Potential Impact of Including Time to First Cigarette in Risk Models for Selecting Ever-Smokers for Lung Cancer Screening
Selecting individuals on the basis of model-predicted risks has been reported to improve lung cancer screening efficiency. On the basis of case-control studies, time to first cigarette (TTFC), a nicotine dependency measurement, has been a strong independent predictor of lung cancer risk. Our objective was to verify the TTFC–lung cancer association in the prospective National Lung Screening Trial and evaluate whether adding TTFC can improve lung cancer risk-prediction models.Methods:
Using Cox models, we examined associations between TTFC (≤5, 6–14, 15–29, 30–59, and ≥60 minutes) and lung cancer incidence and death in 18,729 National Lung Screening Trial participants, adjusting for smoking and other lung cancer risk factors comprehensively. We estimated 5-year individual lung cancer incidence by using models without and with TTFC and dichotomized into screening or no-screening groups based on risk thresholds of 1% and 2%. Area under the receiver operating curve values were calculated for models without and with TTFC.Results:
Smokers with a TTFC of 60 minutes or more had a much lower standardized 5-year lung cancer incidence risk—1.54% (1.52%–1.56%) for TTFC 60 minutes or more versus 4.07% (4.04%–4.09%) for TTFC 5 minutes or less—and lung cancer death risk—0.59% (0.57%–0.61%) for TTFC 60 minutes or more versus 2.26% (2.23%–2.28%) for TTFC 5 minutes or less (p trend < 0.001). Adding TTFC to the lung cancer incidence model improved the area under the receiver operating curve by 0.0079 (95% confidence interval = 0.0019–0.0138 [p = 0.0085]). Among 8994 smokers without a lung cancer diagnosis, 180 (2.00%) and 155 (1.67%) more people would be assigned to the no-screening group by adding TTFC to the model (p values for percent changes <0.001) at the 1% and 2% risk thresholds, respectively.Conclusion:
Including TTFC, which can be elicited by a single question at very low cost and noninvasively question, into risk models might better identify smokers with lower risk and could therefore be a safe, convenient tool to improve identification of those who benefit less from lung cancer screening.