A Comparison of the PanCan Model and Lung-RADS to Assess Cancer Probability Among People With Screening-Detected, Solid Lung Nodules.

Authors

Sundaram V, Gould MK, Nair VS

Affiliations

• Quantitative Sciences Unit, Stanford University School of Medicine, Stanford, CA. Electronic address: sund@stanford.edu.
• Division of Health Services Research and Implementation Science, Kaiser Permanente Southern California, La Cañada Flintridge, CA; Department of Health Systems Science, Kaiser Permanente Bernard J. Tyson School of Medicine, Pasadena, CA.
• Clinical Research Division, Fred Hutchinson Cancer Research Center, Seattle, WA; Division of Pulmonary, Critical Care and Sleep Medicine, University of Washington School of Medicine, Seattle, WA.

Abstract

BACKGROUND: The Pan-Canadian Early Detection of Lung Cancer (PanCan) risk model and the Lung CT Screening Reporting & Data System (Lung-RADS) estimate cancer probability for screening-detected nodules. The accuracy and agreement of these models require further study.

RESEARCH QUESTION: What is the performance of the PanCan model and Lung-RADS to estimate the probability of cancer in screening-detected solid nodules?

STUDY DESIGN AND METHODS: We analyzed data for newly identified, solid nodules detected on any screening round in the low-dose CT arm of the National Lung Screening Trial to assign a PanCan risk and Lung-RADS score. We compared PanCan risk with the corresponding Lung-RADS category according to the expected prevalence of cancer and examined accuracy using logistic regression and between-test agreement. We also analyzed baseline screen-detected nodules only, high (defined as ≥ 5% probability of cancer) vs low-risk nodules, "risk-gap" nodules with a 3% to 5% PanCan probability and no equivalent Lung-RADS category, and procedure use by model.

RESULTS: Participants with solid nodules (6,956) had a calculable PanCan risk and Lung-RADS score. PanCan accuracy by cancer probabilities < 1%, 1% to 2%, 5% to 15%, and > 15% was similar to corresponding Lung-RADS categories 2, 3, 4A, and 4B for any solid nodule (area under the curve, 0.84 vs 0.84; P = .95) and for nodules identified at baseline (area under the curve, 0.85 vs 0.84; P = .17). When dichotomized by high/low risk, PanCan and Lung-RADS were discordant (P < .001). Participants with risk-gap nodules (n = 543) were distributed across Lung-RADS categories 2 through 4; 41 (8%) had invasive procedures with 23 (4%) having unnecessary invasive procedure use for solid, benign nodules.

INTERPRETATION: PanCan and Lung-RADS had similar overall accuracy for assessing cancer in screening-detected, solid lung nodules with evidence of discordance by subgroup. The existence of Lung-RADS category 4 nodules with a ≥ 3% to 5% PanCan risk may result in unnecessary procedures.