A model incorporating volume doubling time (VDT) and select radiomic features was able to predict tumor behavior for screen-detected lung cancers, an analysis of low-dose CT (LDCT) scans from the National Lung Screening Trial (NLST) showed.
Using a VDT of 234 days and radiomic features of compactness and average concurrence, very high-risk patients had a 5-year overall survival (OS) of 21.4%, compared with 82.4% for the low-risk group (P<0.0001), reported Jaileene Pérez-Morales, PhD, of Moffitt Cancer Center in Tampa, Florida.
The model also “identified a vulnerable group of early-stage lung cancer patients who had a high risk of experiencing poor survival outcomes,” she said during her presentation at the virtual North America Conference on Lung Cancer.
For this group, the decision tree was able to discriminate between high-risk tumors with a 5-year OS of 39.9%, and more indolent tumors with a 5-year OS of 80.8% (P<0.0001).
“Use of volume doubling time was a hallmark of the NELSON study,” noted discussant Betty Tong, MD, MHS, of Duke University Medical Center in Durham, North Carolina.
“In the study protocol, a volume doubling time of less than 400 days was considered to be positive,” she said. “In contrast to the current study, the NELSON authors determined that the volume doubling time in later rounds of screening was more variable, and that decreasing the volume time less than 400 days was not recommended in later screening rounds.”
For their study, Pérez-Morales and colleagues used LDCT scans from NLST involving 88 patients with malignancy at first follow-up to generate radiomic features that could help classify risk. Participants in the trial underwent three LDCT scans — at baseline and years 1 and 2. A decision-tree analysis was developed to stratify patients into four risk groups (low, intermediate, high, and very high), which showed differences in both progression-free survival (PFS) and OS.
“We also noticed that more aggressive VDTs occurred in the later screening follow-up than the first follow-up,” said Pérez-Morales. “These high-risk patients may require aggressive follow-up and/or adjuvant therapy to mitigate their poor outcomes.”
As VDT requires temporal imaging, the researchers also looked at whether radiomics features could identify high-risk tumors from an initial LDCT scan, and found that two peritumoral features (non-uniformity in Gray level size zone [GLSZM] and average 3D run length) were both predictive of VDT.
“The authors are no strangers to the use of radiomics in lung cancer screening,” Tong noted, pointing to a recently published paper from the group that examined radiomic features of non-small cell lung nodules in the NLST dataset. Again, they created a risk classification system for patients and showed significant differences in PFS and OS between low-, intermediate-, and high-risk groups.
“Even more interesting was the radio-genomic analysis that was performed using a separate dataset of resected tumors and their associated scans — there were differential levels of gene expression associated with select radiomic features,” said Tong.
“One can imagine a possible future state where volume doubling time, radiomics, and even radio-genomics