Big data analysis of outcomes in non-small cell lung cancer at a regional cancer institute with a coordinated precision medicine program.

Abstract

e15161 Background: Non-small cell lung cancer (NSCLC) accounts for approximately 85% of all lung cancer cases, and the 5-year survival rate for patients with advanced-stage disease is low. These tumors often harbor specific actionable biomarkers that can be identified using advanced genomic technologies and have been shown to improve survival. Clinical care gaps, however, can adversely impact the application of test results. At our institution, several operational changes, such as the implementation of standardized reflex testing protocols and expert review of biomarker testing results, have created faster pathways toward targeted treatments for NSCLC patients. This study aims to apply a data analytics approach to the NSCLC population to identify the factors that impact survival in the era of precision medicine. Methods: To evaluate the outcomes of patients with NSCLC, we retrospectively joined and analyzed three different data sources: 1) Electronic Health Records (EPIC) of lung cancer patient encounters from 2015 to 2024; 2) Internal genomics database (cBioPortal); 3) Cancer registry data with patients survival outcome information. By merging these three disparate data sources, we could comprehensively study the NSCLC patient population at our hospital, including their clinical characteristics, genomic profiles, and long-term outcomes utilizing Kaplan-Meir curves and XGBoost classifier to assess patient survival. Results: The study comprised 2,187 patients diagnosed with NSCLC from 2015-2024—48.56% of the patients presented with stage III or IV disease. Kaplan-Meier analysis showed a median overall survival of 22 months and a 12-month survival rate of 55% for distant stage, 71% for regional stage, and 84% for local stage patients. Patients were followed for a median of 24 months after diagnosis, with last follow-up ranging from 1 to 96 months. At the end of the follow-up, 57% of the patients were censored due to death (47.2% due to an unknown cause, 45.3% died from cancer, and the rest died due to other causes). Among the cohort diagnosed at our hospital, 54% underwent molecular profiling. Pathologist-initiated reflex testing reduced overall pathology-to-test results times by 12 days compared to standard oncologist ordering. Profiling before treatment initiation and positive guideline-driven biomarkers were associated with improved survival by the XGBoost classifier. Conclusions: Data analytics is a promising approach to measuring, monitoring, and improving cancer care. We showed in a highly coordinated program, profiling before treatment initiation and having a positive biomarker are associated with improved outcomes in a large and heterogeneous cohort of NSCLC patients. We also document that reflex testing improves overall testing times. The results herein may help guide improved delivery of precision medicine approaches for NSCLC.