Patient-Reported Outcomes, Machine Learning May Predict Breast Cancer Recurrence Earlier

Although patient-reported outcomes (PROs) have become increasingly important for evaluating quality of life and treatment tolerability, investigators sought to determine whether longitudinal changes in symptoms could also function as early biomarkers of disease progression.¹

Researchers retrospectively analyzed 445,239 longitudinal PRO entries from 2738 patients with early- and advanced-stage breast cancer enrolled in 4 clinical trials. The dataset included 15 validated PRO measures assessing symptoms, physical functioning, and overall quality of life. Machine learning and Cox proportional hazards models were used to evaluate associations between symptom deterioration, tumor characteristics, recurrence, progression-free survival (PFS), and overall survival (OS).¹

Among patients with radiographically confirmed recurrence or disease progression, approximately 89.2% experienced worsening in at least 1 PRO before relapse was identified on imaging. Symptom deterioration occurred a median of 706 days before radiographic confirmation compared with 85 days before recurrence was clinically detected, suggesting that patient-reported symptom changes may provide substantially earlier warning of disease progression.¹

The findings indicate that subtle changes in symptoms experienced by patients may emerge long before conventional imaging detects recurrent disease, supporting the potential role of PROs as complementary clinical biomarkers rather than solely measures of treatment tolerability.¹

Machine Learning Improved Survival Prediction

Beyond recurrence detection, investigators evaluated whether integrating PRO data with machine learning could improve prognostic modeling.

Specific symptom changes were associated with distinct clinical outcomes. Appetite loss demonstrated the strongest correlation with increasing tumor size, whereas worsening pain and diarrhea emerged as the strongest predictors of both OS and PFS.¹

Gradient boosting machine learning models that combined deterioration across all PRO domains with PFS achieved the highest predictive performance for OS, correctly classifying survival outcomes with an area under the receiver operating characteristic curve (AUC-ROC) of 0.932. These outperformed models using PRO data alone (AUC-ROC, 0.806) or PFS alone (AUC-ROC, 0.880).¹

The authors noted that integrating multiple patient-reported symptom trajectories captured clinically meaningful information that traditional disease end points alone may overlook, potentially enabling more personalized surveillance strategies.¹

Implications for Oncology Pharmacists

The findings build upon growing evidence supporting electronic PRO (ePRO) monitoring as part of routine oncology care. Previous studies have demonstrated that remote symptom monitoring can improve adverse event recognition, reduce emergency department visits and hospitalizations, prolong time on treatment, and improve survival through earlier intervention.²

During the 2026 American Society of Clinical Oncology Annual Meeting, Debra Patt, MD, PhD, MBA, emphasized that successful implementation of ePRO programs depends on rapid response workflows supported by dedicated triage teams. She noted that oncology pharmacists play a critical role in evaluating reported toxicities, optimizing supportive care, recommending dose modifications, and helping patients remain adherent to therapy through earlier symptom management.²

For pharmacists, the current findings suggest that longitudinal symptom monitoring may eventually extend beyond toxicity management to include earlier identification of disease recurrence and risk stratification. Changes in symptoms—such as appetite loss, pain, or gastrointestinal toxicities—could become clinically meaningful signals that warrant additional assessment alongside routine imaging and laboratory evaluation.

The investigators also noted that their findings align with ongoing efforts by regulatory agencies, including the FDA and European Medicines Agency, to further integrate PROs into oncology endpoint assessment.¹ As artificial intelligence continues to advance across breast cancer care—from diagnosis and risk prediction to treatment planning and disease monitoring—the integration of machine learning with patient-reported data may offer a scalable, patient-centered approach that could identify disease progression earlier while supporting more individualized care.³

REFERENCES

1. Wang W, Zhang C, Muluneh B, et al. Patient-reported outcomes as early indicators of recurrence, disease progression, and survival via machine learning in breast cancer. NPJ Breast Cancer. Published online June 27, 2026. Accessed July 15, 2026. doi:10.1038/s41523-026-01001-3

2. Patt D, Valletti D. EPRO-Based Monitoring Improves Early Symptom Management in Patients With Cancer. Pharmacy Times. June 1, 2026. Accessed July 15, 2026. https://www.pharmacytimes.com/view/epro-based-monitoring-improves-early-symptom-management-in-patients-with-cancer

3. NPJ Breast Cancer. Artificial Intelligence in Breast Cancer: From Discovery to Clinical Impact. Nature. Accessed July 15, 2026. https://www.nature.com/collections/gbfibebfjd