With advancing therapeutics, lung cancer survivors are rapidly increasing in number. While mounting evidence suggests lung cancer survivors have high risk of second primary lung cancer (SPLC), there are no evidence-based consensus screening guidelines for SPLC. To implement effective screening programs for lung cancer survivors, it is essential to evaluate individuals’ risk of SPLC and identify high-risk subgroups to be screened by computed tomography. Our group developed a SPLC prediction model, called Second Primary Lung Cancer Risk Assessment Tool (SPLC-RAT), for lung cancer patients who ever smoked by integrating key clinical factors, medical histories, tumor characteristics, and smoking histories based on a large prospective population-based cohort, the Multiethnic Cohort Study (MEC).

The development cohort included participants (N=6,325) with an ever-smoking history who were diagnosed with IPLC in 1993-2017 in the MEC. The primary outcome was defined as the time from initial primary lung cancer (IPLC) diagnosis to SPLC, death, or censored at the end of follow-up, whichever occurred first. SPLC was defined by the Martini and Melamed criteria: (1) histology is different from the IPLC, (2) the new tumor is diagnosed 2 years after the IPLC diagnosis, or (3) the new tumor, with same histology and developed within 2 years, is diagnosed in a different lobe or segment with no evidence of metastasis.

We applied a set of competing-risk models to obtain unbiased estimates of SPLC risks among lung cancer patients, a substantial proportion of whom tend to die before developing SPLC due to high comorbidities. We applied a cause-specific Cox regression (CSC) to build a prediction model for SPLC risk at the time of IPLC diagnosis. The model development was guided using three predictive performance metrics including discrimination (time-varying AUC), calibration, and the Brier score. Different forms of variables (categorical vs. continuous, or linear vs. non-linear effects) and all possible pairs of interactions between the selected predictors in the models were examined. After we finalized the proposed SPLC model, we used a bootstrap method to obtain unbiased internal assessments of the predictive accuracy metrics by optimism correction. External validation was performed using 2,963 and 2,844 IPLC patients in PLCO and NLST, respectively.

The proposed model showed a clear separation among low versus high-risk groups for SPLC among lung cancer patients. Decision curve analysis indicated that in a wide range of 10-year SPLC risk thresholds from 1% to 20%, the model yielded a larger net-benefit versus hypothetical all-screening or no-screening scenarios. This SPLC prediction model was validated using two large-scale randomized screening trials (PLCO and NLST) with diverse smoking histories and racial/ethnic distribution, demonstrating the potential generalizability.