summarizes the demographic details for the subjects by smoking status. Current smokers (350 cases, 244 controls) were well matched on age and gender. Among former smokers (375 cases, 371 controls), controls were, on average, 2 years younger than the case patients but well within the 5-year age matching criterion mandated by the study design. Case patients predictably were significantly heavier smokers that their respective control subjects. In former smokers, the mean DNA repair capacity (%) for the case patients was 8.2 (±2.6), compared with 9.0 (±3.5) for the control subjects (P < 0.001). The mean bleomycin sensitivity breaks per cell data were 0.8 (±0.4) and 0.7 (±0.4), respectively, for the case patients and control subjects (P < 0.001). The comparable data for current smokers were 8.3 (±2.9) and 9.2 (±3.6) for DNA repair capacity (P = 0.002) and 0.7 (±0.4) versus 0.6 (±0.3) for bleomycin sensitivity (P < 0.001).
Distribution of case patients and control subjects by selected variables
summarizes the results of the multivariable logistic regression analyses for former and current smokers separately. Although the data were derived from only a subset of subjects included in the original analysis, the same panel of risk factors identified from the univariate analysis were statistically significant in these multivariable logistic models. Specifically, for former smokers, lung cancer was statistically significantly associated with personal history of emphysema [odds ratio (OR), 2.08], exposure to dust (OR, 1.64), family history of any cancer (OR, 1.71), age at smoking cessation (OR, 2.61 for those who quit after age 54), and no prior history of hay fever (OR, 1.69). For current smokers, the ORs were 2.48 for emphysema, 1.68 for the heaviest smoking category, 2.02 for smoking-related family history, and 1.74 for asbestos exposure. The risk estimates for dust exposure (OR, 1.39) and no prior hay fever (OR, 1.51) were only of borderline significance (P = 0.09 and P = 0.08, respectively).
Multivariable logistic models for predicting lung cancer
For former smokers, the OR (95% CI) for DNA repair capacity (%) as a continuous variable was 1.11 (1.06-1.17). For mutagen sensitivity (breaks per cell), the OR was 1.87 (1.24-2.82). The comparable risk estimates for current smokers were 1.07 (1.02-1.14) and 4.33 (2.45-7.66), respectively. There was only a weak inverse correlation between DNA repair capacity and bleomycin sensitivity that achieved statistical significance only in currently smoking control subjects (data not shown), but this correlation did not lead to a significant colinearity or affect the overall model fit.
For former smokers, the AUC (95% CI) was 0.67 (0.63-0.71) for the model with the epidemiologic variables, 0.69 (0.65-0.73) with the addition of DNA repair capacity, 0.68 (0.64-0.72) for bleomycin, and 0.70 (0.66-0.74) when both assay data were added to the epidemiologic model (). For current smokers, the AUC for the baseline model was 0.68 (0.64-0.72) compared with 0.69 (0.64-0.73) with DNA repair capacity added, 0.72 (0.68-0.77) with only bleomycin sensitivity included, and 0.73 (0.69-0.77) with incorporation of both assays. Pairwise comparisons of the receiver operator characteristic curves showed that the expanded models incorporating both markers were statistically significantly better than the baseline models for both current and former smokers (P = 0.006 and P = 0.0048, respectively; data not shown).
Final models and results of 3-fold cross-validation analyses
The mean AUC for the overall expanded model derived from the 3-fold cross-validation showed good calibration as indicated by nonstatistically significant Hosmer-Lemeshow goodness of fit test statistics (). The cross-validation AUC values were 0.70 (0.63-0.77) and 0.68 (0.62-0.75) for current and former smokers, respectively.
In our previous article, we computed 1-year absolute risks for lung cancer using national incidence and mortality data for hypothetical scenarios with extreme risk profiles (1
). For example, a currently smoking 75-year-old White man, with a 58-pack-year smoking history, report of physician-diagnosed emphysema, one or more first-degree relatives diagnosed with a smoking-related cancer, and prior asbestos exposure, had a 1-year absolute risk (compared with a man of similar age, but without these risk factors) of 8.7% (). If, in addition, he exhibited suboptimal DNA repair capacity, his 1-year absolute risk was increased to 8.9%. When we added the mutagen-sensitive phenotype into the equation, his risk almost doubled to 16.3%. Likewise, for a female former smoker, age 66 y, who quit smoking at age 54 y, reported regular exposure to dusts, but denied any family cancer history, her baseline 1-year risk was 1.2% (). Adding poor repair capacity yielded an estimated 1-year risk of 2.3%. Together with mutagen sensitivity, this risk doubled from baseline to 3.1%.
One-year absolute risks of lung cancer for two hypothetical risk profiles