Development and performance of a multiomics lung cancer screening blood test

Lung cancer is the leading cause of cancer death in the US. Early detection is crucial for improving survival rates, yet only 18% of eligible high-risk adults (aged 50-80 with ≥20 pack-years of cigarette smoking history) are up to date with recommended annual low-dose computed tomography screening. Non-invasive blood tests could increase screening participation. Here, we evaluate a multiomics approach for lung cancer detection in patients in the intended use population (IUP). Our test uses base-resolution methylation sequencing of circulating cell-free DNA (cfDNA), as previously described, and plasma protein immunoassays. An artificial intelligence/machine learning classifier was trained on tissue (n = 136) and plasma (n = 6,716) samples. Accuracy was evaluated in a cohort of 673 plasma samples, including lung cancer cases (n = 363) and cancer-negative controls (n = 310) from the IUP. This cohort encompassed all cancer stages and three major subtypes - adenocarcinoma, squamous cell carcinoma and small-cell lung cancer (SCLC) - and its age and pack-year smoking history distributions reflected the IUP. Two specificities were considered: 50% (prioritizing sensitivity) and 75% (prioritizing specificity). Reported sensitivities were adjusted for stage and subtype to address differences between the evaluation cohort and literature-reported distributions for the IUP. (Adenocarcinoma and squamous data were weighted per stage and subtype; due to small sample size, SCLC data were weighted only per subtype.) 95% confidence intervals (CIs) were computed via Wilson's method. The multiomics test had an IUP-adjusted sensitivity of 90.7% (CI: 86.8 - 93.7%) at 50% specificity and 80.4% (75.4 - 84.7%) at 75% specificity. Results for a methylation-only test were 85.8% (81.3 - 89.5%) and 78.2% (73.1 - 82.8%), respectively. The multiomics test showed higher sensitivity at both specificities but the differences were not statistically significant. At 50% specificity, the multiomics test had adjusted sensitivity for stages I through IV of 77.7% (73.3 - 89.0%), 95.2% (85.0 - 98.1%), 98.8% (93.6 - 99.7%) and 97.1% (90.7 - 98.9%), respectively. Corresponding results at 75% specificity were 63.5% (51.9 - 71.6%), 89.8% (76.6 - 94.1%), 90.9% (85.0 - 96.3%) and 95.9% (89.8 - 98.6%). At 50% specificity, the multiomics test's adjusted sensitivity for adenocarcinoma and squamous cell carcinoma was 88.7% (81.4 - 92.0%) and 91.1% (84.3 - 96.1%), respectively; nominal sensitivity for SCLC was 97.1% (85.1 - 99.5%). Corresponding results at 75% specificity were 73.0% (63.1 - 77.6%), 88.6% (81.6 - 94.6%) and 91.2% (77.0 - 97.0%). Our multiomics platform demonstrated promising initial performance for blood-based lung cancer detection. The complementary nature of cfDNA methylation and plasma protein may enhance performance relative to a methylation-only approach, and this will be further assessed in a future study with a previously unseen evaluation cohort.