2016-Statistical Modelling in Environmental and Health Studies


Statistical Modelling in Environmental and Health Studies 
Chair: Tim Swartz (Simon Fraser University)
Organizer: Paramjit Gill (British Columbia-Okanogan) 

CINDY FENG, University of Saskatchewan
Modeling Spatially Correlated Survival Data: Impact of Misspecification of Correlation Structure on the Parameter Estimates  [PDF]
In epidemiological and environmental studies, time to event data are often grouped into clusters (e.g. families, clinical sites and geographical regions, etc.). Spatial survival models have been proposed in literature for modeling geographically correlated survival data, which includes a flexible spatially varying baseline hazard function to control for unmeasured spatial confounders and to borrow information across geographical units. The identified spatial pattern may highlight the regions requiring attention, which may assist public health professionals in their decision making. Few studies have been conducted to investigate the consequence of ignoring modeling the spatial correlation on the parameter estimates, so we conducted a simulation study to determine how the bias and efficiency in the parameter estimates change when misspecifying the correlation structure. 
RENJUN MA, University of New Brunswick
Mixed Models for Correlated Environmental/Health Data with Left Censoring and Right-Skewness [PDF]
Environmental and health data are often subject to left censoring at detection limit. Such censored data are usually right-skewed continuous, but with a point mass at the detection limit. Examples of such data are water pollutants in fish, tumor size due to radioactivity treatments, lesion depth related to ultrasound and precipitation. A special case of such data is so-called zero-inflated semi-continuous data where negligible amount is ignored. In this talk, we propose compound Poisson mixed models to characterize both the occurrence of the detection limit and size of the correlated data simultaneously. Our approach is illustrated with applications to real data. 
GYANENDRA POKHAREL, University of Calgary
Gaussian Process Emulator-Based Inference for Spatial Models Infectious Disease Systems  [PDF]
Statistical inference for mechanistic models of infectious disease spread is often very computationally expensive. Such models are generally fitted in a Bayesian Markov chain Monte Carlo (MCMC) framework that requires multiple calculation of likelihood function and is often computationally inefficient. This problem is more severe when incorporating large numbers of latent variables. Here, we propose a method of inference based on so-called emulation techniques. The method is again set in a Bayesian MCMC context, but avoids calculation of the computationally expensive likelihood function replacing it with a Gaussian process approximation. We show that such a method offers a significant gain in computation and can be used to infer the model parameters and underlying characteristics of spatial disease systems.