2023 North America Conference on Lung Cancer (NACLC) - Posters and Abstracts-PP01.105 Melissa Santorelli Abstract

PP01.105 Melissa Santorelli Abstract

Pdf Summary

Researchers have developed a risk prediction model for the early detection of non-small cell lung cancer (NSCLC) using electronic health record (EHR) data. The study incorporated two EHR cohorts: the Mass General Brigham (MGB) Lung Cancer Mart, which includes lung cancer patients identified using a machine learning algorithm, and the MGB Biobank, which enrolls patients from a broader EHR population. The researchers used a pretrained knowledge graph to identify NSCLC associated features and generated risk scores through multivariate logistic regression. These risk scores were then integrated and calibrated using a Cox Proportional Hazard (CPH) model in a prospective cohort to build the final risk score. <br /><br />The study found that highly predictive features for early-stage NSCLC diagnosis included smoking, socioeconomic conditions, relevant lab results such as platelet mean volume and number of neutrophils in blood, and chronic lung diseases like chronic obstructive pulmonary diseases. The prediction model achieved an AUC of 0.80 for identifying one-year early-stage NSCLC risk, higher than a baseline model that only incorporated demographic and smoking information.<br /><br />The researchers also developed a prediction model for all-stage NSCLC diagnosis, which identified similar features but with different relative importance. Lab results were more influential in predicting all-stage NSCLC, while smoking features were less important.<br /><br />Overall, the developed risk prediction model demonstrated superior performance for the early detection of NSCLC compared to a model based solely on demographic and smoking information. This suggests the potential of incorporating EHR data for personalized cancer screening and early detection.

Keywords

risk prediction model

non-small cell lung cancer

EHR data

machine learning algorithm

lab results

chronic lung diseases

AUC

early-stage NSCLC risk

baseline model

personalized cancer screening