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

Abstract

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

A convolutional neural network-based model that predicts acute graft-versus-host disease after allogeneic hematopoietic stem cell transplantation

“…While we and other groups have previously developed machine learning-based models to predict outcomes after HSCT [7][8][9] , but the arbitrariness of variable settings has not been solved, especially regarding HLA information. In this study, for the first time, we incorporated raw information about specific antigens and/or alleles of both donors and recipients into a machine learning-based prediction model.…”

“…Recent application of machine learning algorithms, which perform statistical calculations without the assumptions required by conventional methods, is beginning to provide novel insights into clinical practice [7][8][9] . However, these previous studies utilizing machine learning algorithms have not solved the arbitrariness of variable settings and failed to incorporate detailed, raw clinical data.…”

Survival Prediction of Children Undergoing Hematopoietic Stem Cell Transplantation Using Different Machine Learning Classifiers by Performing Chi-Square Test and Hyperparameter Optimization: A Retrospective Analysis

“…The dataset used in this study was retrieved from the ML repository at the University of California, Irvine, and the version utilized in this study was extracted from [ 27 ]. It covers medical information for children who have been diagnosed with a variety of hematologic diseases and who underwent unmodified allogeneic unrelated donor HSCT [ 43 ]. Hence, this dataset comprises 187 occurrences and 37 attributes that contain information about individuals who have been diagnosed with a range of hematologic, malignant, or benign diseases.…”

“…Moreover, Arabyarmohammadi et al used the Cox regression model to estimate the probability of patient relapse after acute myeloid leukemia posthematopoietic cell transplantation [ 42 ]. Similarly, Iwasaki et al created a stacked ensemble of the Cox proportional hazard (Cox-PH) regression and 7 machine learning algorithms and discovered prediction accuracy with a C -index of 0.670 utilizing the ensemble model [ 43 ]. On the other hand, Morvant et al used machine learning (support vector machine (SVC) and Ridge logistic regression (LR Ridge)) with leave-one-out cross-validation to compare several combinations for predicting bone marrow minimal residual disease (MRD) before autologous stem cell transplant consolidation (ASCT) and discovered AUCs of up to 0.63 and 0.82 for negative vs. positive MRD in the lesion with the highest uptake [ 44 ].…”

Machine learning applications and challenges in graft-versus-host disease: a scoping review

“…HLA matching algorithms are continually evolving because of technological advancements. ML-based techniques are being investigated to improve HLA donor-recipient matching [46,47 ▪▪ ,48 ▪▪ ]. Further research is needed to develop an algorithm to identify the best match between the donor and the recipient to increase the possibility of a successful match.…”