To systematically explore how machine learning-powered clinical decision support systems (CDSS) have been employed to predict treatment outcomes in cancer care, synthesizing evidence across diverse cancer types, data modalities, and algorithmic architectures.
Approach:
Methodological Framework: The review followed a five-stage process based on Arksey and O’Malley’s framework, focusing on mapping the literature across oncology, artificial intelligence, and clinical informatics.
Research Questions: The primary question was how ML-enabled CDSS are utilized to predict treatment outcomes in cancer care, with secondary questions addressing common ML models, predicted clinical outcomes, and model performance evaluation.
Eligibility Criteria: Studies included cancer patients and those deploying ML techniques within CDSS or as predictive models for treatment outcomes.
Key Findings:
Machine learning algorithms can enhance clinical decision-making by predicting treatment efficacy and risks.
Real-world implementation of ML-enabled CDSS is limited due to barriers such as interoperability, model interpretability, and inadequate external validation.
Existing systematic reviews have focused on specific cancer types or predictive tasks, highlighting a gap in comprehensive synthesis across diverse settings.
Interpretation:
The review aims to map the breadth of evidence on ML models in CDSS for cancer treatment outcome prediction, addressing methodological strengths and limitations.
Limitations:
Limited real-world implementation of ML-enabled CDSS due to barriers such as interoperability and model interpretability.
Challenges in external validation of models.
Heterogeneity in cancer types and treatment regimens complicates predictive modeling.
Conclusion:
The review provides insights into the integration of predictive functionalities in clinical workflows.
