Predicting cognitive decline from routinely collected data

2018 Fiscal Year Final Research Report

• Project/Area Number: 17H07421
• Research Category: Grant-in-Aid for Research Activity Start-up
• Allocation Type: Single-year Grants
• Research Field: Applied health science
• Research Institution: Fujita Health University (2018); National Cardiovascular Center Research Institute (2017)
• Principal Investigator: Ogata Soshiro 藤田医科大学, 医療科学部, 講師 (00805012)
• Project Period (FY): 2017-08-25 – 2019-03-31
• Keywords: 認知機能 / 予測モデル / 自治体保有情報 / ビッグデータ / 循環器病リスク要因 / 機械学習 / AI

Outline of Final Research Achievements

The present study conducted an epidemiological study collaborating with Nobeoka city, and collected data related to cognitive function from 480 community-dwelling people aged 75 years old and over. By utilizing a sparse modeling as a machine learning method, we developed a diagnostic prediction model of cognitive impairment (i.e., the Routine prediction model) only based on information routinely collected by Japanese cities. This had moderate predictability (AUC：0.70~0.71). Additionally, we developed an additional diagnostic prediction model (i.e., the Additional prediction model) by adding education level, category fluency test, and short-term memory test to the Routine prediction model. This had relatively high predictability (AUC：0.74~0.96). Thus, the present study developed the diagnostic prediction models of cognitive impairment with moderate to high predictability, which just required information collected routinely by cities and/or additional information collected easily.

Free Research Field

応用健康科学

Academic Significance and Societal Importance of the Research Achievements

認知症対策として認知機能低下を早期発見し予防につなげることが重要な戦略である。本研究で開発した地域在住後期高齢者を対象とした認知機能障害予測モデルは、日本の市町村が既に業務の一環として収集している情報を主として使用している。そのため、市町村は本研究で開発した予測モデルを低コストで今すぐ運用することが可能である。以上のことから、本研究は認知症早期発見システム構築に貢献すると考える。