Machine Learning Maps Three Fatigue Subgroups in Breast Cancer Patients

Breast cancer patients undergoing treatment often experience cancer-related fatigue, a symptom that can undermine health outcomes and quality of life. A new cross-sectional study published in Supportive Care in Cancer applied latent profile analysis and five machine learning algorithms to survey data from 1,005 patients, identifying three distinct fatigue subgroups and a short list of psychological factors that best predict where an individual patient falls. The researchers report that XGBoost, an ensemble algorithm, performed best among the tested models, achieving an area under the curve of 0.88.

Participants completed a self-administered questionnaire measuring fatigue with the Cancer Fatigue Scale, recruited through a stratified sampling approach. Latent profile analysis, a statistical technique that uncovers unobserved subgroups within a population, sorted patients into three classes: a stable fatigue group comprising 9.8 percent of the sample, a low fatigue unstable group at 27.4 percent, and a high fatigue unstable group accounting for 62.9 percent. The labels reflect both the severity of reported fatigue and the stability of symptoms, suggesting that for most patients fatigue levels fluctuate rather than remain constant during treatment.

The team then trained five machine learning models—K-nearest neighbors, Adaptive Boosting, Gaussian Naive Bayes, Extreme Gradient Boosting, and Random Forest—to identify which measured variables best predicted class membership. Area under the curve values ranged from 0.83 to 0.88, indicating moderately strong discrimination across all five algorithms. The final model singled out perceived stress, depression, and self-compassion as the most important predictors of cancer-related fatigue in this population.

The authors argue that these findings support a stratified approach to fatigue management. For the high fatigue unstable group, interventions targeting perceived stress and depression may be most relevant; for the low fatigue unstable group, they suggest focusing on self-compassion and stress management; and for the stable fatigue group, anxiety reduction and enhanced social support. Such individualized planning, they note, could give clinicians a practical framework rather than a one-size-fits-all protocol.

Several caveats deserve emphasis. The study is cross-sectional, meaning it captured fatigue and its correlates at a single point in time; it can identify associations but cannot establish that stress, depression, or low self-compassion cause fatigue, nor that modifying them would reduce symptoms. All data came from self-reported questionnaires, and the authors themselves describe machine learning here as a way to identify factors associated with fatigue rather than proven causal drivers. The proposed intervention targets are presented as a framework for future work, not as validated treatment effects. Even so, the study addresses a recognized gap: machine learning research on cancer-related fatigue in breast cancer populations has remained limited, and the combination of subgroup identification with predictive modeling offers a template for larger, longitudinal studies that could test whether addressing these psychological factors actually changes fatigue trajectories over the course of treatment.

Subject of Research: Fatigue subgroups and predictive factors in breast cancer patients identified with latent profile analysis and machine learning

Article Title: Characterizing potential subgroups and predictive factors of cancer-related fatigue among breast cancer patients through latent profile analysis and machine learning algorithms

Article References: Zheng, J., Zhao, X., Cui, Y., Liu, B., Chen, Z., Lu, X., & Yang, X. (2026). Characterizing potential subgroups and predictive factors of cancer-related fatigue among breast cancer patients through latent profile analysis and machine learning algorithms. Supportive Care in Cancer, 34(11), Article 1172. https://doi.org/10.1007/s00520-026-11417-z

Image Credits: AI Generated

DOI: 10.1007/s00520-026-11417-z

Keywords: breast cancer, cancer-related fatigue, machine learning, latent profile analysis, XGBoost, perceived stress, depression, self-compassion, cross-sectional study, Supportive Care in Cancer, Characterizing, potential