Time and temporality in machine learning methods to improve cancer clinical decision support: A literature review.

OBJECTIVE: This systematic literature review explores how temporal and time-related dimensions are incorporated into Machine Learning (ML) models used in Clinical Decision Support Systems (CDSS) for cancer. The study examines current applications, identifies research trends and limitations, and proposes future directions for enhancing temporal modeling in ML-based cancer decision support. METHODS: Following the PRISMA guidelines, a systematic search was conducted in the Web of Science database using combinations of keywords related to machine learning, clinical decision support, time, and cancer. After applying inclusion and exclusion criteria, 83 peer-reviewed studies published between 2014 and 2023 were analyzed. Each study was examined to determine how temporal aspects were integrated into ML models and categorized using the Ancona time framework to capture conceptions of time, actors relating to time, and mapping activities to time. RESULTS: The findings show increasing research activity since 2015, with rapid growth from 2020 onwards. Most studies focused on survival analysis, time-series modeling, and time-to-event prediction, emphasizing their value in prognosis and treatment planning. However, temporal constructs related to annotation efficiency, biological timing, and longitudinal data remain underexplored. Many approaches still rely on static datasets, lack external validation, and provide limited interpretability. The mapping to the Ancona framework revealed fragmented consideration of time across studies, with limited attention to synchronization, temporal orientation, and patient or clinician experiences of time. CONCLUSION: The review highlights both progress and persistent limitations in applying temporal dimensions to ML-based cancer CDSS. Future research should strengthen longitudinal modeling, improve temporal data integration, and consider the clinical and human aspects of time to enhance decision support accuracy and relevance.