ROC Curves for Low-Dose CT in the National Lung Screening Trial
Authors:
Open Access
- 30 August 2013
- journal article
- research article
- Published by SAGE Publications in Journal of Medical Screening
- Vol. 20 (3) , 165-168
Abstract
The National Lung Screening Trial (NLST) reported a 20% reduction in lung cancer specific mortality using low-dose chest CT (LDCT) compared with chest radiograph (CXR) screening. The high number of false positive screens with LDCT (around 25%) raises concerns. NLST radiologists reported LDCT screens as either positive or not positive, based primarily on the presence of a 4+ mm non-calcified lung nodule (NCN). They did not explicitly record a propensity score for lung cancer. However, by using maximum NCN size, or alternatively, radiologists’ recommendations for diagnostic follow-up categorized hierarchically, surrogate propensity scores (PSSZ and PSFR) were created. These scores were then used to compute ROC curves, which determine possible operating points of sensitivity versus false positive rate (1-Specificity). The area under the ROC curve (AUC) was 0.934 and 0.928 for PSFR and PSSZ, respectively; the former was significantly greater than the latter. With the NLST definition of a positive screen, sensitivity and specificity of LDCT was 93.1% and 76.5%, respectively. With cutoffs based on PSFR, a specificity of 92.4% could be achieved while only lowering sensitivity to 86.9%. Radiologists using LDCT have good predictive ability; the optimal operating point for sensitivity and specificity remains to be determined.Keywords
This publication has 6 references indexed in Scilit:
- National Lung Screening Trial: Variability in Nodule Detection Rates in Chest CT StudiesRadiology, 2013
- Reduced Lung-Cancer Mortality with Low-Dose Computed Tomographic ScreeningNew England Journal of Medicine, 2011
- The National Lung Screening Trial: Overview and Study DesignRadiology, 2011
- Computed tomography screening for lung cancer: Review of screening principles and update on current statusCancer, 2007
- Diagnostic Performance of Digital versus Film Mammography for Breast-Cancer ScreeningNew England Journal of Medicine, 2005
- A non‐parametric method for the comparison of partial areas under ROC curves and its application to large health care data setsStatistics in Medicine, 2002