Selection Criteria for Lung-Cancer Screening

Dr. Martin C. Tammemägi Professor (Epidemiology) Brock University Department of Community Health Sciences Walker Complex – Academic South, Room 306 St. Catharines, Ontario, Canada L2S 3A1Medical
Author Interview: Dr. Martin C. Tammemägi

Professor (Epidemiology) Brock University
Department of Community Health Sciences
St. Catharines, Ontario, Canada L2S 3A1

Medical What are the main findings of the study?

Dr. Tammemägi: Our study accomplished three things:

1. We presented an updated Lung Cancer Risk Prediction Model, which compared to our previously JNCI-published model, incorporates more predictors but is simpler to use because we changed the way we modeled nonlinear effects.

2. We demonstrated that using the Lung Cancer Risk Prediction Model to select individuals for lung cancer screening was much more effective than using the National Lung Screening Trial (NLST) enrolment criteria. 41.3% fewer lung cancers were missed. Sensitivity and positive predictive value of identifying individuals who develop lung cancer were significantly improved. Shortly after our NEJM paper was published, Ma et al published in CANCER their findings that 8.6 million Americans are NLST-criteria positive and if they were CT screened under ideal conditions 12,000 lung cancer deaths would be averted. Our NEJM article findings indicate that an additional 2,764 lives would be saved if the selection criteria had enrolled 8.6 million individuals for screening based on highest risk by our Lung Cancer Risk Prediction Model.

3. Importantly, using NLST data we demonstrated that the beneficial effect of CT screening did not vary by model predicted lung cancer risk.
Medical Were any of the findings unexpected?

Dr. Tammemägi: Many organizations have made recommendations endorsing CT screening of high-risk individuals and many institutions have begun or are planning to implement lung cancer screening programs shortly, enrolling participants based on the NLST enrollment criteria or some variant of it.

We were pleased to find that our prediction model substantially out-performed the NLST criteria in identifying individuals who would subsequently be diagnosed with lung cancer, but we were not surprised by this finding. The enrollment criteria for the NLST are simple, which served the trial purpose well enough. Our model includes many more lung cancer risk factors and did so in greater detail, did not dichotomize continuous variables, and included nonlinear relationships. Our model is considerably more sophisticated than the NLST criteria for selecting individuals who will be diagnosed with lung cancer.

Medical What should clinicians and patients take away from your report?

Dr. Tammemägi: Clinicians and institutions who are or are planning to implement a low-dose computed tomography lung cancer screening program should enroll individuals based on accurate model-predicted lung cancer risk rather than the NLST criteria. The NEJM article reports probability cupoints which are estimated to capture 80% and 90% of lung cancers, or screen roughly the same number of individuals as would be done using the NLST-criteria. A spreadsheet calculator which easily computes an individual’s risk of lung cancer according to the model is available for download for free for non-commercial users at

Medical What recommendations do you have for future research as a result of your study:

Dr. Tammemägi: The Pan-Canadian Early Detection of Lung Cancer Study enrolled 2537 individuals for lung cancer screening based on a prototype of the PLCO Lung Cancer Prediction Models. In a median follow-up of 3 years, 113 individuals were diagnosed with lung cancer (4.5%). This is much higher than observed in the NLST and was very close to the three-year probability estimated for the cohort by the prediction model. These findings corroborate the findings presented in the NEJM.

There are ways to improve the risk prediction model, but absence of data prevented inclusion of some variables in the current model. Further additional predictors, including pulmonary function and biomarkers, may make small incremental improvements to the existing model.

With quantitative models, we are investigating whether never-smokers or symptomatic individuals should be screened for lung cancer.

For studying special populations who have had occupational or environmental exposures to asbestos, I plan to prepare a model which will additionally include exposure to asbestos.

A major issue with lung cancer screening is the large number of false positives. Two groups, which I am working with, will be publishing risk models based on prospective population-based data which accurately predict which CT-detected nodule is cancer and which abnormal CT screen represents lung cancer. These models are expected to help provide some guidance to clinicians dealing with false positive screens.


Selection criteria for lung-cancer screening.

Tammemägi MC, Katki HA, Hocking WG, Church TR, Caporaso N, Kvale PA, Chaturvedi AK, Silvestri GA, Riley TL, Commins J, Berg CD.
Department of Community Health Sciences, Brock University, St. Catharines, ON, Canada.
N Engl J Med. 2013 Feb 21;368(8):728-36. doi: 10.1056/NEJMoa1211776.