| 研究課題/領域番号 |
18H03336
|
|---|
| 研究機関 | 聖路加国際大学 |
|---|
研究代表者 |
ウォン スイー 聖路加国際大学, 専門職大学院公衆衛生学研究科(公衆衛生大学院), 准教授 (70791599)
|
|---|
| 研究分担者 |
林 邦好 聖路加国際大学, 専門職大学院公衆衛生学研究科(公衆衛生大学院), 講師 (00793217)
笹野 遼平 名古屋大学, 情報学研究科, 准教授 (70603918)
|
|---|
| 研究期間 (年度) |
2018-04-01 – 2021-03-31
|
| キーワード | Patient Safety / Incident reports / Named entity / Adverse drug events / Deep learning / Natural language processing |
|---|
| 研究実績の概要 |
In the last fiscal year, the research team focused on the development of annotation guideline for medication errors (Task 1) and designed named entity recognition (NER) model development framework (Task 2). Our team has performed narrative review on relevant state of science literature, prepared the relevant incident data and information (IRB approved), discussed the proposed guideline with field experts. In terms of the design of NER model, we constructed NER program using biLTSM and BERT methods, explored the model performance using sample annotated data, and carried out intensive experiments. These tasks have established a solid foundation for the team for developing the prototype NER system development (Task 3) in year 2 and 3.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
We closely followed the proposed research schedule, fully accomplished the outlined research tasks and reported progress to team members in regular whole group research meetings. Together with two Co-Is (kenkyu-buntansha) - Dr. Sasano and Dr. Hayashi, Dr. Wong led project team members and research assistants with medicine, pharmacy and computer science backgrounds to achieve the above research tasks. The team has been working closely with other collaborators within Japan and has invited international experts to facilitate global exchange. Some of our ongoing research outcomes have been accepted in international conferences (such as Medinfo 2019, CSHI 2019 etc) and our group disseminated our ongoing outcomes at AI acceleration meeting at Ministry of Health, Labour and Welfare.
|
| 今後の研究の推進方策 |
In the next fiscal year, the research team will focus on the validation of taxonomy for medication error for incident reports (Task 1) and finalizing named entity recognition (NER) model framework for incident reports learning (Task 2). Using our incident report data, our team will validate the annotation guideline and create gold standard annotated incident reports data for NER model training and validation. We will evaluate multiple promising NER program settings that are suitable for incident report learning and carry out extensive experiments to enhance model performance. In the latter part of this fiscal year, we will design and develop prototype NER system development (Task 3) by incorporating the NER model prediction as interactive outputs to aid incident reporting.
|
