Performance of volume and diameter thresholds in malignancy prediction of solid nodules in lung cancer screening

Abstract

Background: Prospective validation and comparison of the performance of nodule management protocols is limited. The aim of this study was to examine the performance of size and risk thresholds for assessing malignancy in solid nodules at baseline low-dose CT (LDCT) in a lung cancer screening (LCS) programme.

Methods: This was an observational study using data from the SUMMIT Study, a prospective longitudinal study investigating LDCT for LCS. Participants were 55-77 years old and met either the United States Preventative Services Task Force (2013) criteria or had a PLCO_m2012 risk of ≥1.3%. LDCTs were reported using computer-aided detection software (Veolity, MeVIS) with semiautomated volumetry. Cancer outcomes were reported for solid nodules reported at baseline CT, with participants represented by the single largest solid nodule where more than one was present. Malignancy risk was stratified by long-axis diameter and volume using predefined size thresholds taken from British Thoracic Society and European Position statement guidelines: a 5/6 mm long axis diameter or 80/100 mm³ volume 'rule out' thresholds for low-risk nodules and ≥300 mm³ or ≥8 mm diameter with or without Brock score ≥10% 'Rule in' thresholds for high-risk nodules.Pearson's χ² test was used to calculate statistical significance for nominal variables, McNemar's test for comparison of sensitivity/specificity and DeLong' test for comparison of areas under the receiver operating characteristic curve (AUROC). Optimal thresholds were determined with Youden's J statistic. Net benefit calculations were undertaken to compare the existing thresholds with 95% CIs calculated by bootstrap sampling.

Results: 11 355 participants were included. Crude risk of malignancy in solid nodules at baseline LDCT was 3.8% (228/5929). Risk of malignancy in solid nodules <6 mm long-axis diameter or <100 mm³ volume was equivalent to that in participants with no nodules at baseline LDCT (0.88% and 0.84% vs 0.77%, p=0.4600 and p=0.7932, respectively). A <80 mm³ volume and <5 mm diameter 'rule out' threshold achieved sensitivity 86.8% and 93.4%, specificity 65.4% and 24.64%, and negative predictive value (NPV) 99.2% and 98.9%, respectively. The <80 mm³ volume threshold encompassed 63.3% of participants with a baseline solid nodule compared with 24.0% by the <5 mm diameter threshold.For nodules ≥8 mm diameter, the addition of a risk score (Brock ≥10%) was associated with a significant net benefit when compared with using size threshold alone by net effect analysis (31.24; 95% CI 26.19 to 35.89).

Conclusions: Solid nodules <100 mm³ or <6 mm diameter are not associated with increased risk of lung cancer compared with participants with no nodules at baseline LDCT. Volumetric rule-out thresholds achieve equivalent NPV to long-axis diameter thresholds while encompassing significantly more participants, reducing the number of interval scans required.

Keywords: Imaging/CT MRI etc; Lung Cancer.

Figure 1. Data presented at a per-participant level. +Completed baseline LDCT in SUMMIT prior to April 2020. *Other nodule types reported in SUMMIT include subsolid, endobronchial and indeterminate focal consolidation. ^ⱡFor the volumetric analysis, the 144 solid nodules with no reliable segmentation were included but were categorised by diameter thresholds of 6 mm for ‘rule out’ and 8 mm for ‘rule in’, reflecting how such nodules would be managed under BTS guidelines. LDCT, low-dose CT.

Figure 2. Receiver-operating characteristic curves for nodule diameter and volume. Area under the curve for diameter 0.822 (95% CI 0.787 to 0.857) and volume 0.844 (95% CI 0.807 to 0.8820), p=0.3852.

Figure 3. Receiver-operator characteristic curve for Brock model in predicting malignancy risk in solid nodules of >8 mm diameter. Area under the curve =0.898 (95% CI 0.8764 to 0.9196).