Variability Among Radiologists in Interpreting Low-Dose CT Lung Cancer Screens.

Abstract

Lung cancer screening (LCS) with low-dose CT (LDCT) reduces lung cancer mortality, but false-positive tests remain a concern. Radiologist variability in interpreting LDCT scans may contribute to higher false-positive rates (FPRs). Radiologists were identified from the LCS registry maintained by the ACR. Registry data were merged with Medicare files to create an LCS registry-Medicare database. Radiologists with ≥50 baseline or ≥50 postbaseline LDCT screens in the database were included in the analysis. The sensitivity and FPR were defined as the proportion of positive (Lung CT Screening Reporting and Data System 3-4) screens among those with and without a lung cancer diagnosis, respectively. Distributions, including median and interquartile range, of radiologists' FPRs were evaluated based on observed data and mixed-effects models, in which the radiologist was the random effect. Sensitivity was computed on an aggregate level based on radiologists' FPR quartiles. In all, 1,130 radiologists read ≥50 baseline screens, and 935 read ≥50 postbaseline screens. Median (interquartile range) FPR for baseline screens was 15.9% (11.3%-21.8%) and 16.3% (11.8%-21.8%) based on observed data and the mixed-effects model, respectively. Corresponding values for postbaseline screens were 8.5% (5.7%-12.8%) and 8.8% (6.3%-11.8%). Sensitivity increased significantly with radiologist FPR quartile for baseline and postbaseline screens. For baseline screens, radiologists in the lower two FPR quartiles had aggregate FPR and sensitivity of 10.9% and 89.6%, respectively, versus 24.1% and 94.4% for radiologists in the upper two quartiles. Variability in radiologists' FPRs with LDCT screening was high. Higher FPRs were associated with increased sensitivity rates.