2023 North America Conference on Lung Cancer (NACLC) - Posters and Abstracts-PP01.105 (Poster) Early detection of non–small-cell lung cancer using electronic health record data

Pdf Summary

A recent study aimed to identify characteristics associated with early-stage diagnosis of non-small cell lung cancer (NSCLC) and build a prediction model for early detection using electronic health records (EHR) data. The study utilized two EHR cohorts: the Mass General Brigham (MGB) Lung Cancer Mart, which included lung cancer patients identified using a machine learning algorithm, and the MGB Biobank, a non-disease-specific population that collects blood samples from patients at MGB clinics. <br /><br />Three study designs were employed: the self-controlled design, the case-control design, and the prospective cohort design. Risk models were built using the data from these designs, and ensemble learning was used to integrate the risk scores. A knowledge graph pre-trained on MGB and Veteran Affair (VA) data was used to identify EHR-derived features related to NSCLC. Univariate association screening and input from domain experts were used to screen the features.<br /><br />The study found a set of 104 EHR-derived features predictive of early NSCLC diagnosis (stages I-III). These features included smoking, socioeconomic conditions, relevant lab results, and chronic lung diseases. The prediction model based on these features demonstrated superior performance for early detection compared to a model that only used demographic and smoking information. The model achieved an overall area under the curve (AUC) of 0.80 for predicting one-year early-stage risk.<br /><br />The results suggest that incorporating EHR data into personalized cancer screening can improve early detection of lung cancer. The study recommends validating the model's performance in an external population to facilitate its translation into clinical practice. The research was sponsored by Merck Sharp & Dohme LLC and presented at the IASLC North America Conference on Lung Cancer.

Asset Subtitle

Melissa Santorelli

Keywords

non-small cell lung cancer

early-stage diagnosis

prediction model

electronic health records

EHR data

smoking

lab results

chronic lung diseases

area under the curve

personalized cancer screening