Cancer risk assessment in patients with persistent pulmonary nodules and its correlation with cancer-free survival.

Abstract

8569 Background: With the wide use of CT for lung cancer screening and diagnosis, detection of pulmonary nodules increases drastically. Brock malignancy risk scoring is a validated risk prediction model for distinguishing malignant nodules, but it only provides a snapshot without incorporating the dynamic changes of lung nodules. Our study tested the performance of Brock malignancy risk scoring system in patients with persistent lung nodules. Methods: We prospectively studied a cohort of 304 patients with persistent lung nodules (at least 2 CT scans, 3 months apart with no evidence of shrinkage) who were under management at MD Anderson Cancer Center from 11/28/2018 to 12/14/2022. These lung nodules were assessed by radiologists with Brock full model. These patients were followed up routinely and subjected to biopsy as determined by treating physicians. The area under the receiver operating characteristic curve (AUC) and the optimal cut-off of Brock model was studied. Additionally, we studied another cohort with 130 patients with histologically confirmed lung cancer. We retrospectively reviewed the CT or PET/CT scans and assessed the corresponding persistent lung nodules as defined above prior to the cancer diagnosis. We explored the correlations among nodule characteristics, demographic factors and Brock cancer risk scores. Cox proportional hazards model was built for multivariate analysis. Results: The median follow-up time for the prospective cohort was 337 days and 40 of the 304 patients (13.16%) were diagnosed with lung cancer with a median lung cancer-free survival of 228 days. The mean risk score was 24.20% (0.12%-62.86%) for histologically confirmed malignant vs 11.01% (0.07%-61.84%) for the remaining lung nodules (P < 0.001). Of note, 4 of 46 (8.70%) patients with persistent lung nodules of risk scores between 5%-10% and 8 of 134 (5.97%) patients with risk score < 5% were diagnosed with lung cancer. The AUC for Brock model was 0.72 and the optimal cut-off value is 10.64% (sensitivity: 0.702, specificity:0.677). Among the retrospective cohort of 130 lung cancer patients (82.3% adenocarcinoma, 13.1% squamous cell carcinoma and 4.6% others), the predicted risk score ranged from 0.09% to 85.82%, including 33.85% of patients with risk score < 10% and 26.15% of patients with predicted risk score < 5%. The risk score was not correlated with age, sex, race, ethnicity, smoking history, family history of lung cancer, emphysema, nodule types or locations. The low-risk patients had a longer median cancer-free time (P = 0.001). Conclusions: Persistency is an important risk factor for malignant lung nodules. Using Brock criteria, a substantial proportion of lung nodules with true malignant potential can be overlooked because of predicted “low risk”. Improved prediction models incorporating the dynamic changes of lung nodules are warranted to guide early diagnosis of lung cancer.