Quantitative CT Imaging Measures and Clinical Risk Factors for Airflow Obstruction in a Lung Cancer Screening Population

Principal Investigator

Name
Sabee Molloi

Degrees
Ph.D.

Institution
The Regents of the University of California- Irvine

Position Title
Professor

Email
symolloi@uci.edu

About this CDAS Project

Study
NLST (Learn more about this study)

Project ID
NLST-1536

Initial CDAS Request Approval
Oct 5, 2026

Title
Quantitative CT Imaging Measures and Clinical Risk Factors for Airflow Obstruction in a Lung Cancer Screening Population

Summary
Low-dose chest CT for lung cancer screening is acquired in millions of adults who share the principal risk factor for chronic obstructive pulmonary disease, yet much of the airflow obstruction in this population remains undiagnosed because spirometry is not part of the screening pathway. Whether quantitative measurements derived from the screening CT itself can identify these individuals, and whether they add information beyond a participant's demographics, smoking history, and medical history, is not settled.

This project evaluates quantitative CT-derived imaging measures against clinical risk factors for the identification of spirometrically defined airflow obstruction (pre-bronchodilator FEV1/FVC < 0.70) in the ACRIN arm of NLST, the subset of participants for whom baseline spirometry was collected. Spirometry is obtained separately through ECOG-ACRIN; the CDAS datasets requested provide the clinical covariates and the screening and technical context that the imaging analysis requires.

Design: retrospective cross-sectional analysis of baseline screening CT paired with contemporaneous baseline spirometry, using images from TCIA and the NCI Imaging Data Commons. Participants are assigned once to a discovery set and a held-out evaluation set at cohort creation, stratified by outcome, sex, and scanner manufacturer. All model selection and tuning occur in discovery; all reported performance comes from pooled out-of-fold predictions in the evaluation set. Established quantitative CT measures (densitometric emphysema indices, airway and texture features), additional image-derived quantitative features, and machine-learning models trained on the images will be compared against a clinical risk-factor model built from the Participant dataset: age, sex, race, education, height and weight, pack-years and smoking intensity, duration and cessation, occupational exposures, and self-reported respiratory disease history.

Analysis will report discrimination (AUC with DeLong confidence intervals), calibration (intercept, slope, Brier score), and decision-curve net benefit, with prespecified subgroup estimation by scanner manufacturer, reconstruction kernel, slice thickness, sex, and severity grade. The screening exam datasets provide the technical parameters needed to characterize and adjust for scanner heterogeneity; the abnormality, lung cancer, and cause of death datasets are used to describe the cohort and to account for prevalent and incident malignancy. Findings are intended for peer-reviewed publication.

Aims

Assemble and characterize the analytic cohort: NLST ACRIN participants with baseline spirometry and a baseline low-dose screening CT, reporting demographic, smoking, anthropometric, occupational, and comorbidity characteristics of the discovery and evaluation sets, together with the scanner, reconstruction kernel, and slice-thickness composition of the imaging.

Establish the clinical risk-factor reference model for spirometrically defined airflow obstruction using variables available without imaging: age, sex, race, education, height and weight, pack-years, smoking intensity, duration and cessation, occupational exposure history, and self-reported respiratory disease history.

Quantify the performance of quantitative CT-derived imaging measures and machine-learning models for the same outcome, using only the baseline screening CT.

Determine whether imaging-derived measures add discrimination, calibration, and clinical net benefit beyond the clinical risk-factor model, and beyond established densitometric emphysema indices.

Evaluate robustness of the imaging models across scanner manufacturer, reconstruction kernel, slice thickness, sex, and obstruction severity, and under prespecified sensitivity definitions of the spirometric outcome.

Collaborators

Sabee Molloi The Regents of the University of California
DALE BLACK The Regents of the University of California