Establishment of a Stacked Ensemble Model for Risk Stratification after Allogeneic HSCT Utilizing Big Data

2023 Fiscal Year Final Research Report

• Project/Area Number: 22K21082
• Research Category: Grant-in-Aid for Research Activity Start-up
• Allocation Type: Multi-year Fund
• Review Section: 0908:Society medicine, nursing, and related fields
• Research Institution: Kyoto University
• Principal Investigator: Iwasaki Makoto 京都大学, 医学研究科, 医員 (40967527)
• Project Period (FY): 2022-08-31 – 2024-03-31
• Keywords: 造血幹細胞移植 / 移植片対宿主病 / 機械学習 / アンサンブル学習 / 生存時間解析 / 競合リスク

Outline of Final Research Achievements

We successfully developed a mid- to long-term prognosis prediction model incorporating these factors using stacked ensemble model. By taking into account post-transplant acute GVHD and its treatment, we demonstrated that the accuracy of the stacked ensemble model gradually improves over time, which underscores the significance of acute GVHD events in mid-to-long-term post-transplantation outcomes. Furthermore, to identify factors influencing prognosis, we used SHapley Additive Explanations, which revealed that severe acute GVHD is an important predictor of non-relapse mortality and overall survival, comparable to known prognostic factors such as sex matching, donor source, and pre-transplant disease status.

Free Research Field

血液学

Academic Significance and Societal Importance of the Research Achievements

造血幹細胞移植は現代においても、再発・移植後合併症で約半数は病気の治癒が難しく、予後を予測して適応を慎重に判断していく必要がある。特に、移植後1年を過ぎた中長期の予後の予測は、移植後の合併症や治療の影響も受けることから現代においても難しい課題の一つである。本研究において、機械学習・アンサンブル学習を用いて、移植後の因子も取り入れた予後の予測を行う事で、従来よりも優れたモデルが開発できる事を示しており、将来の造血幹細胞移植の適応判断や治療法選択に生きるものと考える。