Patient-Centered Modeling of the Breast Biopsy Experience

Isabel Nieto Alvarez^1,2,3*Erick Bojorges-Valdez⁴Elvira Lang⁵Mohammadreza Ranaei Sharif²Göran Köber²Nicolas Rohleder⁶Oliver Amft^2,7

• ¹Chair for Digital Health, Medizinische Fakultät, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany
• ²Intelligent Embedded Systems Lab, Albert-Ludwigs-Universität Freiburg, Freiburg, Germany
• ³Siemens Healthcare, Erlangen, Bavaria, Germany
• ⁴Department of Engineering Studies for Innovation, Universidad Iberoamericana, Mexico City, Mexico
• ⁵Hypnalgesics, LLC, Brookline, MA, United States
• ⁶Chair of Health Psychology, Philosophische Fakultät und Theologische Fakultät. Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Bavaria, Germany
• ⁷Hahn-Schickard-Gesellschaft für angewandte Forschung e.V., Freiburg, Germany

Despite significant advances in breast cancer screening and early detection over recent decades, rising patient volumes, limited resources, and time constraints hinder healthcare teams from anticipating distress and effectively managing the patient experience. We leveraged real-world data from 236 patients during a breast biopsy procedure and follow-up period. The study goal was to model important components of the multifaceted biopsy procedure and its effect on patient experience. We integrated data from patient-reported outcomes, psycho-social assessments, and workflow annotations. We (1) provide a visual model of the patient pathway, (2) predict, with linear mixed models and machine learning, anxiety based on psychological pre-assessments as well as procedural events, and (3) analyze communication between caregiver and patient to understand moderators of the patient experience. Predictive modeling revealed significant correlation between psychological pre-assessments and median anxiety during biopsy (IES β = 0.91, CES-D β = 0.8, PSS β = 0.62, STAI β = 0.58, all with p < 0.001). Higher baseline stress was strongly associated with greater anxiety during biopsy. Centering each individual's procedure time at her first local anesthesia (LA) revealed a significant (βt2 p = 5.43e−06) temporal pattern in anxiety, which increased until LA and decreased afterwards. Using natural language processing, we identified patient expressions of pain and distress alongside workflow annotations. Our findings highlight the potential of combining data to model patient experience during a medical procedure. Our work helps to develop digital twins of medical procedures to support clinicians to provide proactive care and mitigate patient distress.