Application of Machine learning in predicting cancer complications using longitudinal Data: A systematic review and...

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Accurate cancer prognosis, including metastasis, recurrence, and treatment side effects, is crucial for patient outcomes.
• Machine learning (ML) offers potential for improving cancer complication prediction using longitudinal data.
• A systematic review of ML models for cancer complication prognosis based on longitudinal data is lacking.

Purpose of the Study:

• To systematically review and assess the accuracy of ML models utilizing longitudinal data for predicting cancer-related complications.
• To evaluate the performance of ML techniques in oncology prognosis.

Main Methods:

• A systematic literature search was conducted in PubMed, Google Scholar, and IEEE Xplore (2020-2024).
• Seven studies employing ML with longitudinal data for cancer complication prognosis were reviewed.
• Risk of bias and diagnostic accuracy were assessed using Cochrane Risk of Bias and QUADES 2 tools, respectively.

Main Results:

• Pooled area under the curve (AUC) for predicting immune-related adverse events was 0.78.
• Pooled AUCs for cancer recurrence and mortality prediction ranged from 0.70 to 0.75.
• ML models integrating clinical, genomic, and imaging data showed superior predictive accuracy (AUC 0.82 for quality of life deterioration).

Conclusions:

• ML models using longitudinal data effectively predict cancer complications.
• Integrating multimodal data significantly improves the predictive accuracy of these models.
• These ML tools show promise for enhancing clinical decision-making in oncology.