Comparison of methods of point estimation in occupational epidemiologic studies

ABSTRACT

Two common study designs of occupational epidemiologic studies are cohort mortality studies, which use the population at risk as a denominator, and proportionate mortality studies, which use the total number of events as a denominator. This study compared the various methods of point estimation for cohort mortality studies, i.e., RR (risk ratio), OR (odds ratio) and SMR (standardized mortality ratio), and those for proportionate mortality studies, i.e., PRR (proportionate risk ratio), POR (proportionate odds ratio) and PMR (proportionate mortality ratio). This study was based on a real dataset of all workers in Metropolitan Toronto, Canada, who applied for compensation for various types of injuries or diseases from the Workers' Compensation Board in 1980. Results showed that within the cohort mortality or proportionate mortality study designs, OR (or POR) in all cases gave the least conservative estimates (farthest away from the null value), while SMR (or PMR) gave the most conservative estimates. The empirical differences between the point estimators were generally small. Our results showed that between study designs the corresponding point estimators were poorly correlated. In addition, this empirical study indicated that the use of the mortality odds ratio did not improve the proportionate mortality study very much in terms of generating results similar to the risk ratio from the cohort mortality design. We drew two conclusions: first, the point estimators within each study design can generally be a good alternative to one another; and second, proportionate mortality studies are not a good approximation for cohort mortality studies.