Brigham Research Institute Poster Session Site logo-1

Judit Simon, MD, PhD

Pronouns

Rank

Fellow

Institution

Massachusetts General Hospital

Department

Radiology, Thoracic Imaging and Intervention

Authors

Judit Simon MD PhD, Peter Mikhael BSc, Ismail Tahir MB BCh BAO, Alexander Graur Cand Med, Amanda Fata BA, Jo-Anne Shepard MD, Francine Jacobson MD PhD, Regina Brazilay PhD, Lecia V Sequist MD MPH, Lydia E Pace MD, Florian J Fintelmann MD

Principal Investigator

Florian J Fintelmann, MD

Categories:

Sex differences in the prediction of future lung cancer risk based on a single low-dose chest computed tomography scan

Abstract

Introduction
A validated open access deep learning algorithm called Sybil can accurately predict long-term lung cancer risk from a single low-dose chest computed tomography (LDCT) scan and its accuracy exceeds clinical risk assessment. Sybil was trained on predominantly (60%) men and use of artificial intelligence algorithms trained on imbalanced cohorts may lead to inequitable outcomes in real-world settings.

Aims
We aimed to study whether Sybil works equally well in both sexes.

Methods
We included participants who underwent lung cancer screening LDCT at Brigham and Women’s Hospital and Massachusetts General Hospital between 2014 and 2019. Patients without follow-up were excluded. Patients diagnosed with lung cancer according to the institutional cancer registry within 6 years after the baseline LDCT were considered confirmed lung cancers. Those without a lung cancer diagnosis in the cancer registry and one or more negative follow-up LDCT were considered as negative for lung cancer. Area under the curve (AUC) values for women and men were compared with the DeLong-test.

Results
After exclusion, 10,588 LDCTs from 6,141 patients (47.1% women, mean age 64.9±6.2) were analyzed. Sybil achieved AUCs of 0.89 (95%CI: 0.85-0.93) for women and 0.89 (95%CI: 0.85-0.94) for men at 1 year, 0.85 (95%CI: 0.80-0.90) for women and 0.82 (95%CI: 0.77-0.88) for men at 2 years, 0.83 (95%CI: 0.78-0.88) for women and 0.81 (95%CI: 0.76-0.87) for men at 3 years, 0.83 (95%CI: 0.78-0.88) for women and 0.80 (95%CI: 0.75-0.86) for men at 4 years and 0.84 (95%CI: 0.79-0.89) for women and 0.78 (95%CI: 0.72-0.84) for men at 5 years; all p>0.05. At 6 years, AUC was 0.87 (95%CI: 0.83-0.93) for women and 0.79 (95%CI: 0.72-0.86) for men, p=0.009.

Conclusion
Sybil can accurately predict future lung cancer risk in women and men. For predicting long-term lung cancer risk at 6 years, Sybil performs better in women than in men.”

Research Context

Sex is an important variable to be considered for the successful implementation of cancer screening programs since women remain underrepresented in lung cancer research. Women also have different risk profiles as compared to men. Over the last decades, the incidence of lung cancer has decreased far more slowly in women compared to men. Researchers at Massachusetts General Hospital (MGH) and Massachusetts Institute of Technology (MIT) have developed, tested and validated a deep learning-based algorithm called Sybil. Sybil can accurately predict long-term lung cancer risk from a single LDCT scan and its accuracy exceeds clinical risk assessment. Given the different risk profiles of lung cancer between the two sexes, we aimed to study whether the current Sybil algorithm works equally well among men and women.