Identifying Cancer Patients at Risk for Heart Failure Using Machine Learning Methods.

Area of Science:

• Oncology
• Cardiology
• Data Science

Background:

• Cancer therapies can cause cardiotoxicity, negatively impacting patient outcomes and quality of life.
• Early identification of at-risk cancer patients is crucial for implementing preventive measures against cardiotoxicity.

Purpose of the Study:

• To predict the development of heart failure in cancer patients post-diagnosis using historical electronic health record (EHR) data.
• To evaluate the feasibility of machine learning (ML) models in identifying cancer patients at risk for cardiotoxicity.

Main Methods:

• Utilized EHR data from 143,199 cancer patients at UF Health.
• Compared four ML algorithms, including gradient boosting (GB), on 1,958 cases and 15,488 matched controls.
• Evaluated two feature encoding strategies for ML model input.

Main Results:

• The gradient boosting model achieved the highest AUC of 0.9077, with 0.8520 sensitivity and 0.8138 specificity.
• ML models effectively identified clinical factors associated with heart failure development.
• A subgroup analysis of chemotherapy-exposed patients showed a lower specificity score (0.7089).

Conclusions:

• Machine learning models demonstrate feasibility in predicting cancer therapy-related heart failure.
• This approach can aid in identifying high-risk cancer patients for timely intervention.
• Predictive modeling using EHR data holds potential for improving cancer patient care and quality of life.