Development of EHR Based Phenotyping with High Dimensional Patient Information

2018 Fiscal Year Final Research Report

• Project/Area Number: 16K09161
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Medical and hospital managemen
• Research Institution: The University of Tokyo
• Principal Investigator: Kawazoe Yoshimasa 東京大学, 医学部附属病院, 特任准教授 (10621477)
• Project Period (FY): 2016-04-01 – 2019-03-31
• Keywords: EHR Phenotyping / 院内がん登録 / 電子的診療情報

Outline of Final Research Achievements

This study aimed to develop EHR phenotyping algorithms which describes the patient characteristics by high-dimensional features utilizing all items contained in physician’s order entry without manually selecting features. We constructed a dataset containing 100,313 patients, and evaluated the performance of machine learnings on the task of binarizing cancer and non-cancer cases, and the task of multi classification of cancer types. The former task showed better precision than the primary screening performed in hospital cancer registration, but the latter task does not seem to have sufficient accuracy. To improve the accuracy, it was considered to add surgical procedure codes and pathological diagnoses as features.

Free Research Field

医療情報学

Academic Significance and Societal Importance of the Research Achievements

病院における症例の登録業務は人手によるインテンシブな作業が必要であるため、機械学習等の技術を活用して人手による労力を軽減することが期待される。本研究は、日々の診療で発生する電子的診療情報を利用して、院内がん登録業務で行われるがん症例のスクリーニングとがん種別の分類を機械学習によって行った場合の精度を評価し、がん登録業務への応用可能性を検討したことが社会的意義としてあげられる。