Background
Advances in targeted therapy and immunotherapy for lung cancer improves patient outcomes but requires molecular testing of cancer samples. Successful biomarker testing depends on many factors; quality improvement initiatives require access to real-world data. Natural Language Processing (NLP) and Artificial Intelligence (AI) technology automate data abstraction from unstructured electronic health records (EHR), eliminating the need for resource-intensive manual chart review.
Methods
The DARWEN™ automated data abstraction platform (Pentavere, Toronto, Canada) was used to extract data from EHR of advanced lung cancer patients diagnosed and treated at Princess Margaret Cancer Centre (Toronto, Canada) between 01/2015 and 05/2018. Demographics, tumour characteristics including EGFR, ALK and PD-L1 status, treatment and survival data were extracted.
Results
Of 1210 advanced lung cancer patients, 615 had accessible electronic pathology records and were reviewed by DARWEN™. Of these, 246 had non-squamous NSCLC and received systemic therapy (analysis set). Complete biomarker testing was performed in 79% (by minimum standard at time of diagnosis). Never smokers (p=0.001) and patients with better performance status (p=0.014) were more likely to have biomarker testing. After initiation of routine PDL1 testing (09/2016), the rate of PD-L1 testing increased 28% to only 55% and remained high for EGFR and ALK (93% each). Successful testing resulted in more patients accessing targeted therapy (p<0.001) or immunotherapy (p=0.002) as initial treatment, and fewer patients receiving chemotherapy first-line (p<0.001). One-year survival rates were similar pre/post introduction of routine PDL1 testing, 70% (95% CI: 62-77%) and 76% (95% CI: 67-83%), respectively.
Conclusions
NLP and AI technologies like DARWEN™ give clinicians access to previously unavailable real-world data. Herein, we identified opportunities to improve molecular testing and patient outcomes in advanced lung cancer and personalized therapy. Novel technologies that facilitate
