The Research Question The NELSON trial showed that LDCT screening reduces lung cancer mortality, but the trial itself cannot directly measure how sensitive the screening protocol is for detecting each cancer type at each stage. This modeling study used MISCAN-Lung, a validated microsimulation model, to estimate these stage- and histology-specific detection sensitivities for the NELSON protocol.
Why Sensitivity Matters by Subtype Lung adenocarcinoma and squamous cell carcinoma grow and progress differently. Stage IA adenocarcinoma, the most curable form, is also the slowest-growing and most likely to be missed at a given screening round. Understanding which subtypes are detected least reliably helps explain why screening benefits are uneven and guides protocol refinements.
Key Finding The NELSON protocol had a baseline sensitivity of only 41% for stage IA adenocarcinoma in the first screening round. For repeat screening rounds, this rose to 71%. This means that even in the best-performing lung cancer screening protocol in the world, the most curable disease stage was missed in the majority of first-round screens.
Implications for Protocol Design These sensitivity estimates have direct implications for screening frequency, nodule size thresholds, and the potential role of supplementary biomarkers. If early adenocarcinoma is being missed at such high rates, protocol modifications or adjunct technologies that improve detection of slow-growing nodules could substantially increase the mortality benefit.
The NELSON Trial NELSON (Dutch-Belgian Randomized Lung Cancer Screening Trial) was a European randomized controlled trial comparing LDCT screening to no screening in 15,792 high-risk participants. Using a volume-based nodule management algorithm, NELSON demonstrated a 24% mortality reduction in male participants over 10 years - the strongest evidence for CT lung cancer screening outside of NLST.
Volume-Based Nodule Management Unlike the NLST, which used diameter-based thresholds, NELSON classified nodules by volume and volume-doubling time. Nodules below 100 mm3 were managed conservatively, those between 100-300 mm3 triggered repeat CT, and those above 300 mm3 or with volume doubling times under 400 days were referred for further workup. This algorithm was intended to reduce false positives.
MISCAN-Lung Model MISCAN (MIcrosimulation SCreening ANalysis) is a natural history model of lung cancer that simulates individual patient trajectories from tumor initiation through clinical detection or screening detection to treatment and death. By calibrating the model to observed NELSON outcomes and Dutch population data, researchers can estimate quantities not directly measurable in the trial - including stage-specific sensitivities.
Why Modeling Is Needed In a randomized trial, you observe who gets diagnosed and when, but you cannot directly measure what fraction of existing cancers were present but undetected at each screen. Microsimulation fills this gap by estimating the tumor's biological state at each screen visit, allowing calculation of the conditional probability of detection given the tumor's true stage and histology.
Model Calibration to NELSON Data MISCAN-Lung was calibrated to reproduce the observed NELSON trial outcomes including cancer incidence rates, stage distribution at detection, and interval cancer rates. The calibration process adjusted model parameters (tumor growth rates, preclinical detectable phase durations) until simulated outputs matched observed data within acceptable confidence bounds.
Histology-Specific Natural History The model distinguished between adenocarcinoma, squamous cell carcinoma, and other NSCLC subtypes, each with different natural history parameters - particularly tumor volume-doubling times. Adenocarcinoma was modeled with a longer sojourn time (time spent in the preclinical detectable phase) compared to squamous cell carcinoma, reflecting its characteristically slower growth.
Sensitivity Definition Sensitivity was defined as the probability that a lung cancer present at the time of a screen was detected by that screen. Baseline sensitivity referred to the first screening round, while repeat-round sensitivity reflected improved detection due to prior interval growth and volume-doubling time data informing the workup decision.
Stage-Specific Analysis Sensitivities were estimated separately for stage IA, IB, II, III, and IV, and for each major histological subtype. This granularity allows identification of which specific cancer phenotypes are driving the detection gaps and which protocol changes would most efficiently improve the overall mortality benefit.
Adenocarcinoma Stage IA Sensitivity The baseline sensitivity for stage IA adenocarcinoma was 41% in the first screening round, meaning that nearly 6 in 10 existing stage IA adenocarcinomas were not detected. In repeat rounds, this improved to 71% - still missing nearly 1 in 3. The difficulty arises because stage IA adenocarcinomas are often small, slow-growing, and may present as ground-glass opacities that are difficult to characterize on CT.
Squamous Cell Carcinoma Comparison Squamous cell carcinoma showed higher detection sensitivity at equivalent stages compared to adenocarcinoma. This is consistent with squamous carcinoma's typically faster growth rates and more solid, centrally located presentation, which makes it both more likely to have grown to a detectable size and more conspicuous when present.
Advanced Stage Sensitivity Stage III and IV cancers showed higher sensitivity than early-stage disease at the first screen, reflecting their larger size and more obvious imaging appearance. However, detecting advanced-stage disease is less clinically valuable since these patients cannot benefit from curative surgical resection.
Interval Cancer Contribution The model estimated a meaningful proportion of lung cancers as interval cancers - tumors that were either not present or below the detection threshold at screening but became symptomatic between rounds. Adenocarcinoma interval cancers were more likely to be stage IA at screen but progress to later stage before the next scheduled screen, highlighting the cost of the volume-based protocol's conservative management of small nodules.
Threshold Calibration The NELSON protocol's volume thresholds were set to balance sensitivity against false positives. These modeling results suggest the thresholds may be too conservative for slow-growing adenocarcinoma, where the cost of a missed early-stage cancer is high relative to the harm of a false positive. Lowering volume thresholds specifically for ground-glass opacity nodules warrants evaluation.
Screening Frequency Annual screening (as in NLST) versus biennial (as used in later NELSON rounds) has a direct impact on interval cancer rates for slow-growing tumors. Increasing frequency for participants in whom prior rounds identified indeterminate or slow-growing nodules could partially compensate for the 41% first-round sensitivity gap.
Supplementary Biomarkers Given the inherent limits of CT morphology alone for sub-centimeter ground-glass nodules, supplementary biomarkers - blood-based protein or ctDNA markers, bronchial genomic risk scores, or PET - may help differentiate malignant from benign nodules and reduce both missed cancers and unnecessary follow-up.
Volume-Doubling Time Value The repeat-round sensitivity improvement (41% to 71%) largely comes from volume-doubling time data accumulated over multiple screens. This demonstrates that multi-round screening protocols provide substantially more value than single-round programs, supporting the case for sustained engagement of participants over multiple screening cycles.