Four risk prediction models found most accurate for guiding lung cancer screening

These models picked similar numbers of patients who had ever smoked, showed better agreement on size of the screening population, and achieved consensus on 73% of persons chosen, while the other five models generally overestimated risk.

Four risk prediction models have been shown to be most accurate for selecting the highest-risk patients who have ever smoked for lung cancer screening, a study found.

Researchers from the National Cancer Institute (NCI) and the American Cancer Society compared nine risk models to guide lung cancer screening: the Bach model; the Spitz model; the Liverpool Lung Project (LLP) model; the LLP Incidence Risk Model (LLPi); the Hoggart model; the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial Model 2012 (PLCOM2012); the Pittsburgh Predictor; the Lung Cancer Risk Assessment Tool (LCRAT); and the Lung Cancer Death Risk Assessment Tool (LCDRAT).

Model predictions were evaluated in two large U.S. cohorts of patients who had ever smoked: 337,388 people in the National Institutes of Health–AARP Diet and Health Study and 72,338 people in the CPS-II (Cancer Prevention Study II) Nutrition Survey cohort. Results were published May 15 by Annals of Internal Medicine.

At a five-year risk threshold for lung cancer of 2.0%, the models differed widely in the number of patients who had ever smoked that they selected (7.6 million to 26 million), and there was no consensus on which patients to select for screening.

Four models, the Bach model; PLCOM2012; LCRAT, and LCDRAT, performed best, as measured by their ability to accurately predict risk and to distinguish high-risk individuals from low-risk individuals (expected-observed ratio range, 0.92 to 1.12) and by their higher areas under the curve (AUC) (range, 0.75 to 0.79). These models picked similar numbers of patients who had ever smoked, showed better agreement on size of the screening population (7.6 million to 10.9 million), and achieved consensus on 73% of persons chosen. The other five models generally overestimated risk (expected-observed ratio range, 0.83 to 3.69) and had lower AUCs (range, 0.62 to 0.75).

These findings could be used in future guidelines to recommend risk models for lung cancer screening, the researchers suggested.

An accompanying editorial laid out six criteria that need to be met for the routine implementation of these models in practice:

  • Convince policymakers to accept the use of models to identify screening-eligible persons, despite the fact that the U.S. Preventive Services Task Force and CMS do not recommend using model-estimated risk;
  • Overcome the perception that models identify many persons who are too elderly and sick to benefit from screening;
  • Establish thresholds that define high cancer risk, since they vary among models;
  • Determine optimal ways of implementing risk models;
  • Explore whether model-based risk estimation helps to avoid health disparities, and
  • Determine how to incorporate prior screening results to inform individual risk.

“Lung cancer screening is rapidly evolving,” the editorial stated. “Over the next few years, many of the aforementioned issues are likely to be addressed and lung cancer screening is likely to improve.”

The cover article in the May 2018 ACP Internist offers an overview of current screening recommendations for lung cancer and talks to experts about overcoming barriers to best practice.