Which risk models perform best in selecting ever-smokers for lung cancer screening?
, by DCEG Staff
A new analysis by scientists at NCI evaluates nine different individualized lung cancer risk prediction models based on their selections of ever-smokers for computed tomography (CT) lung cancer screening. The paper, published on May 15 in Annals of Internal Medicine, compares the ever-smoker populations chosen for screening by each model to provide guidance on which models could be considered for recommendation by future screening guidelines.
Earlier studies performed by this research group and others have demonstrated that using individualized lung cancer risk models to offer screening to ever-smokers could save more lives, and do so more effectively and efficiently, than current USPSTF recommendations. These existing recommendations are based only on age and smoking history. The 2018 National Comprehensive Cancer Network guidelines in the U.S. now permit the use of individualized risk models, which take into account a variety of demographic and clinical factors. However, several such models are available for use, and their performance had not previously been compared.
A team of researchers, led by Hormuzd Katki, Ph.D., estimated the populations chosen for screening by nine existing lung cancer risk models using data from the 2010-2012 National Health Interview Survey, a representative sample of the US population. They also compared the models’ statistical performance in two cohorts of ever-smokers: more than 330,000 in the NIH-AARP Diet and Health Study, and more than 70,000 in the American Cancer Society Cancer Prevention Study II Nutrition Survey Cohort.
The researchers found wide variation in the size and demographics of the U.S. screening population selected by each model, which ranged from 7.6 million to 26 million ever-smokers. These disagreements occurred because of the differing statistical performance of the models: four of the models estimated lung cancer risk accurately on average, but four other models substantially overestimated lung cancer risk. The remaining model’s performance fell in between that of the four most accurate and four least accurate models. The four top-performing risk models also selected similar numbers of ever-smokers for screening, and they distinguished best between individuals who would develop lung cancer and those who would not. However, even the top models disagreed on how to estimate risk for different racial or ethnic subgroups, which is an important area of future research.
The top models in this analysis are the Bach model, PLCOM2012, the Lung Cancer Risk Assessment Tool (LCRAT), and the Lung Cancer Death Risk Assessment Tool (LCDRAT) (the latter two were developed by authors of this study). The team hopes this analysis will inform future lung cancer screening guidelines to recommend using individualized risk models to select ever-smokers for CT lung cancer screening.
Reference: Katki HA et al. Implications of nine risk prediction models for selecting ever-smokers for CT lung-cancer screening. Annals of Internal Medicine. May 15, 2018. DOI: 10.7326/M17-2701