Predicting adverse events for risk stratification of chemotherapy based stem cell mobilization in multiple myeloma

TL;DR

This paper explores using machine learning to predict adverse events during outpatient chemotherapy for multiple myeloma patients, aiming to reduce hospital bed usage.

Contribution

The novel contribution is the development of machine learning models to predict adverse events and their timing during stem cell mobilization.

Findings

69% of patients experienced severe adverse events requiring hospitalization.

Classification models accurately predicted some adverse events (e.g., elevated creatinine) with high ROC-AUC scores.

Regression models forecasted adverse event onset with a mean error of just over one day.

Abstract

Autologous stem-cell transplantation is a fundamental therapy for multiple myeloma. Although inpatient chemo-based stem-cell mobilization (SCM) is standard care in Germany, outpatient approaches could ease healthcare constraints. We analyzed 109 myeloma patients undergoing SCM and collection at the University Medical Center Göttingen for safety. We then trained machine learning models to predict adverse events (AEs) requiring hospitalization and to forecast AE onset timing for optimized ward management. In our cohort, 97% achieved successful collection, but 69% experienced severe AEs necessitating hospitalization. Simulations suggest a risk-stratified outpatient protocol could cut bed usage by at least one third without compromising safety. Classification models accurately predicted some AE types (e.g., elevated creatinine, ROC-AUC 1.0), though neutropenic fever remained challenging (ROC-AUC 0.67). Regression models forecast AE onset with a mean error of just over one day. These results outline a data-driven roadmap for safely adopting outpatient SCM and optimizing resource allocation in clinical practice.

Figures