Assessing the relationship between marital status and cancer incidence: methodologic considerations.

In registry-based population studies on marital status in relation to cancer, incidence rates sometimes have been calculated using marital status-specific populations that have been estimated by interpolation and extrapolation from census data as a denominator. Alternatively, other cancers from the same registry have been used to estimate the proportion of the population in each marital-status category in the calculation of the relative risk (RR) of a given cancer. Using cancer registry data from four United States populations for the years 1979-87, we compared the relative incidence estimated using each of the two methods. For selected cancers diagnosed during 1979-81, the age-adjusted risks of never-married Black persons were 1.5 to 2.2 times those of married persons when the population size was estimated from census data. The corresponding RRs were 0.7 to 1.1 when the 'control' cancers were used. Among Whites, the differences between the two methods were about 20 to 30 percent. For both races, the difference between the methods was greater still for the years for which we relied on extrapolation to estimate the population (1981-87). The differences between the risk estimates from the two methods may be related to underenumeration in the census, inconsistent definitions of marital status between cancer registries and the census, errors in the extrapolation of the population, and/or the possible association of the incidence of 'control' cancers with marital status. In the US, while each method has some potential for bias, we believe that the likelihood of bias is relatively greater using the census-based method.