Modelling risk factor information for linked census data: The case of smoking.

Area of Science:

• Biostatistics
• Public Health
• Epidemiology

Background:

• Statistics Canada links Census data with health outcomes.
• Linked data requires risk factor imputation, such as smoking status.
• Assessing statistical modeling for assigning smoking status to census respondents is crucial.

Purpose of the Study:

• To evaluate the feasibility of using statistical modeling to assign smoking status to individuals in census data.
• To develop and validate predictive models for smoking status using census variables.
• To assess the impact of modelled versus self-reported smoking status on smoking-related hospitalizations.

Main Methods:

• Developed age-/sex-specific predictive models using Canadian Community Health Survey (CCHS) data (2000/2001) based on 1991 Census variables.
• Validated the predictive models using subsequent CCHS data (2002/2003).
• Evaluated modelled smoking status against self-reported status using linked CCHS and Hospital Morbidity Database data.

Main Results:

• Identified key predictors for current daily smokers (income, education, marital status, dwelling ownership, region of birth) and never smokers (marital status, dwelling ownership, Aboriginal identity, region of birth).
• Modelled current daily smoker status showed increased odds of smoking-related hospitalization compared to never smokers, even after adjusting for covariates.
• The study confirmed the predictive power of socio-economic and identity variables for smoking status.

Conclusions:

• Statistical modeling is a feasible approach for assigning smoking status to census data.
• The availability of socio-economic and identity information is essential for accurate modeling.
• This methodology enhances the utility of linked census and health data for public health research.