Acute Care Events During Systemic Cancer Treatment: Moving From Risk Prediction to Clinical Decision Support Using a...

Area of Science:

• Oncology
• Health Services Research
• Biostatistics

Background:

• Acute care events (ACEs) are a significant burden for cancer patients, often preventable.
• Existing prognostic risk models are underutilized for ACE prevention.
• A Clinical Advisory Panel (CAP) was formed to overcome translational barriers and develop effective risk models.

Purpose of the Study:

• To develop clinically and statistically valid prognostic models for predicting ACEs in cancer patients.
• To enable risk-stratified interventions to reduce the incidence of preventable ACEs.
• To address the translational gap in deploying prognostic models for cancer patient care.

Main Methods:

• Patients aged 21+ initiating cancer treatment were identified from electronic health records.
• Data were split into training (50%), validation (25%), and test (25%) sets.
• Constrained elastic-net logistic regression was used, incorporating clinically informed coefficient constraints.

Main Results:

• Two models were developed: a baseline model and a follow-up model incorporating clinical changes.
• Key predictors included prior ACEs, chemotherapy, heart failure, abnormal INR, and advanced cancer stage.
• Both models demonstrated acceptable statistical performance (C-statistic 0.71 and 0.70) in identifying high-risk patients.

Conclusions:

• A novel approach using two predictive models facilitates ACE risk assessment before and during cancer treatment.
• Risk-stratified interventions targeting comorbidities, treatment toxicities, and disease stage are recommended for ACE prevention.
• Clinical Advisory Panel (CAP) engagement was crucial in developing these clinically relevant models.