Faculty of Medicine, University of Toronto, Toronto, ON, Canada
Ghazal Haddad, Katrina Hueniken, Scott Victor Bratman, John R de Almeida, David Paul Goldstein, Shao Hui Huang, Aaron Richard Hansen, Andrew J. Hope, Anna Spreafico, Wei Xu, Geoffrey Liu
Background: Cancer patients often experience symptoms in clusters, which could contribute to patient outcomes and quality of life; yet most analyses describe only individual symptom scores or a summation of symptom scores rather than symptom clusters. We compared machine learning to traditional statistical methods for grouping patients by symptoms in a heavy symptom burden patient population: oropharyngeal cancer patients undergoing initial treatment with radiation or chemoradiation (C/RT). Methods: K-means clustering was compared to traditional statistical methods (i.e., histogram-analysis; use of quantiles; categorization by clinico-demographic features) for classifying patients based on Edmonton Symptom Assessment System (ESAS) symptom scores in newly diagnosed oropharyngeal cancer patients before and after C/RT initiation. Results: 278 patients were classified into 3 K-means groups at each of two timepoints: pre-C/RT and post-C/RT. These groupings were formed primarily based on differences by symptom burden and were thus labeled: low, moderate, and high symptom-burden patient groups. The table shows dynamic change as patients moved from group to group during C/RT treatment. Greatest symptom-burdens in pre-C/RT patients were anxiety and tiredness; in post-C/RT, poor appetite and tiredness. Chi-squared residuals attributed being female (residuals of -3.95 for low, -2.46 for moderate, and 2.41 for high burden) and HPV-negative cancer (residuals of -1.48, -0.52, and 2.57, respectively) to being associated with higher symptom burden pre-C/RT; no clinico-demographic characteristics were associated with high symptom burden post-C/RT, suggesting this was a previously unidentified patient group. Although separation into K-means clusters in this population was primarily related to symptom burden, K-means better identified the number of patient groups and classified patients at the boundaries between groups (involving up to a quarter of patients) when compared to traditional statistical methods. Conclusions: Machine learning clustering analysis separated patients into discrete groups by symptom burden. K-means provided an objective means for clustering patients in a clinically-meaningful way when compared to traditional statistical methods for grouping patients.
Pre-C/RT Symptom Group | Post-C/RT Low | Post-C/RT Moderate | Post-C/RT High |
---|---|---|---|
Low - 164 (100%) | 80 (49%) | 70 (43%) | 14 (8.5%) |
Moderate - 68 (100%) | 18 (26%) | 35 (51%) | 15 (22%) |
High - 46 (100%) | 6 (13%) | 27 (59%) | 13 (28%) |
Disclaimer
This material on this page is ©2024 American Society of Clinical Oncology, all rights reserved. Licensing available upon request. For more information, please contact licensing@asco.org
Abstract Disclosures
2024 ASCO Quality Care Symposium
First Author: Anh B. Lam
2021 ASCO Annual Meeting
First Author: Loretta A. Williams
2024 ASCO Quality Care Symposium
First Author: Elif Andac-Jones
2023 ASCO Quality Care Symposium
First Author: Jibran Ahmed