk factors, including fasting glucose, cholesterol, high density lipoprotein cholesterol (HDL-c), LDL-c, blood pressure, body mass index (BMI), C-peptide, creatinine kinase, smoking, alcohol use, physical activity, C-reactive protein as well as homocysteine levels and cardiovascular events.

Methods: Data from 1545 men aged ≥40?years, with testosterone deficiency (TD) (<300?ng/dL) and non-TD (≥300?ng/dL) which were extracted from the National Health and Nutrition Examination Survey database 2011–2012 and analyzed.

Results: Multivariate logistic regression analysis showed positive associations between TD and BMI (≥35 vs.?p?=?.016), HDL-c (<0.91 vs. ≥0.91: OR?=?1.60, 95% CI: 1.14–2.24, p?=?.006) and diabetes (diabetes vs. non-diabetes: OR?=?1.48, 95% CI: 1.14–1.92, p?=?.004) as well as negative associations between TD and metabolic equivalent scores (≥12 vs. <12: OR?=?0.69, 95% CI: 0.52–0.91, p?=?.009) and smoking (Ever vs. never: OR?=?0.69, 95% CI: 0.51–0.94, p?=?.018). Furthermore, total serum testosterone levels were lower in patients with heart failure (p?=?.04) and angina/angina pectoris (p?=?.001) compared with subjects without these cardiac problems.

Conclusion: Low serum testosterone was associated with multiple risk factors for CHD.     

Predicting objective physical activity from self-report surveys: a model validation study using estimated generalized least-squares regression

Physical activity measurements derived from self-report surveys are prone to measurement errors. Monitoring devices like accelerometers offer more objective measurements of physical activity, but are impractical for use in large-scale surveys. A model capable of predicting objective measurements of physical activity from self-reports would offer a practical alternative to obtaining measurements directly from monitoring devices. Using data from National Health and Nutrition Examination Survey 2003–2006, we developed and validated models for predicting objective physical activity from self-report variables and other demographic characteristics. The prediction intervals produced by the models were large, suggesting that the ability to predict objective physical activity for individuals from self-reports is limited.