Machine Learning for Prediction of High-Risk Infections in Patients With Cancer

Area of Science:

• Oncology
• Medical Informatics
• Computational Biology

Background:

• Infectious complications significantly impact cancer patient outcomes, leading to hospitalizations, treatment delays, and mortality.
• Accurate risk stratification for infection-related hospitalizations (IRHs) is crucial for optimizing patient care and resource allocation.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting IRHs in adult cancer patients.
• To assess the performance of different ML approaches and data configurations in risk stratification.

Main Methods:

• Adult patients with lymphoma, multiple myeloma, chronic lymphocytic leukemia, colorectal, or lung cancer were included (2013-2023).
• Data from national registries and health records were utilized to define serious infections (sepsis, positive blood culture, ICU admission, or death).
• Multiple ML models were trained and tested using data up to admission, up to 48 hours post-admission (ML48), and a reduced feature set (ML48,reduced).

Main Results:

• The study included 9874 patients, with 2661 IRHs observed.
• The ML model achieved the highest performance (ROC-AUC: 0.79), outperforming ML48 (0.76) and ML48,reduced (0.75).
• Key predictors included previous blood cultures, carbamide, and neutrophil counts; performance varied by cancer type and ECOG score.

Conclusions:

• ML models demonstrate moderate effectiveness in stratifying IRH risk among cancer patients.
• These models can serve as valuable decision support tools for hospitalisation and discharge planning.