Environmental health tracking: epidemiologic study of respiratory health

Participants: Joseph Abraham, Adrienne Ettinger, Tom Burke, Jon Samet, Scott Zeger

The objective of this project is to evaluate current surveillance methodology and develop cost-effective strategies for the ongoing assessment of the effect of air pollution on respiratory disease. The epidemiologic study will examine the relationship between chronic respiratory disease and air pollution using datasets which are publicly available.


Estimating distributed lag cumulative models with missing air pollution data

Participants: Brian Caffo, Francesca Dominici, Tom Louis, Scott Zeger

In this project we develop methods for estimating distributed lag models for time series analyses of air pollution and mortality when the air pollution data are missing. Our approaches include methods of momentís approximations, and likelihood solutions implemented via Markov chain Monte Carlo.


Point process methods for on-line disease surveillance

Participants: Peter Diggle, Ting-li Su, Barry Rowlingson

In this project we develop statistical methods for the early detection of spatially and temporally localized anomalies in the incidence pattern of disease, using data on cases of non-specific gastrointesinal disease in central southern England. We use a non-stationary Cox process model, in which a latent spatio-temporal Gaussian process is used to model anomalies after adjustment for the spatial distribution of the population at risk, normal seasonal variation in disease incidence and artefactual day-of-the-week effects due to reporting biases.


Geostatistical methods for tropical disease epidemiology

Participants: Peter Diggle, Barry Rowlingson

The aim of this project is to construct smoothly interpolated maps of spatial variation in disease risk, based on village-level survey data in which prevalence is estimated by testing a random sample of village residents. We use geostatistical models in which the observed prevalences follow a logistic regression with a binomial error distribution and a latent spatial Gaussian process within the linear predictor. Current applications include:

  1. Mapping loaloa prevalence in Cameroon, using elevation and green-ness of vegetation as explanatory variables;
  2. Calibrating a questionnaire-based instrument for the rapid assessment of loaloa prevalence against parasitological assessment, using bivariate models;
  3. Investigating the spatial variation in lymphatic filariasis prevalence in tropical west Africa in relation to malaria prevalence and mosquito species composition.


Bayesian geostatistical design

Participants: Peter Diggle, Soren Lophavn (Copenhagen)

When monitoring networks are used to produce interpolated maps of spatial variation in pollutant levels, the accuracy of the interpolation depends on the number and locations of monitoring stations. Geostatistical interpolation, or spatial prediction, methods require estimation of parameters in the geostatistical model, but designs which are efficient for prediction may not be efficient for parameter estimation, and vice versa. In this project, we propose and investigatea Bayesian geostatistical design criterion which seeks to minimize prediction uncertainty whilst making proper allowance for model parameter uncertainty.


Applications of spatial case-control methodology

Participants: Peter Diggle, Barry Rowlingson, Ting-li Su

In this project we apply recently developed methodology for estimating residual spatial variation in risk whilst adjusting for individual-level confounders and/or testing for the significance of risk factors of interest, using spatially referenced case-control data. Current or prospective applications include case-control studies of:

  1. Multiple sclerosis in the north west of England (in collaboration with Clive Hawkins and colleagues, North Stafforshire General Hospital,UK);
  2. Childhood asthma in East Baltimore (in collaboration with Greg Diette and colleagues, Johns Hopkins University).


Improved semi-parametric models for time series analyses of air pollution and mortality

Participants: Francesca Dominici, Aidan McDermott, Trevor Hastie

The nature and characteristics of time series data make risk estimation challenging, requiring complex statistical methods sufficiently sensitive to detect effects that can be small relative to the combined effect of other time-varying covariates. One widely used approach for a time series analysis of air pollution and health involves a semi-parametric Poisson regression with daily mortality or morbidity counts as the outcome, linear terms measuring the percentage increase in the mortality/morbidity associated with elevations in air pollution levels and smooth functions of time and weather variables to adjust for the time-varying confounders. We develop bandwidth selection methods for determining the number of degrees of freedom in the smooth functions of time and temperature that eliminate confounding bias in the estimation of the air pollution effects.


Joint analyses of cohort and time series data

Participants: Sorina Eftim and Francesca Dominici

To resolve discrepancies in time-series and cohort studies, in this project we develop statistical methods that allow us to jointly estimate short and long-term effects of pollution on health. We will apply the new methods to the National Medicare Cohort to jointly estimate acute and chronic health effects associated with short and long-term exposures to PM10 and PM2.5. Other data sources that will be used are the National Claim History Files (NCHF), the Medicare Current Beneficiary Survey (MCBS) and the National Air Pollution Monitoring Network with data for 1999-2000.


Bias in the exposure-response from spatial-temporal aggregation

Participants: Tom Louis, Francesca Dominici, Yijie Zhou

In this project we develop statistical methods and carry out empirical studies to assess bias in the estimated exposure-response slope rising from spatio-temporal aggregation of the environmental exposures. We develop the methods for Poisson regression and survival analyses.


Seasonal analyses of particulate matter and mortality

Participants: Roger Peng, Francesca Dominici, Scott Zeger

The focus of this project is to examine changes across seasons of the short-term effects of particulate matter on mortality. We are developing models for time series data that allow for interactions between season, pollution, and possibly other confounders. In addition we are looking at how the seasonality of pollution effects varies across regions of the U.S. We apply out methods to the recently updated NMMAPS data base which includes daily time series for the period 1987-2000 for the largest 100 cities in the US. One important component of this project is to explore associations between spatio-temporal variations of short-term effects of particulate matter on mortality and spatio-temporal variations of PM constituents available from the EPAís PM2.5 Chemical Speciation Trends Network (STN)


Higher-Order distributed lag models for quantifying effects of particulate matter air pollution and temperature on mortality

Participants: Leah Welty and Scott Zeger

The effects of daily air pollution and temperature on mortality may be distributed over several days; i.e. today's air pollution levels and temperature values may affect today's mortality as well as mortality on subsequent days. Distributed lag models allow for this type of exposure-response relationship. Our focus is on developing suitably flexible classes of distributed lag models to account for complexities in how particulate matter air pollution and temperature may individually and interactively affect mortality.


Resolving concerns of confounding by temperature and seasonality on the relationship between particulate matter air pollution and mortality, using NMMAPS

Participants: Leah Welty, Scott Zeger

A sensitivity analysis demonstrating that the air pollution estimates from the NMMAPS results are not substantively altered by considering different and more flexible models to account for effects of temperature and seasonality on mortality.


Bayesian hierarchical distributed lag models for estimating the association between summer ozone and cardiovascular mortality

Participants: Yi Huang, Francesca Dominici, Michelle Bell, Aidan McDermott, Jonathan Samet, Scott Zeger

In this research we develop Bayesian Hierarchical Distributed Lag Models for estimating associations between daily variations in summer ozone levels and daily variations in cardiovascular and respiratory (CVDRESP) mortality counts for the largest 19 cities included in NMMAPS for the period 1987 - 1994. We estimate an overall relative rate of CVDRESP mortality associated with exposure to summer ozone in the last week and investigate the sensitivity of this overall relative rate to confounding adjustment, model choices, and prior distributions.


Ozone and mortality in 95 U.S. cities from 1987 to 2000

Participants: Michelle Bell, Francesca Dominici, Aidan McDermott, Scott L Zeger, and Jonathan M Samet

Using analytical approaches developed for the National Morbidity, Mortality and Air Pollution Study (NMMAPS), we investigated the short-term effects of ozone le on mortality for 95 largest US cities over a 14 year period. We provide a national average estimate by use of hierarchical models. Our multi-city approach has several strengths: avoidance of selection bias, uniform analytic approaches, large statistical power, and information gained from heterogeneity of effects across cities.