Purpose: Tobacco smoking and alcohol consumption are risk factors for several types of cancer and may act as confounders in aetiological studies. Large register-based cohorts often lack data on tobacco and alcohol. We present a method for computing estimates of cancer risk adjusted for tobacco and alcohol without exposure information.
Methods: We propose the use of confirmatory factor analysis models for simultaneous analysis of several cancer sites related to tobacco and alcohol. In the analyses, the unobserved pattern of smoking habits and alcohol drinking is considered latent common factors. The models allow for different effects on each cancer site, and also for appropriate latent site-specific factors for subgroup variation. Results may be used to compute expected numbers of cancer from reference rates, adjusted for tobacco smoking and alcohol consumption. This method was applied to results from a large, published study of work-related cancer based on census data (1970) and 21 years of cancer incidence data from the national cancer registry.
Results: The results from our analysis were in accordance with recognised risks in selected occupational groups. The estimated relative effects from tobacco and alcohol on cancer risk were largely in line with results from Nordic reports. For lung cancer, adjustment for tobacco implied relative changes in SIR between a decrease from 1.16 to 0.72 (Fishermen), and an increase from 0.47 to 0.95 (Forestry workers).
Conclusions: We consider the method useful for achieving less confounded estimates of cancer risk in large cohort studies with no available information on smoking and alcohol consumption.
Keywords: Alcohol drinking; Bias; Confounding factors; Epidemiological methods; Neoplasms; Tobacco smoking.