Machine Learning-Based Early Warning Systems for Acute Care Utilization During Systemic Therapy for Cancer

Area of Science:

• Oncology
• Health Informatics
• Machine Learning

Background:

• Emergency department visits and hospitalizations are common during cancer systemic therapy.
• A longitudinal warning system was developed to predict acute care utilization.

Purpose of the Study:

• To develop and evaluate a machine learning-based warning system for predicting acute care use in cancer patients receiving systemic therapy.
• To identify key features that predict the need for acute care interventions.

Main Methods:

• A retrospective population-based cohort of 105,129 patients receiving intravenous systemic therapy for nonhematologic cancers was analyzed.
• Predictive models incorporated static (demographics, cancer type) and dynamic (symptoms, lab values) features.
• The system predicted the probability of acute care utilization within 30 days post-treatment.

Main Results:

• The developed ensemble model achieved an area under the receiver operating characteristic curve of 0.742 in the test cohort.
• Key predictors included prior acute care use, treatment regimen, and laboratory test results.
• The system could flag a significant portion of acute care events but showed underestimation for certain regimens and populations.

Conclusions:

• Machine learning warning systems show promise in identifying cancer patients at risk for acute care utilization.
• These systems can support preventive interventions and tailored treatment strategies.
• Further research is needed to address potential biases and prospectively evaluate system impact.