Feasibility Assessment of Telehealth-Based Cancer Pain Management Through Machine Learning: A Prospective Clinical...

Area of Science:

• Oncology
• Health Informatics
• Machine Learning

Background:

• Telehealth is effective for cancer pain management, but optimal pathways for tailoring interventions and resource allocation are challenging.
• Artificial intelligence (AI) and machine learning (ML) may enhance prediction of patient needs for remote or in-person evaluations.

Purpose of the Study:

• To evaluate the predictive performance of various ML models in determining the need for televisits in cancer pain management.
• To assess the feasibility of using structured telemedicine data for clinical workload modeling and follow-up optimization.

Main Methods:

• Analysis of data from two cancer pain patient cohorts, including sociodemographic and clinical variables.
• Harmonization of datasets and testing of six ML models (logistic regression, RF, GBM, SVM, KNN, MLP) using cross-validation.
• Evaluation of model performance using F1-score, accuracy, and AUC-ROC, with sensitivity analysis for class weighting and cohort variable exclusion.

Main Results:

• No statistically significant associations were found between analyzed variables and the number of televisits.
• ML model performance varied, with the MLP showing the highest F1-score (0.65) but exhibiting instability.
• No significant differences were observed between algorithms; class weighting offered minor, non-significant improvements for SVM and LR.

Conclusions:

• Current ML models do not demonstrate strong predictive power for optimizing telehealth in cancer pain care.
• The developed ML framework highlights the potential of structured telemedicine data for modeling clinical workload.
• Further research is needed to refine models for more accurate prediction and resource allocation in cancer pain management.