Application of sample survey methods for modelling ratios to incidence densities

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Prospective epidemiologic studies generate complex incidence density data.
• Analyzing incidence densities requires methods that handle variability in at-risk periods.

Purpose of the Study:

• To introduce and illustrate novel ratio estimation methods for multivariate analysis of incidence densities.
• To assess the association between passive tobacco smoke exposure and lower respiratory illness (LRI) rates in children.
• To evaluate the persistence of this association after adjusting for covariates.

Main Methods:

• Employs a two-step ratio estimation process: estimating subgroup-specific incidence densities and their covariance matrix using Taylor series approximation.
• Tests differences in marginal LRI rates between exposed and unexposed groups.
• Fits a log-linear model to estimated ratios to assess covariate effects.
• Compares results with survey logistic regression and generalized estimating equation (GEE) methods (logistic and Poisson).

Main Results:

• The ratio method allows direct estimation of adjusted incidence density ratios for risk factors.
• Demonstrates application in a study of LRI in children, examining tobacco smoke exposure.
• Highlights the method's ability to incorporate age and season, and adjust for socioeconomic status, crowding, race, and feeding type.

Conclusions:

• Ratio estimation methods offer a robust approach for analyzing incidence densities in prospective studies.
• The methods are flexible and require minimal distributional assumptions.
• Applicable beyond epidemiology, including clinical trial adverse event analysis.