Department of Biostatistics
Environmental Statistics Seminar
2014 - 2015
ABSTRACT: Human capital, a determining factor in individual labor market and macroeconomic outcomes, is malleable to early-life investments and insults. This study examines the long-term human capital impacts of early-life exposure to criteria air pollutants in the developing economy context of Metropolitan Cebu, Philippines. A three-decade, longitudinal survey containing frequent measures of human capital is combined with macro- and micro-environmental databases characterizing exposure to carbon monoxide and ozone. An instrumental variable strategy corrects the bias from unobserved heterogeneity and measurement error. Findings indicate that height and cognition - correlates and measures of human capital - are negatively affected by increased early-life exposures. Impacts to labor market outcomes, including hours worked and earnings, vary by gender and labor sector. Carbon monoxide exposure is consistently detrimental to both height and cognition while the effects of ozone exposure grow over time and are highly detrimental to cognition and earnings. In present value terms, a nationwide 10% policy reduction in carbon monoxide and ozone levels would annually generate approximately $5.15 billion in discounted lifetime earnings per annual birth cohort.
ABSTRACT: Exposure to air pollutants adversely affects human health, but the full scope of this impact is unknown. Studies to date have largely examined the magnitude of air pollution's effect on a set of pre-specified health conditions, rather than investigating a wide spectrum of conditions making no a priory assumptions. We aim to identify the full range of reasons for hospitalization in the older US population associated with short-term exposure to fine-particulate matter (PM2.5) air pollution, while accounting for temporal trends in hospital admission rates and PM2.5 levels.
ABSTRACT: Rational priority setting in global health requires rigorous quantification of worldwide, population-level trends in health status. Because global-level surveys are not available, researchers are forced to rely on country-level and local data that are often sparse, fragmentary, or unreliable. We present a Bayesian model that addresses this problem by systematically combining data from disparate sources to make country-level estimates of trends in important health metrics for all nations. The model uses Markov random field methods to allow for nonlinear trends and a hierarchical structure to borrow strength within and across regional country clusters. MCMC sampling facilitates inference in a high-dimensional, constrained parameter space, while providing posterior draws that enable straightforward inference on the wide variety of functionals of interest. Throughout, the Bayesian approach accounts for uncertainty resulting from data missingness, as well as sampling and parameter uncertainty.
In this talk, I will present results for two example health metrics. First, I will discuss trends in hypertension, a primary risk factor for cardiovascular disease – the leading cause of death worldwide. I will then turn to malnutrition, an important contributor to childhood morbidity and mortality in low-income regions. As all levels of mild, moderate, and severe malnutrition are of clinical and public health importance, I will present an extension of the model that uses a finite normal mixture to estimate the shape of the distributions for markers of malnutrition. The model incorporates both individual-level data when available, as well as aggregated summary statistics from studies for which individual-level data could not be obtained.
This work addresses three important problems that often arise in the fields of public health surveillance and global health monitoring. First, data are always incomplete. Second, data from different sources are often incomparable. Third, standard techniques fail to provide estimates of the full distributions of health metrics, the tails of which are often of substantive interest.
ABSTRACT: Evaluating environmental health risks in communities requires models characterizing geographic and demographic patterns of exposure to multiple stressors. These exposure models can be constructed from multivariable regression analyses using individual-level predictors (microdata), but these microdata are not typically available with sufficient geographic resolution given privacy concerns. In this study, we developed synthetic geographically-resolved microdata for a low-income community (New Bedford, Massachusetts) facing multiple environmental stressors, and we used these microdata to predict smoking behavior and other stressor exposures at high geographic and demographic resolution. Our simulation approach can be used to predict high-resolution patterns of multiple exposures and vulnerability attributes in community settings.
ABSTRACT: Fine particulate matter (PM2.5) is composed of many sources of pollution, each potentially varying at different spatial scales. An unanswered question in the air pollution literature is at which spatial scales PM2.5 is most harmful. Identifying whether harmful effects are driven by local, finely varying pollution or from regional, slowly varying sources is important to future air pollution regulations. We propose a two-dimensional wavelet decomposition that alleviates restrictive assumptions about the spatial surface required for standard Wavelet decompositions. Using this method we can decompose the surface of PM2.5 to identify which sources of pollution are driving adverse health outcomes. We apply our method to a study of birth weights in Massachusetts and find that both local and regional sources of pollution negatively impact birth weight, though the effect is greater for local sources.
ABSTRACT: Exposures to heavy metal mixtures (ie. lead, manganese, zinc, etc.) may significantly impact neurodevelopment in early life. Due to sequential neurodevelopmental processes, there may be certain time windows of susceptibility during which vulnerability to metal mixtures is increased; studies have shown effect modification of one metal's exposure in the presence of other metals, and a possibility for detectable mixture effects on health at low doses of exposure below individual no observable adverse effect levels. We present a Bayesian penalization framework that identifies time windows of susceptibility in the context of exposure to metal mixtures. Simulations demonstrate the ability of the method to detect time-varying nonlinear and quadratic effects, and the method is applied to a study of heavy metal mixture exposure in children.
ABSTRACT: In clinical settings often the scientific question of interest is isolating the effect of a primary treatment. Other exposures such as: subject demographics, related biologic mechanisms, and epidemiological factors that are associated with the exposure and outcome may act as confounders (clouding or completely inverting the true effect of the primary exposure). The biggest barrier typically is the decision of which covariates out of a high-dimensional set covariates to include, since the true confounders are unknown.
We propose a 2-Step Bayesian Model Averaging (2-Step BMA) technique that targets the primary exposure of interest, characterizing the treatment effect while controlling for a high dimensional set of unknown confounders using propensity scores. This method improves on existing methods by averaging over the entire model space of both the exposure and outcome models to control for cofounding while targeting treatment effect without making assumptions about the underlying model and without need of an arbitrary number of confounders to include a priori.
ABSTRACT: None Given
ABSTRACT: The Seychelles Child Development Study (SCDS) examines the effects of prenatal exposure to methylmercury on the functioning of the central nervous system. The SCDS data include 20 outcomes measured on 9-year old children that can be classified broadly in four outcome classes or "domains": cognition, memory, motor, and social behavior. We first consider the problem of estimating the effect of exposure on multiple outcomes in a single model when each outcome belongs to one domain and domain memberships are known. We then extend this to the situation in which the outcomes may belong to more than one domain and where we also want to learn about the assignment of outcomes to domains. In this talk I will discuss the Seychelles data, motivate and give results of the models with the aid of pictures, and briefly compare these models to a structural equation model.
ABSTRACT: Epidemiologic studies suggest maternal ambient air pollution exposure during critical periods of pregnancy is associated with adverse effects on fetal development. In this work, we introduce new methodology for identifying critical periods of development during post-conception gestational weeks 2-8 where elevated exposure to particulate matter less than 2.5 micrometers (PM2.5) adversely impacts development of the heart. A majority of past studies have focused on highly aggregated temporal levels of exposure during the pregnancy and have failed to account for anatomical similarities between the considered congenital heart defects (CHDs). We introduce a multinomial probit model in the Bayesian setting that allows for joint identification of susceptible daily periods during pregnancy for 12 types of CHDs with respect to maternal PM2.5 exposure. We apply the model to a dataset of mothers from the National Birth Defect Prevention Study where daily PM2.5 exposures from post-conception gestational weeks 2-8 are assigned using predictions from the downscaler pollution model. This approach is compared to an aggregated exposure model that defines exposure as the average value over post-conception gestational weeks 2-8. Results suggest an association between increased PM2.5 exposure on post-conception gestational day 53 with the development of pulmonary valve stenosis and exposures during days 50-51 with tetralogy of Fallot. The aggregated exposure model fails to identify any significant associations. Simulation study results suggest that the findings are generally robust to multiple sources of error. The general form of the introduced model allows for different exposures and health outcomes to be considered in future applications.
|Back to SPH Biostatistics||
Maintained by the
Last Update: April 24, 2015