Lung cancer screening for current or former smokers and others at high risk of developing lung cancer reduces deaths from lung cancer, but at present these screening programs capture only a fraction of the individuals who could benefit.
Sybil could help, say researchers.
Sybil is a validated deep learning model that predicts future lung cancer risk based on a single low-dose CT (LDCT) scan, they explain.
The model was developed using more than 12,000 LDCT scans from the National Lung Screening Trial (NLST) and was validated in independent datasets from the NLST, Massachusetts General Hospital (MGH), and Chang Gung Memorial Hospital (CGMH).
Sybil achieved area under the receiver-operator curves of 0.92, 0.86, and 0.94 for lung cancer prediction at 1 year on the NLST, MGH, and CGMH datasets, respectively. Concordance indices over 6 years were 0.75, 0.81, and 0.80, respectively, reported co–first investigators Peter G. Mikhael and Jeremy Wohlwend, both of Massachusetts Institute of Technology, Cambridge.
The findings were published online on January 12 in the Journal of Clinical Oncology.
These findings suggest that Sybil accurately predicts future lung cancer risk from a single LDCT, according to the investigators, who noted that Sybil was able to forecast both short-term and long-term lung cancer risk on the NLST test set and maintained accuracy across diverse patient datasets.
"On the basis of our initial results, one potential clinical application is to use Sybil to decrease follow-up scans or biopsies among patients with nodules that are low risk," they concluded.
Furthermore, anecdotal cases mentioned in the study "spark contemplation about whether Sybil could be harnessed to decrease follow-up intervals or increase prioritization by the patient navigator and other tools to ensure those at highest risk are followed most closely," they noted.
However, they acknowledged that "[t]he benefit of such interventions will require confirmation in prospective clinical trials."
First, however, it will be necessary to "gain confidence" that the model is generalizable, because 92% of patients in the NLST development dataset were White, they said.
In an accompanying editorial, two experts explore the potential for Sybil to accelerate widespread lung cancer screening implementation. The editorialists are Gerard A. Silvestri, MD, of the Medical University of South Carolina, Charleston, and James R. Jett, MD, Emeritus, National Jewish Health, Denver, and Biodesix, Inc, Boulder.
They point out that two large trials have shown that screening reduces deaths from lung cancer — by 20% in the NLST in the United States and by 24% in the Dutch–Belgian lung cancer screening trial NELSON in Europe. These findings led to changes in screening guidelines and greater focus on implementing screening programs.
However, thus far, the "news on implementation is mixed," they commented.
The good news is that more than 90% of persons undergoing screening meet the eligibility criteria, according to an analysis of the first 1 million persons screened in the United States.
The bad news is that only about 6% of the eligible population is being screened, the screening rate has been about the same since 2019, and only 22% of people who are eligible for a repeat annual screen have received one, they said.
"Even more worrisome, a recent study found that in patients with abnormal/suspicious findings on their screening LDCT, follow-up testing occurred in only 42% of the patients, suggesting the possibility of delays in diagnosis which might result in patients who would have had early stage resectable disease being diagnosed with advanced cancer and the resultant poorer outcomes," they added.
Sybil may help change that. The model represents "an important first step toward a precision approach to lung cancer screening," Silvestri and Jett noted, citing the "outstanding" performance of the model.
From a practical perspective, the findings are important because "the model does not require patient demographics, risk factors or manual identification, and characterization of nodules, each of which take time and expertise, limiting the practical use of some of the other [deep learning] model," they wrote.
One advantage with Sybil is that "[i]n theory, the model code can be run in real time in the background of any radiology reading station and act as a second reader."
"Second, after prospective evaluation, one could envision this model being utilized to improve adherence to follow-up in those where the model suggests a high likelihood of the development of cancer in the future," they added.
The model might also decrease unnecessary workup and invasive testing and, with additional testing, might have the ability to identify patients with low lung cancer risk who would benefit from lengthening the interval between screens or discontinuing screening and identify at-risk groups that currently do not meet criteria for screening.
The latter is "an intriguing possibility because nearly half the lung cancers diagnosed in the United States currently do not qualify for screening," the editorialists noted.
"Understanding who would truly benefit from this technology will require significantly more investment in prospective studies targeting groups with differing risk profiles. To that end, the authors have graciously offered to provide the model code to other investigators to validate the usefulness or limitations of the model. Our hope is that investigators worldwide will take them up on that," they said.
The Sybil research was supported by the Bridge Project, the MIT Jameel-Clinic, Quanta Computing, Stand Up To Cancer, and the Massachusetts General Hospital Center for Innovation in Early Cancer Detection, including support from the Bralower and Landry Families, and Upstage Lung Cancer. Funding was also provided for by the Eric and Wendy Schmidt Center at the Broad Institute of MIT and Harvard. Mikhael reported a consulting or advisory role for Outcomes4Me, and Wohlwend reported having no disclosures. Silvestri reported relationships with AstraZeneca, Olympus Medical Systems, Biodesix, AstraZeneca/Daiichi Sankyo, Seer, and Amgen. Jett is employed by and has stock and other ownership interest in Biodesix.
Sharon Worcester, MA, is an award-winning medical journalist based in Birmingham, Alabama, writing for Medscape, MDedge, and other affiliate sites. She currently covers oncology, but she has also written on a variety of other medical specialties and healthcare topics. She can be reached at firstname.lastname@example.org or on Twitter: @SW_MedReporter
Lead image: Mopic/Getty Images
Medscape Medical News © 2023 WebMD, LLC
Send news tips to email@example.com.
Cite this: Deep Learning Model May Help Expand Lung Screening Programs - Medscape - Mar 01, 2023.