2010 i2b2/VA challenge on concepts, assertions, and relations in clinical text

Area of Science:

• Natural Language Processing
• Clinical Informatics
• Biomedical Informatics

Background:

• The 2010 i2b2/VA Workshop addressed challenges in processing clinical records.
• Key tasks included concept extraction, assertion classification, and relation classification.
• An annotated corpus was provided by i2b2 and the VA.

Purpose of the Study:

• To evaluate systems for natural language processing (NLP) on clinical records.
• To explore methods for extracting medical concepts, assertions, and relations.
• To assess the effectiveness of machine learning and rule-based approaches.

Main Methods:

• Development of 22 systems for concept extraction, 21 for assertion classification, and 16 for relation classification.
• Utilized an annotated reference standard corpus for training and evaluation.
• Investigated the augmentation of machine learning with rule-based systems.

Main Results:

• Systems demonstrated that machine learning can be enhanced by rule-based approaches.
• Rule-based systems proved effective as either pre-processing or post-processing steps.
• Ensembles, unlabeled data, and external knowledge sources addressed data inadequacy.

Conclusions:

• Hybrid approaches combining machine learning and rule-based systems are effective for clinical NLP tasks.
• The workshop highlighted the potential of NLP in understanding clinical narratives.
• Strategies for overcoming limited training data were identified.