Predicting Chemotherapy-Related Symptom Deterioration Using Hybrid Deep Learning Architecture

Area of Science:

• Oncology
• Artificial Intelligence
• Digital Health

Background:

• Chemotherapy patients experience various physical and mental symptoms.
• Predicting symptom escalation is crucial for timely intervention and improved patient outcomes.
• High class imbalance in symptom escalation data poses a challenge for predictive modeling.

Purpose of the Study:

• To develop and evaluate a hybrid deep learning model (CNN-LSTM) for predicting symptom escalation in chemotherapy patients.
• To assess the effectiveness of different data aggregation intervals for improving predictive performance.
• To enable AI-driven decision support for real-time clinical interventions.

Main Methods:

• Utilized a hybrid Convolutional Neural Network with Long Short-Term Memory (CNN-LSTM) deep learning architecture.
• Aggregated daily self-reported symptom data into 3- to 7-day intervals to address class imbalance and enhance temporal resolution.
• Employed five-fold cross-validation for robust model training and generalization assessment.

Main Results:

• The CNN-LSTM model achieved 83% accuracy, 89% precision, 86% recall, 88% F1-score, and 83% AUC for physical symptom prediction using 5-day intervals.
• Optimal data aggregation into 5-day intervals significantly improved predictive accuracy for physical symptoms.
• Demonstrated the model's capability in capturing temporal dependencies for symptom progression forecasting.

Conclusions:

• Hybrid deep learning models like CNN-LSTM are effective for continuous symptom monitoring and early detection in chemotherapy patients.
• AI-driven decision support systems can enhance clinical interventions through accurate symptom prediction.
• Integration into digital health platforms can improve the quality of care by facilitating real-time symptom tracking and prediction.