Abstract CN07-02: Personalized Prediction of Bladder Cancer Risk: an Integrative Approach

Xifeng Wu

DOI: https://doi.org/10.1158/1940-6207.prev-08-cn07-02

2008-01-01

Cancer Prevention Research

Abstract:Abstract CN07-02 Background Bladder cancer is the fourth most common cancer in men in the United States. Bladder cancer is primarily an environment-induced cancer, with tobacco smoking and occupational exposure being the top two risk factors for the development of the disease. However, there is also compelling evidence that genetic predisposing play a role in bladder cancer etiology. One of the strongest and consistent genetic predisposing factors is mutagen sensitivity, a phenotypic assay that was recently demonstrated to have high genetic heritability in twin study. In this study, we evaluated mutagen sensitivity as a susceptibility marker for bladder cancer and explored the interplay of multiple epidemiologic risk factors and mutagen sensitivity in modulating bladder cancer risk in a large bladder cancer case-control study in M.D. Anderson Cancer Center. We developed the first risk prediction model of bladder cancer using extensive epidemiologic data. In addition, we incorporated mutagen sensitivity to build an enhanced epidemiologic-genetic risk prediction model for bladder cancer. Methods We performed a stepwise logistic regression analysis to identify significant risk factors for bladder cancer. We applied a series of statistical tools to build multivariate risk models and evaluate high-order interaction among these risk factors. We calculated the specificity and sensitivity of the model by constructing receiver operating characteristic (ROC) curves and calculating the area under the curve (AUC). We estimated the absolute risk of developing bladder cancer of representative individuals at 5, 10, and 15 years. We also explored similar approaches to predict the risk of recurrence for superficial bladder cancer patients. Results The significant risk factors included in the basic epidemiologic model included pack-years smoked, past exposures to diesel, aromatic amine, dry cleaning fluids, radioactive materials, and arsenic. This model yielded a good discriminatory ability with an AUC of 0.70 (95% CI: 0.67-0.73). When mutagen sensitivity data was incorporated into the model, the AUC increased to 0.80 (95% CI: 0.72-0.82), a substantial improvement in prediction accuracy. These two models both showed excellent concordance in our internal validation. Classification and regression tree (CART) revealed complex interplay between risk factors, which categorized individuals into subgroups with varying degrees of risk for bladder cancer. The subgroup with the highest risk exhibited a 67-fold increased risk over the reference subgroup. We also computed an ordinal risk score based on regression coefficients from the multivariate risk models. As the score increased, the percentage of controls in each category decreased while the percentage of cases increased, for example, the percentage of cases in the lowest-risk group (scores ≤ 2) derived from the enhanced model was 9.88%, whereas the percentage of cases in the highest-risk group (scores >25) was 84.42%, indicating that our model had a high discriminatory ability. Finally, we also applied similar multivariate analysis to predict clinical outcomes of bladder cancer, such as the recurrence of superficial bladder cancer with or without BCG treatment. Conclusions We constructed the first bladder cancer prediction model. Our data suggest that multivariate analyses of epidemiologic and genetic factors may improve prediction of cancer risk and clinical outcome. Citation Information: Cancer Prev Res 2008;1(7 Suppl):CN07-02.