Machine Learning for Prediction of High-Risk Hospitalizations in Lymphoma Patients: A Danish Population-Based Study

Area of Science:

• Oncology
• Infectious Diseases
• Medical Informatics

Background:

• Infections are a significant cause of hospitalization and mortality in lymphoma patients.
• Accurate risk stratification of infection-related hospitalizations (IRH) is crucial for timely intervention.
• Distinguishing between serious-IRH (S-IRH) and non-serious-IRH (NS-IRH) can guide clinical management.

Purpose of the Study:

• To develop and evaluate a machine learning (ML)-based risk stratification method for classifying IRH in lymphoma patients.
• To predict serious-IRH (S-IRH) using clinical data and an XGBoost model.
• To assess the clinical utility of the ML model as a decision support tool.

Main Methods:

• An XGBoost model was constructed using clinical data from health records and registries.
• S-IRH was defined by criteria including death, positive blood culture, bloodstream infection, and sepsis.
• The model was trained and validated on data from 727 adult lymphoma patients treated with first-line therapies between 2013-2023.

Main Results:

• A total of 591 IRHs were identified, with 119 classified as S-IRH.
• The ML model achieved a ROC-AUC of 71.0% for predicting S-IRH.
• At a threshold of 0.3, the model demonstrated 63.0% sensitivity and 63.0% specificity, with a negative predictive value of 87.5%.

Conclusions:

• The developed ML model shows acceptable performance in predicting S-IRH in lymphoma patients.
• While not sufficient as a standalone tool, the model can serve as a valuable decision support system for clinicians.
• Further refinement may enhance its utility in managing infection risks during lymphoma treatment.