JMIR Publications

ABSTRACT

Background:

Prognosis, diagnosis and treatment of many genetic disorders and familial diseases significantly improves if family history of the patient is known. Such information is often written in free-text of clinical notes.

Objective:

Our aim is to develop automated methods that enable access to family history data through natural language processing.

Methods:

We use information extraction using transformers for extracting disease mentions from notes. We also experiment with rule-based methods for family member extraction from text as well as coreference resolution techniques. We provide a thorough error analysis of contributing factors that affect such information extraction system.

Results:

Our experiments show that a combination of domain-adaptive pretraining together with intermediate-task pretraining method achieves a F1 Score of 81.63\% for extraction of diseases and family members from notes when tested on a public shared task dataset by National NLP Clinical Challenges. In comparison, the 2019 n2c2/OHNLP Shared-Task the median F1 score of all the 17 participating teams is 76.59\%.

Conclusions:

Our approach, which leverages state-of-the-art named entity recognition for disease mention detection, coupled with a hybrid method for family member mention detection, achieved effectiveness close to top three systems participating in the 2019 n2c2 family history extraction challenge, with only the top system outperforming it convincingly in terms of precision.