Establishment of a predictive model for GVHD-free, relapse-free survival after allogeneic HSCT using ensemble learning

• DOI: 10.1182/bloodadvances.2021005800
• Publication Date: 2021-12-21T22:01:23Z
• Authors: Makoto Iwasaki; Junya Kanda; Yasuyuki Arai; Tadakazu Kondo; Takayuki Ishikawa; Yasunori Ueda; Kazunori Imada; Takashi Akasaka; Akihito Yonezawa; Kazuhiro Yago; Masaharu Nohgawa; Naoyuki Anzai; Toshinori Moriguchi; Toshiyuki Kitano; Mitsuru Itoh; Nobuyoshi Arima; Tomoharu Takeoka; Mitsumasa Watanabe; Hirokazu Hirata; Kosuke Asagoe; Isao Miyatsuka; Le My An; Masanori Miyanishi; Akifumi Takaori-K...
• Keywords: Ensemble Learning

Graft-versus-host disease-free, relapse-free survival (GRFS) is a useful composite end point that measures without relapse or significant morbidity after allogeneic hematopoietic stem cell transplantation (allo-HSCT). We aimed to develop novel analytical method appropriately handles right-censored data and competing risks understand the risk for GRFS each component of GRFS. This study was retrospective data-mining on cohort 2207 adult patients who underwent their first allo-HSCT within Kyoto Stem Cell Transplantation Group, multi-institutional joint research group 17 centers in Japan. The primary A stacked ensemble Cox Proportional Hazard (Cox-PH) regression 7 machine-learning algorithms applied prediction model. median age 48 years. For GRFS, model achieved better predictive accuracy evaluated by C-index than other state-of-the-art models (ensemble model: 0.670; Cox-PH: 0.668; Random Survival Forest: 0.660; Dynamic DeepHit: 0.646). probability 2 years 30.54% high-risk 40.69% low-risk (hazard ratio compared with group: 2.127; 95% CI, 1.19-3.80). developed analysis showed superior stratification existing methods using multiple algorithms.