Modeling for Assessment of Environmental Health Data

• Activity Number: 178
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Shailendra Banerjee*+
• Companies: CDC
• Keywords: logistic regression ; modified Poisson ; log-binomial

Abstract

The effect of environmental factors can be assessed by fitting multivariate logistic regression model. The model evaluates the risk factors in presence of confounders. But, this model can give large odds ratio especially when the probability for the modeled event is large. So, other alternative models like log-binomial and modified Poisson regression models can be used. Log-binomial is a log-link generalized linear model. Since, odds ratio does not approximate to relative risk with large event probability, log-binomial model is used in this situation. In this presentation, we used several alternative models like log-binomial, logistic regression, modified Poisson regression on a lead poisoning data, where, probability of death due to lead poisoning was as high as 0.26. Logistic regression model gave a higher estimate of risk with a higher confidence interval. Log-binomial model was found to have failures with non-convergence and out-of-bounds predicted probabilities. Modified Poisson regression model resulted in good estimates of risk factor effects without any convergence problem. This study will be further investigated with simulated data and Bayesian modeling.