Selecting High-Risk Individuals for Lung Cancer Screening: A Prospective Evaluation of Existing Risk Models and Eligibility Criteria in the German EPIC Cohort

Lung cancer risk prediction models are considered more accurate than the eligibility criteria based on age and smoking in identification of high-risk individuals for screening. We externally validated four lung cancer risk prediction models (Bach, Spitz, LLP, and PLCO(M2012)) among 20,700 ever smokers in the EPIC-Germany cohort. High-risk subjects were identified using the eligibility criteria applied in clinical trials (NELSON/LUSI, DLCST, ITALUNG, DANTE, and NLST) and the four risk prediction models. Sensitivity, specificity, and positive predictive value (PPV) were calculated based on the lung cancers diagnosed in the first 5 years of follow-up. Decision curve analysis was performed to compare net benefits. The number of high-risk subjects identified by the eligibility criteria ranged from 3,409 (NELSON/LUSI) to 1,458 (NLST). Among the eligibility criteria, the DLCST produced the highest sensitivity (64.13%), whereas the NLST produced the highest specificity (93.13%) and PPV (2.88%). The PLCO(M2012) model showed the best performance in external validation (C-index: 0.81; 95% CI, 0.76-0.86; E/O: 1.03; 95% CI, 0.87-1.23) and the highest sensitivity, specificity, and PPV, but the superiority over the Bach model and the LLP model was modest. All the models but the Spitz model showed greater net benefit over the full range of risk estimates than the eligibility criteria. We concluded that all of the lung cancer risk prediction models apart from the Spitz model have a similar accuracy to identify high-risk individuals for screening, but in general outperform the eligibility criteria used in the screening trials.

Screening and early detection of lung cancer

The greatest news of the past year in this field was the first large-scale early detection trial that could prove a 20% reduction in lung cancer-related mortality by screening high-risk individuals with low-dose computed tomography (LDCT). Several expert groups and medical societies have assessed the data and concluded that LDCT screening for lung cancer is, however, not ready for large-scale population-based implementation. Too many open questions remain, such as definition of the at-risk population, timing and intervals of screening, optimal method of acquisition and interpretation of the images, how to handle (false) positive findings, and especially cost-effectiveness in relation to other lung cancer prevention strategies, mainly smoking cessation. Further analyses and several ongoing European trials are eagerly awaited. Much hope also resides in the use of biomarkers, as their use in, e.g., blood or exhaled air may provide more easy-to-use tests to better stratify high-risk populations for screening studies. While exciting research is ongoing in this domain--e.g. with microRNAs--none of the tests has yet reached sufficient validation for clinical use. Early central lung cancers are more difficult to visualise by CT. For these patients, standard bronchoscopy, complemented by autofluoresence endoscopy, has been studied in different screening and follow-up settings.

Lung cancer risk prediction to select smokers for screening CT--a model based on the Italian COSMOS trial

Screening with low-dose helical computed tomography (CT) has been shown to significantly reduce lung cancer mortality but the optimal target population and time interval to subsequent screening are yet to be defined. We developed two models to stratify individual smokers according to risk of developing lung cancer. We first used the number of lung cancers detected at baseline screening CT in the 5,203 asymptomatic participants of the COSMOS trial to recalibrate the Bach model, which we propose using to select smokers for screening. Next, we incorporated lung nodule characteristics and presence of emphysema identified at baseline CT into the Bach model and proposed the resulting multivariable model to predict lung cancer risk in screened smokers after baseline CT. Age and smoking exposure were the main determinants of lung cancer risk. The recalibrated Bach model accurately predicted lung cancers detected during the first year of screening. Presence of nonsolid nodules (RR = 10.1, 95% CI = 5.57-18.5), nodule size more than 8 mm (RR = 9.89, 95% CI = 5.84-16.8), and emphysema (RR = 2.36, 95% CI = 1.59-3.49) at baseline CT were all significant predictors of subsequent lung cancers. Incorporation of these variables into the Bach model increased the predictive value of the multivariable model (c-index = 0.759, internal validation). The recalibrated Bach model seems suitable for selecting the higher risk population for recruitment for large-scale CT screening. The Bach model incorporating CT findings at baseline screening could help defining the time interval to subsequent screening in individual participants. Further studies are necessary to validate these models.

Comparison of discriminatory power and accuracy of three lung cancer risk models

Background:

Three lung cancer (LC) models have recently been constructed to predict an individual's absolute risk of LC within a defined period. Given their potential application in prevention strategies, a comparison of their accuracy in an independent population is important.

Methods:

We used data for 3197 patients with LC and 1703 cancer-free controls recruited to an ongoing case–control study at the Harvard School of Public Health and Massachusetts General Hospital. We estimated the 5-year LC risk for each risk model and compared the discriminatory power, accuracy, and clinical utility of these models.

Results:

Overall, the Liverpool Lung Project (LLP) and Spitz models had comparable discriminatory power (0.69), whereas the Bach model had significantly lower power (0.66; P=0.02). Positive predictive values were highest with the Spitz models, whereas negative predictive values were highest with the LLP model. The Spitz and Bach models had lower sensitivity but better specificity than did the LLP model.

Conclusion:

We observed modest differences in discriminatory power among the three LC risk models, but discriminatory powers were moderate at best, highlighting the difficulty in developing effective risk models.

The potential for using risk models in future lung cancer screening trials

Computed tomography screening for early diagnosis of lung cancer is one of the more potentially useful strategies, aside from smoking cessation programmes, for reducing mortality and improving the current poor survival from this disease. The long-term success of lung cancer screening will be dependent upon identifying populations at sufficient risk in order to maximise the benefit-to-harm ratio of the intervention. Risk prediction models could potentially play a major role in the selection of high-risk individuals who would benefit most from screening intervention programmes for the early detection of lung cancer. Improvements of developed lung cancer risk prediction models (through incorporation of objective clinical factors and genetic and molecular biomarkers for precise and accurate estimation of risks), demonstration of their clinical usefulness in decision making, and their use in future screening programmes are the focus of current research.

Incorporation of a genetic factor into an epidemiologic model for prediction of individual risk of lung cancer: the Liverpool Lung Project

The Liverpool Lung Project (LLP) has previously developed a risk model for prediction of 5-year absolute risk of lung cancer based on five epidemiologic risk factors. SEZ6L, a Met430IIe polymorphic variant found on 22q12.2 region, has been previously linked with an increased risk of lung cancer in a case-control population. In this article, we quantify the improvement in risk prediction with addition of SEZ6L to the LLP risk model. Data from 388 LLP subjects genotyped for SEZ6L single-nucleotide polymorphism (SNP) were combined with epidemiologic risk factors. Multivariable conditional logistic regression was used to predict 5-year absolute risk of lung cancer with and without this SNP. The improvement in the model associated with the SEZ6L SNP was assessed through pairwise comparison of the area under the receiver operating characteristic curve and the net reclassification improvements (NRI). The extended model showed better calibration compared with the baseline model. There was a statistically significant modest increase in the area under the receiver operating characteristic curve when SEZ6L was added into the baseline model. The NRI also revealed a statistically significant improvement of around 12% for the extended model; this improvement was better for subjects classified into the two intermediate-risk categories by the baseline model (NRI, 27%). Our results suggest that the addition of SEZ6L improved the performance of the LLP risk model, particularly for subjects whose initial absolute risks were unable to discriminate into "low-risk" or "high-risk" group. This work shows an approach to incorporate genetic biomarkers in risk models for predicting an individual's lung cancer risk.