Patient perspectives on active surveillance and machine learning–driven risk stratification for low-risk breast cancer.

Abstract

e13776 Background: Growing evidence suggests that overdiagnosis from routine screening may result in overtreatment among older women with indolent breast cancers. Active surveillance has emerged as a potential strategy to reduce treatment-related morbidity without compromising survival outcomes. The use of machine learning (ML) in imaging studies has demonstrated promise in supporting risk stratification and prognostication and may allow for selection of candidates for active surveillance. However, little is known about how older women perceive surveillance-based cancer care or the role of ML in clinical decision-making. Methods: We conducted focus groups to understand patient perceptions on active surveillance and ML-driven risk stratification for low-risk breast cancer. Participants were women aged ≥ 70 either unaffected by or diagnosed with early-stage breast cancer. We used purposive sampling to recruit patients from primary care, oncology, and community-based settings within our catchment area. Focus group transcripts were coded using NVivo qualitative research software. Codes were then analyzed and synthesized into central themes using a deductive approach. Results: We held 4 focus groups with 11 participants (mean age, 75 years). Overall, 64% of participants had a history of breast cancer and 36% reported educational attainment of “Some College” or “High School Degree”. Participants were generally receptive to the concept of active surveillance, though willingness to pursue it varied, and was shaped by perceived cancer risk, prior experiences, desired information, and risk tolerance. Trust in physician recommendations and doctor-patient relationships were central to decision-making about pursuing active surveillance. Participants weighed anticipated long-term outcomes, including treatment burden, quality versus quantity of life, the psychological impact of surveillance, and the perceived value of definitive treatment, while emphasizing autonomy, informed choice, and flexibility to change management over time. Participants favored ML as a supportive tool integrated with human oversight rather than a replacement for clinician judgment. Comfort with ML varied by age, educational attainment, values, previous experiences, and family influences. Accuracy and reliability were viewed as essential for high-stakes decisions. Conclusions: Older women expressed nuanced and conditional acceptance of both active surveillance and ML-supported cancer care. Trust, autonomy, risk tolerance, and human oversight emerged as critical determinants of acceptability. These findings will directly inform the development of a population-based survey and provide a patient-centered foundation for a future prospective trial evaluating ML-driven risk stratification to support active surveillance in older women with indolent breast cancers.