Premium
A comparison of methods for the analysis of recurrent health outcome data with environmental covariates
Author(s) -
Fung Karen Y.,
Khan Shahedul,
Krewski Daniel,
Ramsay Tim
Publication year - 2006
Publication title -
statistics in medicine
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.996
H-Index - 183
eISSN - 1097-0258
pISSN - 0277-6715
DOI - 10.1002/sim.2554
Subject(s) - covariate , estimator , statistics , poisson distribution , outcome (game theory) , crossover , generalized linear model , poisson regression , constant (computer programming) , mathematics , econometrics , computer science , medicine , artificial intelligence , population , environmental health , mathematical economics , programming language
Recurrent events such as repeated hospital admissions for the same health outcome occur frequently in environmental health studies. Dewanji and Moolgavkar proposed a flexible parametric model and a conditional likelihood analysis for recurrent events based on a Poisson process formulation. In this paper, we examine the statistical properties of the Dewanji–Moolgavkar (DM) estimator of the risk of an adverse health outcome associated with environmental exposures based on recurrent event data using computer simulation. We also compare the DM approach with both case–crossover analysis for multiple observations and time series analysis when there are no subject‐specific covariates. When using a correctly specified model, the DM method produced better estimates with respect to relative mean square error when each subject had constant or curved baseline intensity functions than it did when baseline intensities were increasing or decreasing in a linear fashion. For under‐specified models, the DM method outperformed case–crossover analysis for decreasing straight line intensity functions, was outperformed by case–crossover analysis for increasing straight line intensity functions, and was roughly equivalent to case–crossover analysis for constant and curved intensity functions. Case–crossover analysis produced superior risk estimates more frequently than the other two methods in the cases considered here, especially for linear representations of the baseline intensities. Copyright © 2006 John Wiley & Sons, Ltd.