Identifying high risk individuals for targeted lung cancer screening: Independent validation of the PLCO_m2012 risk prediction tool

Lung cancer screening with computerised tomography holds promise, but optimising the balance of benefits and harms via selection of a high risk population is critical. PLCO_m2012 is a logistic regression model based on U.S. data, incorporating sociodemographic and health factors, which predicts 6-year lung cancer risk among ever-smokers, and thus may better predict those who might benefit from screening than criteria based solely on age and smoking history. We aimed to validate the performance of PLCO_m2012 in predicting lung cancer outcomes in a cohort of Australian smokers. Predicted risk of lung cancer was calculated using PLCO_m2012 applied to baseline data from 95,882 ever-smokers aged ≥45 years in the 45 and Up Study (2006–2009). Predictions were compared to lung cancer outcomes captured to June 2014 via linkage to population-wide health databases; a total of 1,035 subsequent lung cancer diagnoses were identified. PLCO_m2012 had good discrimination (area under the receiver-operating-characteristic-curve; AUC 0.80, 95%CI 0.78–0.81) and excellent calibration (mean and 90th percentiles of absolute risk difference between observed and predicted outcomes: 0.006 and 0.016, respectively). Sensitivity (69.4%, 95%CI, 65.6–73.0%) of the PLCO_m2012 criteria in the 55–74 year age group for predicting lung cancers was greater than that using criteria based on ≥30 pack-years smoking and ≤15 years quit (57.3%, 53.3-61.3%; p < 0.0001), but specificity was lower (72.0%, 71.7–72.4% versus 75.2%, 74.8–75.6%, respectively; p < 0.0001). Targeting high risk people for lung cancer screening using PLCO_m2012 might improve the balance of benefits versus harms, and cost-effectiveness of lung cancer screening.