1169P - Clustering of patients with lung neuroendocrine neoplasms using machine learning and its association with survival: A population based study from the U.S. SEER database

Background

Lung neuroendocrine neoplasms (NENs) have incidence of 0.2 to 2 /100,000/year. Grouped by morphology and stage alone there is some correlation with survival.

Methods

Patients with NENs diagnosed from 1975 to 2020 were extracted from U.S. Surveillance, Epidemiology, and End Results (SEER) database registries using SEER*Stat software. The total number of patients was 26,839 after excluding patients with missing data and patients who had large and small cell lung cancers. The patients were clustered into 3 clusters using unsupervised machine learning K-means clustering model. Clustering was evaluated using silhouette (S), Davies Boulden (DB) and Caliniski Harabasz (CH) scores. Clusters were visualized using principal component analysis (PCA) and T-distributed stochastic neighbor embedding (T-SNE). SHapley Additive exPlanations (SHAP) were used to identify the clinical features that are most useful in clustering. Five year survival was calculated using Cox-regression and Kaplan Meier (KM) statistics. A decision tree was developed to help clustering individual patients into one of the 3 clusters.

Results

The cohort median age was 66 years. There were more females (62%) than males (38%). Most patients (47%) had localized stage. Most patients (58%) had NET morphology. Patients were clustered into 3 clusters (0,1,2). S, DB and CH scores were 0.36, 1.01, and 17090 respectively. PCA and T-SNE visualizations showed distinct clusters with some overlap. SHAP showed that the 3 most important variables for clustering were stage, age, and morphology in order. Clusters description and survival are shown in the table. The 3 clusters were statistically different in survival based on cox regression, log rank test and KM statistics. Decision tree overall accuracy for clustering patients was 98.4%. Table: 1169P

Conclusions

Machine learning models could be helpful in clustering and prognosticating patients with NENs for survival.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

Hampshire Hospitals NHS Foundation Trust, UK.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.