This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Rights & Permissions Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Wu, X. Articles by Spitz, M. R. Search for Related Content PubMed PubMed Citation Articles by Wu, X. Articles by Spitz, M. R. Social Bookmarking                            What's this?

Projecting Individualized Probabilities of Developing Bladder Cancer in White Individuals

Xifeng Wu, Jie Lin, H. Barton Grossman, Maosheng Huang, Jian Gu, Carol J. Etzel, Christopher I. Amos, Colin P. Dinney, Margaret R. Spitz

From the Departments of Epidemiology and Urology, The University of Texas M.D. Anderson Cancer Center, Houston, TX

Address reprint requests to Xifeng Wu, MD, PhD, Department of Epidemiology, The University of Texas M.D. Anderson Cancer Center, 1155 Hermann Pressler Blvd, Houston, TX 77030; e-mail: xwu{at}mdanderson.org

	ABSTRACT

TOP ABSTRACT INTRODUCTION PATIENTS AND METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix 1 Appendix 2 Appendix 3 REFERENCES

 
Purpose There has been no risk assessment model for bladder cancer (BC). We developed the first model incorporating mutagen sensitivity and epidemiologic factors to predict BC risk.

Patients and Methods We used epidemiologic and genetic data from a large case-control study to build the models and constructed receiver operating characteristic curves. The area under the curve (AUC) was used to evaluate model discriminatory ability. We also projected absolute risk of developing BC by taking into account competing causes of death.

Results The study included 678 white BC patients and 678 controls. Significant risk factors in the epidemiologic model included pack-years smoked and exposures to diesel, aromatic amines, dry cleaning fluids, radioactive materials, and arsenic. This model yielded good discriminatory ability (AUC = 0.70; 95% CI, 0.67 to 0.73). When mutagen sensitivity data were incorporated, the AUC increased to 0.80 (95% CI, 0.72 to 0.82). The models showed excellent concordance in the internal validation. We also computed an easy to use ordinal risk score and provided examples for projecting absolute risk.

Conclusion We have developed the first risk prediction model for BC. The enhanced model integrating the genetic factor exhibited excellent discriminatory ability. Our model only requires an individual to answer a few simple questions during a clinic visit to project individualized probability. This model may be used as a basis for developing a Web-based tool for BC risk assessment. Validation of our model in an external population is an essential next step towards practical use in the clinical setting.

	INTRODUCTION

TOP ABSTRACT INTRODUCTION PATIENTS AND METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix 1 Appendix 2 Appendix 3 REFERENCES

 
Bladder cancer (BC) is the fourth most common cancer in US men and the second most common urologic malignancy, with approximately 67,160 new diagnoses and approximately 13,750 deaths in 2007.¹ The age-adjusted incidence of BC was 20.95 per 100,000 person-years for 2002 to 2003 from the Surveillance, Epidemiology, and End Results (SEER) Program. Cigarette smoking is the key environmental risk factor, followed by exposure to aromatic amines.^2-4

Screening for BC in the US general population is not currently recommended because there is no model to identify high-risk subgroups. Statistical models incorporating multiple risk factors for cancer can identify high-risk individuals who may be recruited to prevention trials.⁵ To date, risk prediction models have been developed to estimate individual risk for coronary artery disease,⁶ breast cancer,^7,8 melanoma,^9,10 colorectal cancer,¹¹ and lung cancer.^12-14 With the elucidation over the last three decades of major BC risk factors^2-4 and the identification of molecular and cytogenetic biomarkers for cancer risk assessment, we now have the means to more accurately define subgroups at high risk for BC. In this study, in addition to traditional epidemiologic factors, we incorporated an in vitro measure of latent genetic instability, mutagen sensitivity, to construct a multivariable risk prediction model for BC. The model was evaluated with respect to both goodness of fit and discriminatory ability. We also computed absolute risks of BC by taking into account competing causes of death.

	PATIENTS AND METHODS

TOP ABSTRACT INTRODUCTION PATIENTS AND METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix 1 Appendix 2 Appendix 3 REFERENCES

 
Study Population
Patients were enrolled from a large ongoing BC case-control study, which started patient recruitment in 1999 and is ongoing. BC patients were recruited from The University of Texas M.D. Anderson Cancer Center and from Baylor College of Medicine. The procedures for patient recruitment and eligibility criteria have been described elsewhere.¹⁵ Briefly, patient cases are patients with newly diagnosed and histologically confirmed urinary BC who have not previously received any chemotherapy or radiotherapy. There are no recruitment restrictions on age, sex, race, or cancer stage. The controls were recruited from the Kelsey-Seybold clinics, the largest private multispecialty physician group consisting of more than 23 clinics and more than 300 physicians in the Houston metropolitan area. The majority of control participants were healthy individuals coming to the clinics for annual health check-ups.¹⁶ Controls had no prior history of cancer (except nonmelanoma skin cancer). Controls were frequency matched to the patient cases by age (± 5 years), sex, and race. The study design to recruit patient cases and controls has proven to be valid and efficient for large-scale molecular epidemiologic studies for cancer.¹⁶ We only included white people in this analysis due to the small number of minority populations.

Data Collection
Written informed consent was obtained from both the patient cases and controls. Data were collected through a 45-minute interview and included demographic characteristics, lifestyle factors, occupational history, exposures to a defined list of substances at work or from hobbies, prior medical history, family history of cancer in first-degree relatives. Data on dietary patterns, which were not included in this analysis, were collected through another 45-minute interview (Appendix 1, online only). Human participant approval was obtained from the M.D. Anderson, Baylor, and Kelsey-Seybold Institutional Review Boards. Participation rates are approximately 92% for patient cases and 75% for controls.

Immediately after the interview, each participant donated a blood sample for molecular and cytogenetic analyses. For each individual, three parallel blood cultures, as previously described,¹⁷ were set up and incubated at 37^oC for a total of 4 days. One culture was treated with bleomycin 0.03 U/mL for 5 hours; one culture was treated with benzo[a]pyrene diol-epoxide (BPDE) with a final concentration of 2 µmol/L for 24 hours; and one culture was treated with 1.25 Gy of -radiation delivered from a cesium-137 irradiator. The number of chromatid breaks in 50 metaphases per sample was counted and expressed as the average number of breaks per cell.

Statistical Analysis and Model Building
Our model-building process started with the main effects model. Univariate logistic regression analysis was performed to assess the main effects of each individual risk factor. Then, we performed a stepwise logistic regression analysis to identify significant predictors in a multivariate model. The interaction between variables was tested by adding pair-wise product terms into the logistic regression model, after which higher order interactions were explored by performing a classification and regression tree (CART) analysis.¹⁸ We then evaluated the model goodness of fit and its discriminatory ability. In all analyses, the data were randomly and equally split into a training set to guide the building of the risk model and a validation set to assess the model prediction. Only the results from the final model as estimated from the full data set are presented. Throughout this article, we use the term relative risk for odds ratio (OR). We derived risk models in two levels; the epidemiologic model included only epidemiologic variables that are easy to ascertain in a clinical setting, whereas the epidemiologic-genetic model included all variables in the epidemiologic model plus the mutagen sensitivity phenotype.

Absolute risk of BC was estimated according to the methods of Gail et al⁷ and Dupont and Plummer,¹⁹ in which relative risks estimated from this case-control study were combined with the following two public data resources: the US incidence rates of BC obtained from the SEER registries for the years 2001 to 2003 and the mortality from causes other than BC derived from the National Center for Health Statistics (NCHS) report for the year 2003. See Appendix 2 (online only) for detailed methods for model building and absolute risk calculation. The CIs of absolute risk were calculated as previously described.²⁰

To further evaluate the discriminatory prediction accuracy of the model, we calculated each individual relative risk based on the coefficients from the multivariate logistic regression models, and we categorized individuals into five risk score subgroups on the basis of the distribution of their relative risk (see Appendix 2 for corresponding relative risk and risk score groups).

	RESULTS

TOP ABSTRACT INTRODUCTION PATIENTS AND METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix 1 Appendix 2 Appendix 3 REFERENCES

 
A total of 678 white patients with BC and 678 controls were included. There were 530 male (78.2%) and 148 female (21.8%) patient cases and sex-matched controls. The mean ages were 63.8 and 62.9 years for the patient cases and controls, respectively (P = .09). The percentage of current smokers in patient cases was higher than in controls (P < .0001). The smoking patient cases were self-reported heavier smokers (mean pack-years, 43.5; standard deviation, 30.9 pack-years) than the smoking controls (mean pack-years, 29.9; SD, 27.8 pack-years; P < .0001).

In the univariate analyses (Table A1, online only), significant main effects were observed for smoking status (former smoker: OR = 1.67; current smoker: OR = 5.39) and pack-years smoked (light smoker, < 40 pack-years: OR = 1.66; heavy smoker, 40 pack years: OR = 3.51). Positive exposure to diesel fuel/fumes, tar/mineral oil, dry cleaning fluids, leather and tanning solutions, rubber products, glues, pesticides/insecticides/herbicides, fertilizers, arsenic, zinc, radioactive materials, or aromatic amines was also associated with significantly increased risk (Table A1). A significant main effect was observed for a physician-diagnosed cystitis (OR = 1.47; 95% CI, 1.14 to 1.90). Because BC symptoms might coincide with benign urinary tract diseases, persons diagnosed with these diseases within 1, 3, 5, or 10 years before the interview or BC diagnosis were excluded, and BC risk was reanalyzed to determine the impact of exclusions on ORs. No medical history variables remained significant after these exclusions.

Mutagen sensitivity, a phenotypic marker reflecting the host DNA repair capacity (DRC) for in vitro mutagen challenges, was quantified by counting the number of induced chromatid breaks. We found that sensitivity to two challenge mutagens had significant main effects (BPDE: OR = 1.48; 95% CI, 1.33 to 1.64; and -radiation: OR = 1.92; 95% CI, 1.71 to 2.15; Table A1).

Epidemiology Multivariable Risk Model
In the epidemiologic model, a panel of risk factors was identified as significant predictors of BC after stepwise selection (Table 1); these risk factors included pack-years smoked and past exposures to diesel fuels/fumes, aromatic amines, dry cleaning fluids, radioactive materials, and arsenic. In the CART analysis, there were eight terminal nodes (data not shown). Consistent with the results from the stepwise regression analysis, significant variables were pack-years of smoking and exposures to aromatic amines, dry cleaning fluids, radioactive materials, arsenic, and diesel fuels.

In CART analysis, defining variables were the same as those identified by stepwise logistic regression selection. The tree structure resulted in 11 terminal nodes, defining a range of risk subgroups (Fig 1). Compared with the reference group (terminal node 1), the ORs for terminal nodes 2 to 11 ranged from 5.6 (node 2) to 67.1 (node 11).

View larger version (21K): [in this window] [in a new window] [PowerPoint Slide for Teaching]   Fig 1. High order interactions among epidemiologic-genetic risk factors in classification and regression tree analysis. Note, odds ratios and 95% CIs are presented underneath each terminal node box. BPDE, benzo[a]pyrene diol-epoxide.

View larger version (16K): [in this window] [in a new window] [PowerPoint Slide for Teaching]   Fig 2. Receiver operating characteristic curves from various models showing improvement of discrimination ability. Occupation, exposed to aromatic amines, radioactive materials, dry cleaning fluids, diesel fuel/fumes, and arsenic; EPI, pack-years; BPDE, benzo[a]pyrene diol-epoxide.

Consider first a 67-year-old man who is a heavy smoker (pack-years 40) and has been exposed to diesel fuels for 14 years but who reports no other exposures. His mutagen sensitivity phenotype for BPDE challenge is 0.75 breaks/cell. According to the epidemiologic-genetic model in Table 1, his relative risk would be r = 3.93 x 2.75 x 3.73 = 40.31. The age-specific BC incidence rate for white men 65 to 69 years old from the SEER database is * = 150.288 per 100,000. The AR for males in our epidemiologic-genetic model is 0.929. Thus, the estimated baseline hazard for men 65 to 69 years old is obtained as = *(1 –AR) = 0.00150288 (1 –0.929) = 0.0001067. From the NCHS data, the mortality rate for white men 65 to 69 years old from causes other than BC is µ = 0.02126. His projected absolute risk of developing BC would be 2.5% and 5.2% in the subsequent 5 and 10 years, respectively. The SEER incidence rates for white men aged 72 and 77 years are 0.22% and 0.29%, respectively. Therefore, the absolute risks would be 11.4 and 17.9 times the SEER rates upon his 72nd and 77th birthdays.

Risk Score
The distribution of the ordinal risk score in cases and controls based on the two models is shown in Figure 3. For example, the percentage of patient cases in the low-risk category (OR 0 to 2) was 9.88%, whereas the percentage of controls in the same risk category was 90.12%. However, in the high-risk category (OR > 25), the percentage of patient cases was 84.42%, whereas the percentage of controls was only 15.58% (Fig 3B).

View larger version (21K): [in this window] [in a new window] [PowerPoint Slide for Teaching]   Fig 3. Distributions of ordinal risk score in patient cases and controls. (A) Epidemiologic model. Summation of scores derived as follows: 1, odds ratio (OR) 1.00; 2, OR > 1.00 and 1.50; 3, OR > 1.50 and 3.50; 4, OR > 3.50 and 6.50; and 5, OR > 6.50. (B) Epidemiologic-genetic model. Summation of scores derived as follows: 1, OR 2.0; 2, OR > 2.0 and 5.0; 3, OR > 5.0 and 10.0; 4, OR > 10.0 and 25.0; and 5, OR > 25.0.

	DISCUSSION

TOP ABSTRACT INTRODUCTION PATIENTS AND METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix 1 Appendix 2 Appendix 3 REFERENCES

So far, to our knowledge, no risk assessment model for BC has been described in the literature. The most important characteristics of risk model performance are calibration, discrimination, and accuracy.⁵ In the current study, we used data collected in a large BC case-control study to develop the model and then internally validated its predictability. The cross-validation results showed that the model fitted our data with a relatively high level of calibration. Compared with published models, our epidemiologic model also has comparable discriminatory ability with an AUC of 0.70 (95% CI, 0.67 to 0.73). The Gail et al⁷ model has a concordance index of 0.67 (95% CI, 0.65 to 0.68), with Chen et al²³ recently reporting that adding the mammographic density improved the discriminatory ability. The lung cancer model of Bach et al¹³ has a concordance index of 0.72. The melanoma models reported a discriminatory accuracy of 0.62 (95% CI, 0.58 to 0.65)¹⁰ and 0.70 for women 50 years and older⁹ to 0.80 for men aged 20 to 49 years, an indication of good discriminatory ability.⁹ An important observation of our study was that the AUC increased from 0.7 to 0.8 when a genetic susceptibility marker was incorporated into the model.

The Harvard Cancer Risk Index²⁴ is a simple scoring system that yields a personalized risk of various cancers, including BC. In this system, risk scores are assigned according to the strength of association and then translated into points, taking into account the prevalence of risk factors in the United States. In this system, occupational exposure and smoking are the two risk factors with the strongest association with BC. Our model supports smoking and several high-risk occupational exposures as important risk factors of BC. The etiologic role of smoking in bladder carcinogenesis is now well established, with the relative risk of BC in smokers being two to four times that of never smokers.^25-31

Our model identifies exposure to aromatic amines as an important etiologic factor. Exposure to aromatic amines is the most important determinant of increased incidence of BC observed in workers in dyestuff and rubber manufacturing industries.^2,4,32-34 Our model is consistent with diesel exhaust exposure also having a strong etiologic role in BC, with numerous studies showing an excess incidence of BC in truck drivers and those exposed to diesel exhausts.^35-44 Consistent with numerous previous epidemiologic studies,^45-57 exposures to arsenic, dry cleaning fluids, and radioactive materials were identified as significant predictors of BC. Comparison of the percentages of exposed patients in our study with published BC studies conducted in the United States showed that the percentages are comparable for diesel fume exposures.^58-60 Our study population is comparable to other US populations in terms of prevalence of putative risk genotypes for BC.^61,62 Frequencies of other exposures are low in both our study and in other reports, so there is more variability in the estimates.

The improvement in the discriminatory ability of our model underscores the importance of considering genetic susceptibility in BC etiology. The mutagen sensitivity phenotype is considered as an indirect measure of DRC.⁶³ Numerous epidemiologic studies have shown significant associations between increased mutagen sensitivity and cancer risk.^64-69 Using a classical twin study design, we demonstrated that mutagen sensitivity has a strong genetic determinant.⁷⁰ The two phenotypes, sensitivity to BPDE and -radiation–induced DNA damage, are believed to reflect the DRC of the nucleotide excision repair pathway and the base excision repair and double-strand break repair pathways, respectively.

The National Cancer Institute's workshop, Cancer Risk Prediction Models: A Workshop on Development, Evaluation, and Application, endorsed incorporating biomarkers into risk assessment models. We did this in our model, which showed excellent discriminatory ability. Thus, our model is not only the first for BC, but also the first BC risk assessment model to include genetic markers. The fact that a simple noninvasive blood test increases discriminatory ability significantly demonstrates the promising future of cancer risk prediction.

Although our model demonstrated a high level of goodness of fit and excellent discriminatory ability, validation of our model in an external population is an essential next step towards practical use in the clinical setting. Furthermore, because the model was developed based on a white population, the results may not be generalizable to other racial groups. Although the data set from which our relative risk model is developed consisted predominantly of males, our model is also applicable to females because the association between identified risk factors in the relative risk model and BC risk is in the same direction and trend as in males. It should also be noted that the accurate projection of absolute risk greatly depends on the accuracy of relative risk estimation, which varies over time. Finally, the SEER incidence rates are cross-sectional rates (data based on individuals of different ages); therefore, the temporal fluctuation in age-specific incidence rates can have an appreciable impact in the estimation of projected absolute risk.

In conclusion, we developed the first risk prediction model for BC that could greatly facilitate the identification of high-risk populations. The excellent discriminatory ability of the epidemiology-genetic model demonstrates the promising future of cancer risk prediction by incorporating genetic factors. To estimate individualized probability of developing BC, our epidemiology model only requires an individual to answer a few simple questions during a clinic visit. If a person is willing to donate a small blood sample (a few milliliters), we will be able to incorporate results from a simple blood test to refine risk estimates based on the enhanced epidemiology-genetic model. The ordinal risk score may be used as a basis for developing a Web-based tool for BC risk assessment.

	AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

TOP ABSTRACT INTRODUCTION PATIENTS AND METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix 1 Appendix 2 Appendix 3 REFERENCES

 
The author(s) indicated no potential conflicts of interest.

	AUTHOR CONTRIBUTIONS

TOP ABSTRACT INTRODUCTION PATIENTS AND METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix 1 Appendix 2 Appendix 3 REFERENCES

 
Conception and design: Xifeng Wu, Margaret R. Spitz

Financial support: Xifeng Wu

Provision of study materials or patients: Xifeng Wu, H. Barton Grossman, Colin P. Dinney

Collection and assembly of data: Xifeng Wu, Jie Lin, Jian Gu

Data analysis and interpretation: Xifeng Wu, Jie Lin, Maosheng Huang, Jian Gu, Carol J. Etzel, Christopher I. Amos, Margaret R. Spitz

Manuscript writing: Xifeng Wu, Jie Lin, Jian Gu,

Final approval of manuscript: Xifeng Wu

	Appendix 1

TOP ABSTRACT INTRODUCTION PATIENTS AND METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix 1 Appendix 2 Appendix 3 REFERENCES

 

View larger version (19K): [in this window] [in a new window] [PowerPoint Slide for Teaching]   Fig A1. Modeled univariable and multivariable relationships between relative risk and each of the predictors derived from the final model. BPDE, benzo[a]pyrene diol-epoxide.

Smoking history. Data collected on smoking history include smoking duration, age at smoking initiation, number of cigarettes per day, duration of cessation, and computed pack-years. An individual who has never smoked or has smoked less than 100 cigarettes in his or her lifetime was defined as a never smoker. A former smoker was a person who had quit smoking at least 1 year before diagnosis (patient cases) or before the interview (controls). A current smoker was someone who currently smoked or who had stopped less than 1 year before diagnosis (patient cases) or before the interview (controls).

Family history. Family history data included cancer history in all first-degree relatives (biologic parents, siblings, and offspring). Specific information collected included whether the relative ever had cancer (yes or no), the site of cancer, the age at diagnosis, the vital status of the relative at the time of the interview, the current age or age at death of the relative, and the smoking status of the relative (yes or no).

Occupation and exposures. For the occupation history, participants were asked to describe the occupations they had held for at least 1 year. Participants were asked to give the job title and to describe the major duties of the job, equipment/materials/chemicals used while performing the job, and the type of work that the employer company did. Participants were also asked how long they had been in each job. Occupations and industries were then coded according to the occupational codes in the Dictionary of Occupational Titles (US Department of Labor, 1991). Data were also collected regarding prior regular (8 h/wk) and prolonged (at least 1 year) exposures to solvents, paint thinners, inks and dyes, paints, pigments, motor oils, gasoline, petroleum, car and truck exhaust, diesel fuels, natural gas, tar, mineral oil, photographic materials, hydrochloric acid, bleach/cleaners, dry cleaning fluids, leather and tanning products, glues, plastics, resins, pesticides, insecticides, herbicides, fertilizers, sawdust, wood dust, coal dust, soot, aluminum, arsenic, beryllium, chromium, chromates, lead, nickel, tin, zinc and other metals, dusts or fumes, radioactive materials, asbestos, and fiberglass. Participants were classified as positive to these substances if they self-reported such an exposure or if they had occupational codes suggesting they had held a job within a relevant industry.

Medical history. For prior medical history data collection, participants were asked whether they had ever been diagnosed by a physician with bladder infection, kidney infection, bladder stones, renal stones, or prostate infection (males only). The age of the patients at the time of diagnosis and/or the year of diagnosis were recorded for every positive response. The frequency of occurrence of these urinary tract diseases was also collected. Information on any surgery for removal of a benign bladder or kidney papilloma or polyps was also recorded, including the year of surgery.

Other lifestyle factors. Each participant was also asked about their ever use of hair dye products, the age when they first started using hair dyes, the frequency of use, the types of hair dye used (permanent or semipermanent), and the color of the most frequently used hair dye.

	Appendix 2

TOP ABSTRACT INTRODUCTION PATIENTS AND METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix 1 Appendix 2 Appendix 3 REFERENCES

 
Model Building and Projecting Absolute Risk of Developing Bladder Cancer
Main effects model. We first computed descriptive statistics for continuous and categoric variables separately. Contingency table analysis and the Pearson ² test were used to calculate the distributions of categoric variables between patient cases and controls. For continuous variables, we used the two-sample t test to test the differences between patient cases and controls. Smoothed scatter plots (ie, lowess scatter plots) were also used to ascertain the potential importance of the variables and to identify extreme observations as well as to assess the appropriateness of the scale of the continuous variables. Variables shown by the univariate analysis to be statistically significant (ie, to have a P < .25) were candidates for the multivariate logistic regression analysis. The significance level of P = .25 as a screening criterion was chosen because of the caveat that the traditional P = .05 often fails to identify variables of clinical/biologic importance. Stepwise logistic regression analysis was then performed to choose the final subset of predictors. Both clinical relevance and statistical significance were considered in determining the final variables to be retained in the model. For continuous variables, we further checked the assumption of linearity in the logit via fractional polynomials. Continuous variables that did not satisfy the linearity assumption were transformed accordingly (eg, by quartile smoothing). Before stepwise selection, we also checked correlations between variables to assess the impact of multicolinearity. In the event where two or more variables exhibited multicolinearity, we selected the variable with the most biologic and clinical relevance.

In all analyses, we derived the main effect models using the entire data set and stratified by matching variables. When stratified by sex, models derived for men did not differ from the model generated from the overall data set. In the model for women, however, all occupational variables dropped out because of the small number of exposed participants in the sample. However, the trend in the association between these variables and bladder cancer (BC) was similar to that of the males and of the overall data set. Similarly, models stratified by age group were similar to the master model. Therefore, we only presented models derived from the overall data set.

Interaction. All variables showing significant main effects in the multivariate model were tested for interactions. Interaction terms were tested by adding pair-wise product terms in the preliminary main effect model. Because no interaction terms were shown to be statistically significant by the likelihood ratio test, no interaction terms were included in the final model.

Higher order interactions between variables were explored in a classification and regression tree (CART) analysis in which a tree-based model is created by recursively partitioning the data; CART analysis can identify effect modifications between variables that are less evident in traditional logistic regression analysis and thus further discriminate between low- and high-risk subgroups. CART identifies subgroups with a range of relative risks. The recursive procedure produces subsequent nodes that are more homogeneous (with respect to the response variable) than the original node. Odds ratios (ORs) were calculated at each recursive split or node. We used the Helix Tree software (Golden Helix, Inc, Bozeman, MT) to perform the CART analysis. The terminal node with the lowest risk was chosen as the reference group.

Goodness of fit and discriminatory ability. We calculated the specificity and sensitivity of the model by constructing receiver operating characteristic curves and calculating the area under the curve (AUC) as well as the 95% CIs for the AUC. The AUC statistic measures the model's ability to discriminate between patient cases and controls. An AUC of 0.5 indicates no discrimination of the model, an AUC of 0.7 indicates good discrimination, and an AUC of 0.8 suggests excellent discrimination. We also used the leave one out cross-validation algorithm to predict the probabilities of the case-control status and then compared the predicted case-control status to true cancer status. The average prediction error rate was then calculated as an index of the model's discriminatory ability. A low-average prediction error rate indicates good model discrimination. The Hosmer-Lemeshow test was used to test overall goodness of fit of the model.

To further evaluate the discriminatory prediction accuracy of the model, we calculated each individual relative risk based on the coefficients from the multivariate logistic regression model, and we categorized individuals into five risk score subgroups on the basis of the distribution of their relative risk. For epidemiology risk model, individuals with an OR of 0 to 1.00 were defined as low risk; individuals with an OR of 1.01 to 1.50 were classified as medium-low risk; individuals with an OR of 1.51 to 3.50 were classified as medium risk; individuals with an OR of 3.51 to 6.50 were classified as medium-high risk; and individuals with an OR of 6.51 were classified as high risk. For the epidemiology-genetic risk model, the corresponding ORs for the five risk score subgroups were 0 to 2.0, 2.1 to 5.0, 5.1 to 10.0, 10.1 to 25.0, and 25.1. We then calculated percentages of patient cases and controls belonging to each of these risk subgroups to further evaluate the model's discrimination ability.

Projecting absolute risk of developing BC. Absolute risk of bladder cancer is the probability of developing bladder cancer over a specified time interval given age, sex, race, and risk factors. Suppose P (n, a, R) is defined as the probability that a participant whose current age is a with relative risk of R will be diagnosed with BC in the next n years. P (n, a, R) is then calculated as follows:

where (t) is the baseline incidence of BC for white males and females separately. F (a, R, t) is the probability that a participant aged a years with a relative risk of R will survive and not develop BC until age t. F (a, R, t) is calculated as follows:

where (x) is the baseline incidence as (t) above. µ(x) is the competing risk of BC, which is calculated by subtracting BC mortality rates from the total mortality rates obtained from the National Center for Health Statistics report for the year 2003.

The formula to calculate the baseline hazard (x) derived from the composite age-specific incidence rate from the Surveillance, Epidemiology, and End Results database and population attributable risk (AR) is as follows:

where *(x) is the composite age-specific incidence rate obtained from the Surveillance, Epidemiology, and End Results database for 2001 to 2003. AR(x) is attributable risk, which is calculated using the following equation:

where x is the number of cases in the study population, and r_i is the relative risk of individual i.

We have provided five examples for estimating the probability of developing BC within 5 and 10 years for individuals with different combinations of risk factors. In all examples, we used the enhanced epidemiology-genetic model to estimate relative risks. Table A2 lists the projected probability of developing BC within 5 and 10 years for five hypothetical patients. Exploration of AR estimated from our model showed that AR varied by sex but not by age. In males, the ARs were similar in the four quartile age groups (0.94, 0.93, 0.92, and 0.95 for the first, second, third, and fourth age quartiles, respectively). In females, the ARs were 0.86, 0.86, and 0.89 for the first, second, and third age tertile groups, respectively. Therefore, we calculated AR for males and females separately without further differentiation by age groups. According to the epidemiologic-genetic model, the ARs for males and females were 0.929 and 0.876, respectively. The ARs of the epidemiologic model were 0.56 for males and 0.38 for females (Table A2).

	Appendix 3

TOP ABSTRACT INTRODUCTION PATIENTS AND METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix 1 Appendix 2 Appendix 3 REFERENCES

1. What is your current age?
   
2. Have you smoked 100 or more cigarettes during your lifetime? Yes or no? If yes, how many years have you smoked cigarettes, and how many cigarettes per day?
   
3. Have you ever been in contact with (at least 8 hours a week for a year) aromatic amines either on a job or while working on a hobby? Yes or no?
   
4. Have you ever been in contact with (at least 8 hours a week for a year) diesel fuels/fumes either on a job or while working on a hobby? Yes or no? If yes, how many years?
   
5. Have you ever been in contact with (at least 8 hours a week for a year) dry cleaning fluids either on a job or while working on a hobby? Yes or no?
   
6. Have you ever been in contact with (at least 8 hours a week for a year) arsenic either on a job or while working on a hobby? Yes or no? If yes, how many years?
   
7. Have you ever been in contact with (at least 8 hours a week for a year) radioactive materials either on a job or while working on a hobby? Yes or no?
   
8. Would you like to donate a small sample of blood for further testing and refine risk estimates?
   

If no, stop and the questionnaire is completed.

If yes, proceed to get the blood sample draw.

	ACKNOWLEDGMENTS

 
We thank Waun Ki Hong, MD, for his valuable insights in initiating the risk model project and the continuous encouragement in completing the project. We thank Rebecca Ballow and Brandon Bruner for patient recruitment.

	NOTES

 
Supported by National Cancer Institute Grants No. CA74880 and CA91846.

Authors’ disclosures of potential conflicts of interest and author contributions are found at the end of this article.

	REFERENCES

TOP ABSTRACT INTRODUCTION PATIENTS AND METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix 1 Appendix 2 Appendix 3 REFERENCES

 
1. Jemal A, Siegel R, Ward E, et al: Cancer statistics 2006. CA Cancer J Clin 56 : 106 -130, 2006[Abstract/Free Full Text]

2. La Vecchia C, Airoldi LL: Human bladder cancer: Epidemiological, pathological and mechanistic aspects, in Capen TT (ed): Proceedings of a Consensus Conference: Species Differences in Thyroid, Kidney and Urinary Bladder Carcinogenesis. IARC Scientific Publication No. 147. Lyon, France, International Agency for Research on Cancer, 1999 , pp 139 -157

3. Negri E, La Vecchia C: Epidemiology and prevention of bladder cancer. Eur J Cancer Prev 10 : 7 -14, 2001[CrossRef][Medline]

4. Perlucchi C, Bosetti C, Negri E, et al: Mechanisms of disease: The epidemiology of bladder cancer. Nat Clin Pract Urol 3 : 327 -340, 2006[CrossRef][Medline]

5. Freedman AN, Seminara D, Gail MH, et al: Cancer risk prediction models: A workshop on development, evaluation, and application. J Natl Cancer Inst 97 : 715 -723, 2005[Abstract/Free Full Text]

6. Kannel WB, McGee D, Gordon T: A general cardiovascular risk profile: The Framingham Study. Am J Cardiol 38 : 46 -51, 1976[CrossRef][Medline]

7. Gail MH, Brinton LA, Byar DP, et al: Projecting individualized probabilities of developing breast cancer for white females who are being examined annually. J Natl Cancer Inst 81 : 1879 -1886, 1989[Abstract/Free Full Text]

8. Barlow WE, White E, Ballard-Barbash R, et al: Prospective breast cancer risk prediction model for women undergoing screening mammography. J Natl Cancer Inst 98 : 1204 -1214, 2006[Abstract/Free Full Text]

9. Fears TR, Guerry D, Pfeiffer RM, et al: Identifying individuals at high risk of melanoma: A practical predictor of absolute risk. J Clin Oncol 24 : 3590 -3596, 2006[Abstract/Free Full Text]

10. Cho E, Rosner BA, Feskanich D, et al: Risk factors and individual probabilities of melanoma for whites. J Clin Oncol 23 : 2669 -2675, 2005[Abstract/Free Full Text]

11. Selvachandran SN, Hodder RJ, Ballal MS, et al: Prediction of colorectal cancer by a patient consultation questionnaire and scoring system: A prospective study. Lancet 360 : 278 -283, 2002[CrossRef][Medline]

12. Peto R, Darby S, Deo H, et al: Smoking, smoking cessation, and lung cancer in the UK since 1950: Combination of national statistics with two case-control studies. BMJ 321 : 323 -329, 2000[Abstract/Free Full Text]

13. Bach PB, Kattan MW, Thornquist MD, et al: Variations in lung cancer risk among smokers. J Natl Cancer Inst 95 : 470 -478, 2003[Abstract/Free Full Text]

14. Spitz MR, Hong WK, Amos CI, et al: A risk model for prediction of lung cancer. J Natl Cancer Inst 99 : 715 -726, 2007[Abstract/Free Full Text]

15. Wu X, Gu J, Grossman HB, et al: Bladder cancer predisposition: A multigenetic approach to DNA-repair and cell-cycle-control genes. Am J Hum Genet 78 : 464 -479, 2006[CrossRef][Medline]

16. Hudmon KS, Honn SE, Jiang H, et al: Identifying and recruiting healthy control subjects from a managed care organization: A methodology for molecular epidemiological case-control studies of cancer. Cancer Epidemiol Biomarkers Prev 6 : 565 -571, 1997[Abstract]

17. Wu X, Gu J, Amos CI, et al: A parallel study of in vitro sensitivity to benzo[a]pyrene diol epoxide and bleomycin in lung cancer cases and controls. Cancer 83 : 1118 -1127, 1998[CrossRef][Medline]

18. Zhang H, Singer B: Recursive Partitioning in the Health Sciences. New York, NY, Springer, 1999

19. Dupont WD, Plummer WD: Understanding the relationship between relative and absolute risk. Cancer 77 : 2193 -2199, 1996[CrossRef][Medline]

20. Benichou J, Gail MH: Methods of inference for estimates of absolute risk derived from population-based case-control studies. Biometrics 51 : 182 -194, 1995[CrossRef][Medline]

21. Costantino JP, Gail MH, Pee D, et al: Validation studies for models projecting the risk of invasive and total breast cancer incidence. J Natl Cancer Inst 91 : 1541 -1548, 1999[Abstract/Free Full Text]

22. Fisher B, Costantino JP, Wickerham DL, et al: Tamoxifen for prevention of breast cancer: Report of the National Surgical Adjuvant Breast and Bowel Project P-1 Study. J Natl Cancer Inst 90 : 1371 -1388, 1998[Abstract/Free Full Text]

23. Chen J, Pee D, Ayyagari R, et al: Projecting absolute invasive breast cancer risk in white women with a model that includes mammographic density. J Natl Cancer Inst 98 : 1215 -1226, 2006[Abstract/Free Full Text]

24. Colditz GA, Atwood KA, Emmons K, et al: Harvard report on cancer prevention volume 4: Harvard Cancer Risk Index. Cancer Causes Control 11 : 477 -488, 2000[CrossRef][Medline]

25. Hartge P, Silverman D, Hoover R, et al: Changing cigarette habits and bladder cancer risk: A case-control study. J Natl Cancer Inst 78 : 1119 -1125, 1987[Medline]

26. Augustine A, Hebert JR, Kabat GC, et al: Bladder cancer in relation to smoking. Cancer Res 48 : 4405 -4408, 1988[Abstract/Free Full Text]

27. Vineis P, Esteve J, Hartge P, et al: Effects of the timing of tobacco in cigarette-induced bladder cancer. Cancer Res 48 : 3849 -3852, 1988[Abstract/Free Full Text]

28. D'Avanzo B, Negri E, La Vecchia C, et al: Cigarette smoking and bladder cancer. Eur J Cancer 26 : 714 -718, 1990[Medline]

29. Hartge P, Silverman DT, Schairer C, et al: Smoking and bladder cancer in blacks and whites in the United States. Cancer Causes Control 4 : 391 -394, 1993[CrossRef][Medline]

30. Castelao JE, Yuan JM, Skipper PL, et al: Gender and smoking related bladder cancer risk. J Natl Cancer Inst 93 : 538 -545, 2001[Abstract/Free Full Text]

31. Brennan P, Bogillot O, Cordier S, et al: Cigarette smoking and bladder cancer in men: A pooled analysis of 11 case-control studies. Int J Cancer 86 : 289 -294, 2000[CrossRef][Medline]

32. World Health Organization.Overall Evaluation of Carcinogenicity: An Updating of IRAC Monographs Volumes 1 to 42, Supplement 7—IRAC Monographs on the Evaluation of Carcinogenic Risks to Humans. Lyon, France, International Agency for Research on Cancer, 1987

33. World Health Organization.The Rubber Industry, Volume 28: IRAC Monographs on the Evaluation of Carcinogenic Risks to Humans. Lyon, France, International Agency for Research on Cancer, 1982

34. Bosetti C, Pira E, La Vecchia C: Bladder cancer risk in painters: A review of the epidemiological evidence, 1989-2004. Cancer Causes Control 16 : 997 -1008, 2005[CrossRef][Medline]

35. Silverman DT, Hoover RN, Mason TJ, et al: Motor exhaust-related occupations and bladder cancer. Cancer Res 46 : 2113 -2116, 1986[Abstract/Free Full Text]

36. Wynder EL, Onderdonk KJ, Mantel N: An epidemiological investigation of cancer of the bladder. Cancer 16 : 1388 -1407, 1963[CrossRef][Medline]

37. Jensen OM, Wahrendorf J, Knudsen JB, et al: The Copenhagen case-referent study on bladder cancer: Risk among drivers, painters, and certain other occupations. Scand J Work Environ Health 13 : 129 -134, 1987[Medline]

38. Claude JC, Frentzel-Beyme RR, Kunze E: Occupation and risk of cancer of the lower urinary tract among men: A case control study. Int J Cancer 41 : 371 -379, 1988[Medline]

39. Howe GR, Burch JD, Miller AB, et al: Tobacco use, occupation, coffee, various nutrients, and bladder cancer. J Natl Cancer Inst 64 : 701 -703, 1980[Medline]

40. Hoar SK, Hoover R: Truck driving and bladder cancer mortality in rural New England. J Natl Cancer Inst 74 : 771 -774, 1985[Medline]

41. Risch HA, Burch JD, Miller AB, et al: Occupational factors and the incidence of cancer of the bladder in Canada. Br J Ind Med 45 : 361 -367, 1988[Medline]

42. Silverman DT, Levin LI, Hoover RN, et al: Occupational risks of bladder cancer in the United State: I. White men. J Natl Cancer Inst 81 : 1472 -1480, 1989[Abstract/Free Full Text]

43. Kogevinas M, T'Mannetje A, Cordier S, et al: Occupation and bladder cancer among men in Western Europe. Cancer Causes Control 14 : 907 -914, 2003[CrossRef][Medline]

44. Boffetta P, Silverman DT: A meta-analysis of bladder cancer and diesel exhaust exposure. Epidemiology 12 : 125 -130, 2001[CrossRef][Medline]

45. Chiou HY, Chiou ST, Hsu YH, et al: Incidence of transitional cell carcinoma and arsenic in drinking water: A follow-up study of 8,102 residence in an arseniasis-endemic area in Northeastern Taiwan. Am J Epidemiol 153 : 411 -418, 2001[Abstract/Free Full Text]

46. Steinmaus C, Moore L, Hopenhayn-Rich C, et al: Arsenic in drinking water and bladder cancer. Cancer Invest 18 : 174 -182, 2000[Medline]

47. Cantor KP: Invited commentary: Arsenic and cancer of the urinary tract. Am J Epidemiol 153 : 422 -423, 2001[Free Full Text]

48. Bates MN, Rey OA, Biggs ML, et al: Case-control study of bladder cancer and exposure to arsenic in Argentina. Am J Epidemiol 59 : 381 -389, 2004

49. Steinmaus C, Bates M, Yuan Y, et al: Arsenic methylation and bladder cancer risk in case-control studies in Argentina and the United States. J Occup Environ Med 48 : 478 -488, 2006[CrossRef][Medline]

50. Chen YC, Su HJJ, Guo YLL, et al: Arsenic methylation and bladder cancer risk in Taiwan. Cancer Causes Control 14 : 303 -310, 2003[CrossRef][Medline]

51. Chen CJ, Chiou HY: Chen and Chiou respond to "Arsenic and Cancer of the Urinary Tract" by Cantor. Am J Epidemiol 153 : 422 -423, 2001[Free Full Text]

52. Karagas MR, Tosteson TD, Morris JS, et al: Incidence in transitional cell carcinoma of the bladder and arsenic exposure in New Hampshire. Cancer Causes Control 15 : 465 -472, 2004[CrossRef][Medline]

53. Moore LE, Smith AH, Eng C, et al: Arsenic-related chromosomal alteration in bladder cancer. J Natl Cancer Inst 94 : 1688 -1696, 2002[Abstract/Free Full Text]

54. National Toxicology Program: NTP and carcinogenesis studies of 1,2-dichloropropane in F344/N rats and B6C3F1 mice. Natl Toxicol Program Tech Rep Ser 263 : 1 -182, 1986[Medline]

55. Blair A, Stewart PA, Tolbert PE, et al: Cancer and the other causes of death among a cohort of dry cleaners. Br J Med 47 : 162 -168, 1990

56. Silverman DT, Levin LI, Hoover RN: Occupational risk of bladder cancer in the United States: II. Nonwhite men. J Natl Cancer Inst 81 : 1480 -1483, 1989[Abstract/Free Full Text]

57. Lynge E, Andersen A, Rylander L, et al: Cancer in persons working in dry cleaning in the Nordic countries. Environ Health Perspect 114 : 213 -219, 2006[Medline]

58. Silverman DT, Hoover RN, Albert A, et al: Occupation and cancer of the lower urinary tract in Detroit. J Natl Cancer Inst 70 : 237 -245, 1983[Medline]

59. Colt JS, Baris D, Stewart P, et al: Occupation and bladder cancer risk in a population-based case-control study. Cancer Causes Control 15 : 759 -769, 2004[CrossRef][Medline]

60. Schoenberg JB, Stemhagen A, Mogielnicki AP, et al: Case-control study of bladder cancer in New Jersey: I. Occupational exposures in white males. J Natl Cancer Inst 72 : 973 -981, 1984[Medline]

61. Gu J, Liang D, Wang Y, et al: Effects of N-acetyl transferase I and 2 polymorphisms on bladder cancer risk in Caucasians. Mutat Res 581 : 97 -104, 2005[Medline]

62. Garcia-Closas M, Malats N, Silverman D, et al: NAT2 slow acetylation and GSTM1 null genotypes increase bladder cancer risk: Results from the Spanish Bladder Cancer Study and meta-analyses. Lancet 366 : 649 -659, 2005[CrossRef][Medline]

63. Hsu TC, Johnston DA, Cherry LM, et al: Sensitivity to genotoxic effects of bleomycin in humans: Possible relationship to environmental carcinogenesis. Int J Cancer 43 : 403 -409, 1989[Medline]

64. Spitz MR, Fueger JJ, Beddingfield NA, et al: Chromosome sensitivity to bleomycin-induced mutagenesis, an independent risk factor for upper aerodigestive tract cancers. Cancer Res 49 : 4626 -4628, 1989[Abstract/Free Full Text]

65. Spitz MR, Hsu TC, Wu X, et al: Mutagen sensitivity as a biomarker of lung cancer risk in African Americans. Cancer Epidemiol Biomarkers Prev 4 : 99 -103, 1995[Abstract]

66. Wu X, Delclos GL, Annegers FJ, et al: A case-control study of wood-dust exposure, mutagen sensitivity, and lung cancer risk. Cancer Epidemiol Biomarkers Prev 4 : 583 -588, 1995[Abstract]

67. Cloos J, Spitz MR, Schantz SP, et al: Genetic susceptibility to head and neck squamous cell carcinoma. J Natl Cancer Inst 88 : 530 -534, 1996[Abstract/Free Full Text]

68. Baria K, Warren C, Roberts SA, et al: Chromosomal radiosensitivity as a marker of predisposition to cancers? Br J Cancer 84 : 892 -896, 2001[CrossRef][Medline]

69. Zheng YL, Lofferedo CA, Yu Z, et al: Bleomycin-induced chromosome breaks as a risk marker for lung cancer: A case-control study with population and hospital controls. Carcinogenesis 24 : 269 -274, 2003[Abstract/Free Full Text]

70. Wu X, Spitz MR, Amos CI, et al: Mutagen sensitivity has high heritability: Evidence from a twin study. Cancer Res 66 : 5993 -5996, 2006[Abstract/Free Full Text]

Submitted February 2, 2007; accepted August 9, 2007.

CiteULike    Complore    Connotea    Del.icio.us    Digg    Facebook    Reddit    Technorati    Twitter    What's this?

This article has been cited by other articles:

				M. C. Stern, J. Lin, J. D. Figueroa, K. T. Kelsey, A. E. Kiltie, J.-M. Yuan, G. Matullo, T. Fletcher, S. Benhamou, J. A. Taylor, et al. Polymorphisms in DNA Repair Genes, Smoking, and Bladder Cancer Risk: Findings from the International Consortium of Bladder Cancer Cancer Res., September 1, 2009; 69(17): 6857 - 6864. [Abstract] [Full Text] [PDF]

				J. Lin, A. Kamat, J. Gu, M. Chen, C. P. Dinney, M. R. Forman, and X. Wu Dietary Intake of Vegetables and Fruits and the Modification Effects of GSTM1 and NAT2 Genotypes on Bladder Cancer Risk Cancer Epidemiol. Biomarkers Prev., July 1, 2009; 18(7): 2090 - 2097. [Abstract] [Full Text] [PDF]

				A. N. Freedman, M. L. Slattery, R. Ballard-Barbash, G. Willis, B. J. Cann, D. Pee, M. H. Gail, and R. M. Pfeiffer Colorectal Cancer Risk Prediction Tool for White Men and Women Without Known Susceptibility J. Clin. Oncol., February 10, 2009; 27(5): 686 - 693. [Abstract] [Full Text] [PDF]

This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Rights & Permissions Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Wu, X. Articles by Spitz, M. R. Search for Related Content PubMed PubMed Citation Articles by Wu, X. Articles by Spitz, M. R. Social Bookmarking                            What's this?