project 7

Annotation of ECOG PS from Unstructured Oncology Notes and Survivability Analysis

This project aims at identifing performance status (PS) labels in unstructured clinical notes using a text-based search, trained a CNN model and a transformer-based model to predict the ECOG PS, and evaluated the correlation between ECOG PS and survival outcomes.

This project achieved the following:

  • Developed a model with 95.5% accuracy, and found strong correlation between ECOG PS and survival outcomes.
  • Worked with two medical oncologists and one data scientist at Dana-Farber.
  • Drafted a research paper.