Potential for selection bias with tumor tissue retrieval in molecular epidemiology studies.

Molecular epidemiological studies of cancer generally require tumor tissue to evaluate somatic genetic alterations. Frequently this requires retrieval of fixed tissue blocks from hospital pathology archives. The availability of this material may be associated with disease severity, diagnostic practices, hospitals, or risk factors for disease. Tumor material is not available when the diagnosis is made clinically without histological confirmation. These characteristics create difficulties in defining the study base population. Incomplete access to tumor tissue has implications for description of the natural history of disease, estimates of the prevalence of mutation in the population, and evaluation of environmental exposures and critical target gene mutations. Differential diagnostic practices by age groups or across hospitals may create a biased population with respect to potential risk factors. However, this will not bias case-case comparisons unless the mutation of interest is associated both with the exposure of interest and the presence of a tumor block. When subjects with less severe disease are more likely to have biopsies, information regarding the natural history of the disease will be obscured. Investigation of the interaction of environmental agents and critical target gene mutations may be limited if, for example, an environmental agent is associated with a more aggressive form of the disease. Using an ongoing pancreatic cancer case-control study as an example, we discuss the potential for bias associated with differential availability of tumor blocks including consideration of tumor, patient, and hospital characteristics. Due to incomplete retrieval of tissue, the determinants of selection should be described in all studies using tumor tissue, and the implications for generalizability, power, and interpretation of findings in population-based studies should be considered.