A model developed by machine learning and principal component analysis for predicting bronchiolitis obliterans syndrome after allogeneic HSCT in a Chinese population with hematologic malignancies

Abstract

Bronchiolitis obliterans syndrome (BOS) is a severe pulmonary complication following allogeneic hematopoietic stem cell transplantation (allo-HSCT), with early prediction being crucial. While pulmonary function tests (PFTs) are fundamental for BOS assessment, the predictive value of pretransplant PFT results remains uncertain. In this study involving allo-HSCT recipients with hematologic malignancies who survived over 100 days post-HSCT, we aimed to determine the predictive significance of pretransplant PFTs for BOS. Data from 742 eligible patients were randomly divided into training and validation cohorts, and machine learning algorithms were employed for feature engineering, feature selection, and modeling. Principal component analysis (PCA) was utilized to reduce the dimensionality of PFT parameters. Over a median follow-up of 573.5 days, 57 patients developed BOS, with a median interval of 269 days from transplantation to BOS onset. Our multivariate logistic regression (MLR) model, incorporating the first principal components of PCA-treated PFT results along with age, sex, and previous nonpulmonary chronic graft-versus-host disease (GvHD), demonstrated discriminant AUC values of 0.671 (95 % CI, 0.522-0.820) and 0.669 (95 % CI, 0.588-0.751) in the validation and training sets, respectively. Pretransplant PFT results emerged as pivotal in predicting BOS risk post-allo-HSCT. The MLR model, developed through data-driven machine learning, effectively identifies high-risk BOS populations at an early stage.

Keywords: Bronchiolitis obliterans syndrome; Hematopoietic stem cell transplantation; Machine learning; Predictive model; Pulmonary function test.