JMIR Publications

ABSTRACT

Background:

Electronic Health Records (EHR) store large amounts of patient clinical data. Despite the efforts to structure patient data, clinical notes containing rich patient information remain stored as free text, greatly limiting its exploitation. This includes family history, which is highly relevant for applications such as diagnosis and prognosis.

Objective:

The main goal of this work was to develop automatic strategies for annotating family history information in clinical notes, focusing not only on the extraction of relevant entities such as family members and disease mentions, but also on the extraction of relations between the identified entities.

Methods:

This work extends a previous contribution for the 2019 n2c2 challenge track on family history extraction by 1) improving a previously developed rule-based engine, 2) using Deep Learning approaches for the extraction of entities from clinical notes and 3) combining both approaches in a hybrid end-to-end system capable of successfully extracting family member and observation entities and the relations between those entities. Furthermore, this paper analyses the impact of factors such as the use of external resources and different types of embeddings in the performance of Deep Learning models.

Results:

The approaches developed were evaluated in a first task regarding entity extraction and in a second task concerning relation extraction. The proposed Deep Learning approach improved observation extraction, obtaining F1-scores of 0.869 and 0.791 in the train and test sets, respectively. However, Deep Learning approaches showed limitations in the extraction of family members. The rule-based engine was adjusted to have higher generalizing capability, and managed family member extraction F1-scores of 0.882 and 0.809 in the train and test sets, respectively. The resulting hybrid system obtained F1-scores of 0.874 and 0.798 in the train and test sets, respectively. For the second task, the original evaluator was adjusted to perform a more exact evaluation than the original one, and the hybrid system obtained F1-scores of 0.6480 and 0.5082 in the train and test sets, respectively.

Conclusions:

We evaluated the impact of several factors in the performance of deep learning models, and present an end-to-end system for extracting family history information from clinical notes, which can help in the structuring and reuse of this type of information.