OBJECTIVES: Artificial intelligence (AI) in medical imaging is a rapidly growing field and promises to improve the personalization of breast cancer follow-up, yet evidence supporting comparative effectiveness is lacking. To quantify how the benefits of AI in personalized follow-up compare against usual care, this study aimed to describe currently unpersonalized variation in daily clinical practice using real-world data.

METHODS: Patients with stage I-III invasive breast cancer who received surgical treatment between 2000 and 2020 were included. All imaging activities during follow-up were collected through linkage with hospital-based electronic health records. The INFLUENCE 2.0 nomogram was used for individual risk prediction of recurrence. Process analysis techniques were used to create a mapping of patients and activities to investigate the real-world utilization of care processes and assess the underlying characteristics.

RESULTS: In the period between 2000 and 2020, 3,478 patients were included with a mean follow-up of 4.9 years. In the first 12 months following treatment, patients on visited the hospital between 1-5 times (mean 1.3, IQR 1-1) and then received between 1-9 imaging activities (mean 1.7, IQR 1-2). Mammogram was the prevailing imaging modality, accounting for 70% of the imaging activities. A proportion of patients were recalled to the hospital within 40 days after the annual follow-up imaging activity for a repeat imaging activity. This proportion is 20% in the first year and decreases to 14% in the fifth year of follow-up. Surprisingly, patients with lower predicted risk of recurrence were associated with a higher number of hospital visits.

CONCLUSIONS: This study describes the variation in daily clinical practice for follow-up of breast cancer patients. Variation through AI-based personalization promises to encourage more efficient use of imaging resources and better patient outcomes. Hence, a flexible simulation model is required to determine the impact of AI on the healthcare system regarding healthcare utilization and costs.