Characteristics of new solid nodules detected in incidence screening rounds of low-dose CT lung cancer screening: the NELSON study

Purpose New nodules after baseline are regularly found in low-dose CT lung cancer screening and have a high lung cancer probability. It is unknown whether morphological and location characteristics can improve new nodule risk stratification by size.
Methods Solid non-calcified nodules detected during incidence screening rounds of the randomised controlled Dutch-Belgian lung cancer screening (NELSON) trial and registered as new or previously below detection limit (15 mm3) were included. A multivariate logistic regression analysis with lung cancer as outcome was performed, including previously established volume cut-offs (<30 mm3, 30–<200 mm3 and ≥200 mm3) and nodule characteristics (location, distribution, shape, margin and visibility <15 mm3 in retrospect).
Results Overall, 1280 new nodules were included with 73 (6%) being lung cancer. Of nodules ≥30 mm3 at detection and visible <15 mm3 in retrospect, 22% (6/27) were lung cancer. Discrimination based on volume cut-offs (area under the receiver operating characteristic curve (AUC): 0.80, 95% CI 0.75 to 0.84) and continuous volume (AUC: 0.82, 95% CI 0.77 to 0.87) was similar. After adjustment for volume cut-offs, only location in the right upper lobe (OR 2.0, P=0.012), central distribution (OR 2.4, P=0.001) and visibility <15 mm3 in retrospect (OR 4.7, P=0.003) remained significant predictors for lung cancer. The Hosmer-Lemeshow test (P=0.75) and assessment of bootstrap calibration curves indicated adequate model fit. Discrimination based on the continuous model probability (AUC: 0.85, 95% CI 0.81 to 0.89) was superior to volume cut-offs alone, but when stratified into three risk groups (AUC: 0.82, 95% CI 0.78 to 0.86), discrimination was similar.
Conclusion Contrary to morphological nodule characteristics, growth-independent characteristics may further improve volume-based new nodule lung cancer prediction, but in a three-category stratification approach, this is limited