This study examined arsenic and mercury concentrations in channel catfish (Ictalurus punctatus) caught for human consumption in the Allegheny River. Arsenic is a known human carcinogen and mercury is known to cause neurological disorders, particularly in fetuses and children. Subsistence and semi-subsistence anglers and their families are at risk of exposure. Catfish were caught at 4 distinct sites Pittsburgh, Cheswick, Freeport and Ford City. They were measured for general characteristics such as weight, length, and sex, and tissue samples were taken and analyzed for heavy metal content. The study addressed main questions: Do levels of mercury and arsenic vary among the 4 sites and, if so, how? Do the levels of mercury and arsenic in these fish pose a threat to people who eat them regularly? Analysis of variance was used to determine group differences by location. Contrasts were performed to test for specific differences: Pittsburgh from the other three sites, and Cheswick, Freeport and Ford City from each other. Multiple regression analyses were conducted to determine if any of the other factors, weight, length or sex, had an impact on metals levels in addition to location. Assessments of risk to human consumers of these fish were conducted using US EPA guidelines and formulae. The Pittsburgh fish were found to have significantly different concentrations of both arsenic and mercury than the fish from the other sites. Mean levels of arsenic and mercury were observed to be lower in the Pittsburgh fish. No significant differences in contaminant levels were found between the Cheswick, Freeport and Ford City fish. Subsequent analyses were conducted combining these three locations into the Allegheny River group. Regression analyses showed minimal impact of weight and no impact of any other factor when controlling for location. Public Health Implications: Risk assessments found hazard quotients above 1 for all populations (children 3-8, children 9-15, women of childbearing age, other adults) based on 95% confidence intervals for mean concentrations of mercury. Arsenic levels also showed excess cancer risk for all populations. Current fish consumption advisories are inadequate to protect the health of regular consumers of these fish.

The correlation between any two random variables can be estimated using a variety of techniques including parametric methods based on the Pearson correlation coefficient, nonparametric methods, and regression analysis. While these estimators have been widely used, the computation of these estimates in the presence of missing data has not been as widely studied. There has been some work addressing the estimation of parameters in the presence of missing data for regression analysis; including imputation, inverse probability weighted regression and weighted estimating equations. However, there has been little work focused on the estimation of the correlation coefficient. To assess the usefulness of these methods in a practical setting, we present simulation studies comparing imputation, inverse probability weighting and complete cases and provide recommendations on the basis of these results. Furthermore, computation of Partial Least Squares (PLS) scores with the correlation matrix computed using the above mentioned techniques are also presented. We apply these results in a positron emission tomography data set consisting of several different brain regions as response variables and cognitive tasks as covariates of interest. Alzheimer's disease is a progressive and fatal health disease. The application presented in this work is significant for public health since it provides us with a better understanding of variability in different brain regions as it relates to neuropsychological tests that are helpful in diagnosis of progressive brain disease (i.e Alzheimer's disease).

Children with prenatal alcohol exposure (PAE) tend to show higher rates of conduct disorder (CD), even after the effect of some potentially confounding factors, including parental alcoholism, parental drug abuse, and externalizing disorder, have been taken into account. It is clear that some subgroups of CD may show distinct developmental pathways; for instance, the use of construct of psychopath for subtyping CD children has grown and some research has highlighted a distinction between callous-unemotional traits and highly-impulsive traits. As more and more studies have examined the relationship between PAE and the occurrence of CD, some important questions have been raised. The objective of this study is to determine whether PAE is associated with a specific subtype of CD, or if it is equally associated with both highly impulsive and the callous-unemotional forms of diagnosis. The National Institute of Mental Health Diagnostic Interview Schedule for Children- 4th Edition (DISC-IV) was used to assess the psychiatric disorders and symptoms of 572 children with PAE. Among these 572 children, 67 met the criteria for lifetime diagnosis of CD. We collapsed these children into three groups based on the levels of PAE (unexposed, lightly exposed, heavy exposed). The analyses were conducted to examine the difference of each CD symptoms and clinical information of children. The results suggest that while most of the CD symptoms and clinical information were similar among three groups, the differences of both domains of social impairment and psychiatric treatment in the twelve months preceding the diagnostic interview were statistically significant. Based on the outcome of the analyses, 1ST trimester PAE is associated with an observable increase in the incidence of both callous-unemotional and highly-impulsive subtypes of children with CD, rather than being associated with one or the other of these two subtypes. We would conclude that the CD children with PAE or non-PAE show a similar range of clinical symptoms and subtypes. For public health significance, this might be helpful information for clinicians and public health officials when they discuss the diagnoses or issues about children with PAE. This information may also assist researchers to build an individual and comprehensive intervention for different subtypes of conduct disorder in children.

Objective: To develop a systematic statistical approach to estimate accrual closure date in large scale multi-center clinical trials or large public health studies. It is relevant to the research in public health. Background: In a typical multi-center cancer clinical trial or large public health study, sample size is predetermined to achieve desired power and study participants are enrolled from hundreds of satellite sites. As the accrual is closing to the target size, the coordinating data center needs to project an accrual closure date based on observed accrual pattern and notify participating sites several weeks in advance. In the past, projections were simply based on some crude assessment and conservative measures were incorporated in order to achieve the target accrual size. The resulted excessive accrual size usually leads to unnecessary budget increase considering that the coordinating center needs to pay thousands of dollars for each accrued participant. Method: For multi-center clinical trials, there is very small probability for a site to accrue a patient during a short period and mostly the accrual from different sites is mostly independent from each other. Therefore, the overall accrual could be modeled by a Poisson process. Based on accrual data collected up to a time point, a Poisson process-based method was used to analyze the past accrual pattern. Combining with assumption on the future accruing pattern, two methods were proposed here to predict the accrual closure date. The estimates and their confidence intervals were used to guide clinical practice. The proposed methods were illustrated through analysis of accrual data from NSABP trials B-38 and C-08.

Dengue is an important mosquito-borne viral disease. In Taiwan, there are hundreds to thousands dengue cases each year, and dengue is considered one of the most important public health issues. The objective of this study was to use geographical information systems (GIS) methodology to map and analyze the spatial and temporal distribution of dengue in Taiwan during 2004 to 2007 and to elucidate the association of geographical and climatic risk factors with dengue incidence. Dengue annually occurs starting in summer, peaking in fall and goes down in winter. The spatial distribution: Spatial autocorrelation of dengue was measured using Morans I at the global level and LISA at the local level. The global spatial autocorrelation analysis revealed a significantly positive spatial autocorrelation of dengue for 2004 to 2007, with Morans I=0.171, p-value=0.03. The local spatial autocorrelation analysis showed a significantly high dengue incidence around Tainan county and Kaohsiung county (p-value<0.05), which are located in the southern Taiwan. Based on the geographical features, dengue tended to occur in the southwestern cities/counties in Taiwan with plains and rivers spread. Temperature had a positive relationship with dengue incidence in summer and fall (rs=0.74 and p-value=0.002 in summer, rs=0.53 and p-value=0.003 in fall). Rainfall had a positive relationship with dengue incidence in summer (rs=0.61 and p-value=0.017). However, there was no significant correlation between temperature or rainfall and dengue incidence in winter. The public health importance of this study: Disease maps have been playing a key descriptive role in public health and epidemiology. By this study, areas of the current geographical distribution of the incidence of dengue in Taiwan were identified. Through spatial autocorrelation analyses, the identification of unusual concentration of dengue in Tainan county and Kaohsiung county has been defined. This could prompt health agencies and the government to take a critical look at these risk areas, and make appropriate health planning and resource allocation.

Tree-structured survival analysis (TSSA) is used to recursively detect covariate values that best divide the sample into subsequent subsets with respect to a time to event outcome. The result is a set of empirical classification groups, each of which identifies individuals with more homogeneous risk than the original sample. We propose methods for managing missing covariate data and also for incorporating temporal features of repeatedly measured covariates into TSSA. First, for missing covariate data, we propose an algorithm that uses a stochastic process to add draws to an existing single tree-structured imputation method. Secondly, to incorporate temporal features of repeatedly measured covariates, we propose two different methods: (1) use a two-stage random effects polynomial model to estimate temporal features of repeatedly measured covariates to be used as TSSA predictor variables, and (2) incorporate other types of functions of repeatedly measured covariates into existing time-dependent TSSA methodology. We conduct simulation studies to assess the accuracy and predictive abilities of our proposed methodology. Our methodology has particular public health importance because we create, interpret and assess TSSA algorithms that can be used in a clinical setting to predict response to treatment for late-life depression.

Adaptive treatment regime is a set of rules that governs the assignment of time-varying treatment based on observed covariates and intermediate response. Treatment choices are made sequentially as patients make transition from one health state to another. Specifically, in two stage randomization designs, patients are randomized to one of the initial treatments, and at the end of the first stage, they are randomized to one of the second stage treatments depending on the outcome of the initial treatment. The goal is to find the best treatment regime which produces the best terminal outcome. For time-to-event data, the best outcome is the longest survival time, and for longitudinal data, the best outcome is greatest reduction (or increase) in some scores such as reduction 24-item Hamilton Rating Scale of Depression (HRSD24) score. For time-to-event data, we propose a weighted Kaplan-Meier estimator based on the method of inverse-probability weighting and compare its properties to that of the standard Kaplan-Meier estimator, and two other existing methods such as marginal mean model based estimator and weighted risk set estimator. For longitudinal data, outcome such as HRSD24 scores are collected repeatedly to monitor the progress of the subject. We propose three methods incorporating inverse probability weighting, mixed models, multiple imputations, and pattern mixture models to assess the effect of treatment regimes on the longitudinal HRSD24 scores. Methods are compared through simulation studies with an application to a depression study. Assessing the effect of treatment regimes on longitudinally observed outcome data is important in Public Health since clinicians will be able to identify effective treatment regimes for treating chronic diseases. Proposed statistical methods provide useful tools for unbiased estimation of the effects of treatment regimes from sequentially randomized designs. Availability of these methods will help advance the research in AIDS, cancer, depression, hepatitis and other disease areas.

Generalized additive model (GAM) with natural cubic splines (NS) has been commonly used as a standard analytical tool in time series studies of health effects of air pollution. Standard model selection procedures used in GAM ignore the uncertainty in model fitting. This may lead to biased estimates of the health effects, in particular lagged effects. Moreover, the degrees of smoothing to adjust for time-varying confounders estimated from data-driven methods were found to give biased estimates. We applied Bayesian model averaging (BMA) approach to account for model uncertainty and proposed also a generalized linear mixed model with natural cubic splines (GLMM + NS) to adjust for time-varying confounders. As the posterior model probability derived from BMA contains a hyperparameter to account for model uncertainty and has potential usefulness in this type of studies, we first conducted a sensitivity analysis with simulation studies for BMA with different calibrated hyperparameters. Our results indicated the importance of selecting the optimum degree of lagging for variables, not based on only maximizing the likelihood, but by considering the possible effects of lagging and biological plausibility. Our proposed model, GLMM + NS, was found to produce more precise estimates of the health effects of current day level of PM10 than the commonly used generalized linear models with natural cubic splines (GLM + NS) in our simulation studies. However, more in depth analyses with special attention to inferential procedures in readily available software are needed to have any definitive conclusion about the performance of our proposed model. An illustrative example is provided using data from the Allegheny County Air Pollution Study (ACAPS) where the quantity of interest was the relative risk of cardiopulmonary hospital admissions for a 20 μg⁄m³ increase in PM10 values for the current day and five previous days. Assessing the effect of air pollution on human health is an important public health problem. There are some inconsistencies in the literature as to the magnitude of this effect. The proposed statistical methods are expected to better characterize the true effect of air pollution.

Genome-wide association (GWA) studies are used widely for detecting gene variants contribution to diseases and traits. Recent researches indicate to several methodological challenges in the study design for GWA, for example, sample size issues, power calculations, false positive rate adjustments, and commercial chips coverage of the genome. Chromosomal regions can also influence the observed genetic diversity under certain conditions; mainly the regions of secondary structures and large-scale repeats may affect the fidelity in marker genotyping. This study was to find such regions that contained markers with more variability and to examine the correlation of this variability to the factors relevant in a GWA study design, such as the false positive rate. We enrolled healthy controls from eight independent GWA designs then assigned randomly into case and control status. Minor allele frequency estimates, and case-control association analyses were performed using PLINK for sets with different sample sizes. Marker numbers exhibiting high variability in the allele frequency estimates, and the average number of false positives were calculated for bins across the autosomal genome. We found that SNP variability (in allele frequency) was unrelated to the sample size. More variable regions correlated to regions of more average number of false positives, after adjusting for confounders, such as sample size. We suggested that regions with more variability might have structural characteristics that made them difficult to be scanned during the genotyping process. Our study has great public health relevance because regions with more variability could undermine the effective study of a candidate genes and disease relationship during a research, or worse leading to erroneous conclusions. We advise in studying these regions, the researchers could lower their false positive rates to avoid inaccurately significant levels.

The case-control study design is commonly used in genetic association study with a binary trait using unrelated individuals from a population. To test association with a binary trait in a case-control or cohort study, the standard analysis is a chi-square test or logistic regression model that test to detect a difference in frequencies of alleles or genotypes. In this thesis, we derive the maximum likelihood estimator, using Chen and Chatterjees methods, for standard 1 df genetic tests (dominant, recessive, and multiplicative). We then compare these methods that assume HWE with standard Wald tests and chi-squared tests that do not make the HWE assumption. We consider four different HWE scenarios: 1) when HWE holds in both cases and controls, 2) when HWE does not hold in cases and controls follow HWE, 3) when HWE does not hold in controls, and cases follow HWE and 4) when HWE does not hold in either cases or controls. Our results show that the performances of the three statistics (chi-squared, Wald, and Chen and Chatterjee Wald) are equivalent for multiplicative test under all four HWE scenarios. When HWE holds in both cases and controls, the performances of the three statistics are also equivalent, except for variations attributable to type I error issues. When HWE fails to hold in either cases or controls or both, the 2 df version of the Chan and Chatterjee Wald test (and to a lesser extent the dominant and recessive versions) detects this HWE departure and can therefore "find" a case-control difference even if there is not an allele frequency difference or even a genotype frequency difference. Our results will improve the design and analysis of genetic association studies. Such association studies are a crucial step in understanding the genetic components of many diseases that have a large impact on public health. Better understanding of the etiology of these diseases will lead in the long term to better prevention and treatment strategies.