New research funding from PCORI
This new research contract will support collaborative efforts between MUSC and Vanderbilt University (Glenn Gobbel and his team).
Our general objective is to develop and assess new Natural Language Processing (NLP) methods to improve the accuracy, validity and efficiency of information extraction from Electronic Health Record (EHR) text and validate these new methods with a patient-centered outcomes research (PCOR) use case: extracting and deriving the Glasgow coma scale (GCS). We propose to address gaps in methods allowing for efficient annotation of clinical text and reuse of existing annotated clinical text and gaps in methods to extract information from clinical text with accurate local context and attributes (e.g., clinical note section, negation). To overcome these gaps, we propose to:
- Specific Aim 1: Develop and evaluate new assisted annotation methods to minimize clinical note annotation burdens with novel active learning algorithms that leverage feature expansion and semi-supervised learning.
- Specific Aim 2: Improve and develop new methods for efficient use of existing text annotations and use of non- annotated text data. This aim will implement novel applications of domain adaptation.
- Specific Aim 3: Develop and assess new generalizable methods for analysis of the local context of information extracted from text. This context includes note sections and combines it with information polarity (i.e., negation).