RWD112 Can ML-Extracted Variables Reproduce Real World Comparative Effectiveness Results from Expert-Abstracted Data? A Case Study in Metastatic Non-Small Cell Lung Cancer Treatment

In generating real world data (RWD), machine learning (ML) extraction of clinical characteristics from unstructured text (e.g. clinical notes) in electronic health records (EHRs) is more cost-effective and scalable than manual abstraction. Proper evaluation that goes beyond standard ML metrics is needed to determine whether ML-extracted variables are fit for research use [1]. This study evaluates reproducibility of scientific conclusions when using expert-abstracted versus ML-extracted data in comparing the effectiveness on real-world overall survival (rwOS) of bevacizumab-carboplatin-paclitaxel (BCP) versus carboplatin-paclitaxel (CP) for first-line treatment of non-squamous metastatic non-small cell lung cancer (mNSCLC).