Research on development of prediction method of fall by large scale clinical nursing data and machine learning

2018 Fiscal Year Final Research Report

• Project/Area Number: 16K20977
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Clinical nursing; Medical and hospital managemen
• Research Institution: The University of Tokyo
• Principal Investigator: Yokota Shinichiroh 東京大学, 医学部附属病院, 助教 (90599490)
• Project Period (FY): 2016-04-01 – 2019-03-31
• Keywords: 転倒・転落 / 電子カルテ / 機械学習 / 自然言語処理 / 医療安全管理

Outline of Final Research Achievements

A fall risk assessment model made by machine learning method shows 64.9% sensitivity and 69.6% specificity. Though the result was equivalent to previous studies, it required 40 days of calculation for learning and verification, therefore it was not efficient. Next, we investigated the influence of the implementation of the fall risk assessment tool by comparing the pre- and post-implementation periods. The fall probability of inpatients decreased in the post-implementation period and the fall probability of inpatients was equivalent between the tool-used patients and non-tool-used patients in the post implementation period. Moreover, we investigated the fall reports experimentally by the machine learning method. The results showed that the recognition of the fall related concept in the clinical field might be vague.
All researches were carried out as a retrospective observation study using electronic medical record data.

Free Research Field

看護情報学

Academic Significance and Societal Importance of the Research Achievements

入院患者の転倒は多くの医療機関で入院中の医療インシデントの最も多いもののひとつで、外傷や死亡を引き起こす事があり、患者のQuality of Lifeの向上や医療資源の適切利用の観点からも、喫緊に解決すべき社会的課題である。国内外で患者転倒リスク評価手法はいくつも開発されているが、多くは開発用データが不十分であり、また多くは臨床での実際の転倒発生予防に効果があるかどうかが不明という課題がある。
今回の一連の研究は、医療リアルワールドデータをもとに機械学習を含む人工知能関連手法を活用した判定ロジックの開発、実装評価等を一連としてして実施している点で、学術的にも社会的にも意義があると考える。