Acquisition and use of information on data quality in large population-based tumor registries

Date of Award




Degree Name

Doctor of Philosophy (Ph.D.)


Interdepartmental Studies

First Committee Member

Charles Kurucz, Committee Chair


A survey of North American centralized tumor registries, conducted as part of this study, indicated that quality control procedures and data obtained from quality control activities are used for registry management, but are not used to determine appropriate end use of registry data or to objectively modify registry data interpretation. Most analyses of registry data focus on obtaining point and interval estimates associated with factors involving the disease process. The contribution of the reporting process, by which true cases are transformed into registered cases, is ignored. To illustrate how such quality control data might be used three quality control studies were conducted as part of the routine quality control program of the Florida Cancer Data System, a large population-based tumor registry based at the University of Miami School of Medicine. Two analyses of these data were performed: one to adjust registry data for misclassification, and the other to adjust for incompleteness. In the first analysis, reabstracted data were used in a log-linear, linear, product model with both log-linear and product components to explicitly model the probabilities of misclassification of race and stage in female breast cancer. The model failed to detect strong evidence for differential misclassification of stage over the two races. In the second analysis, cases in the registry and cases found in an intensive multi-source casefinding study were arrayed as in a capture-mark-recapture study, and pure log-linear models were used to estimate age-specific and age-adjusted incidence rates corrected for incompleteness. The corrected estimate of age-adjusted incidence (345.1 cancers/100,000 population) was very similar to the comparable nationally reported figure (341.3) and was about 10% larger than the observed rate (313.8). It is recommended that quality control data, of the type used in the analyses in this dissertation, be collected routinely and be made a permanent part of any disease registry database.


Biology, Biostatistics

Link to Full Text