Repeatability of self-reported information for population-based studies of cancer.

PURPOSE: To measure the repeatability of a cancer risk factor questionnaire in a population-based case-control study. METHODS: Questionnaires were completed on two occasions by patients with cancer of the ovary (n=25) or esophagus (n=23) and by 37 controls without cancer. We assessed general cancer risk factors including height and weight (for calculating body mass index (BMI)), smoking and anti-inflammatory (NSAID) use. Risk factors specific for ovarian and esophageal cancers were also assessed. Agreement was measured by the correlation coefficient and weighted kappa statistic (kw) for continuous and categorical variables respectively. RESULTS: We observed very high levels of agreement for BMI (kw=0.84) and smoking history, including ages at initiation and quitting (Pearson correlation = 0.87 and 0.86 respectively). There was moderate to substantial agreement for use of anti-inflammatory drugs (aspirin kw =0.52, other NSAIDS kw =0.72). Agreement for lifetime prevalence of medical conditions varied from almost perfect (e.g. history of benign breast disease (k =0.86)) to moderate (e.g. heartburn (k =0.57)). Item repeatability was not materially altered by case-control status, age or sex of respondents or interval between completions. CONCLUSIONS: Self-reported cancer risk factor information demonstrates moderate to almost perfect levels of agreement, suggesting these items are suitable for risk estimation and epidemiologic inference.

Searching for cancer deaths in Australia: National Death Index vs. cancer registries.

OBJECTIVE: To compare the accuracy, costs and utility of using the National Death Index (NDI) and state-based cancer registries in determining the mortality status of a cohort of women diagnosed with ovarian cancer in the early 1990s. METHODS: As part of a large prognostic study, identifying information on 822 women diagnosed with ovarian cancer between 1990 and 1993, was simultaneously submitted to the NDI and three state-based cancer registries to identify deceased women as of June 30, 1999. This was compared to the gold standard of "definite deaths". A comparative evaluation was also made of the time and costs associated with the two methods. RESULTS: Of the 450 definite deaths in our cohort the NDI correctly identified 417 and all of the 372 women known to be alive (sensitivity 93%, specificity 100%). Inconsistencies in identifiers recorded in our cohort files, particularly names, were responsible for the majority of known deaths not matching with the NDI, and if eliminated would increase the sensitivity to 98%. The cancer registries correctly identified 431 of the 450 definite deaths (sensitivity 96%). The costs associated with the NDI search were the same as the cancer registry searches, but the cancer registries took two months longer to conduct the searches. CONCLUSIONS AND IMPLICATIONS: This study indicates that the cancer registries are valuable, cost effective agencies for follow-up of mortality outcome in cancer cohorts, particularly where cohort members were residents of those states. For following large national cohorts the NDI provides additional information and flexibility when searching for deaths in Australia. This study also shows that women can be followed up for mortality with a high degree of accuracy using either service. Because each service makes a valuable contribution to the identification of deceased cancer subjects, both should be considered for optimal mortality follow-up in studies of cancer patients.