Machine learning-assisted prognosis prediction and surgical decision-making for glioblastoma: perceived benefits and concerns of patients, caregivers, and neurosurgeons.

Introduction: It is becoming more common for machine learning (ML) models to aid prognostication and clinical decision-making, including for glioblastoma (GBM) cases. However, there is a lack of empirical research on how end-users view potential benefits and risks of implementing such models in clinical practice. Methods: This study examines the perspectives of GBM patients ( Results: All three groups thought a major benefit of the ML model was its ability to take into account a large amount and scope of patient data, which could help facilitate communication and decision-making among patients and neurosurgeons when planning treatment strategies or end-of-life care. Participants also expressed concerns about potential inaccuracies or biases in model output, and shared unease about the possibility of the model replacing a neurosurgeon's clinical judgment entirely. Some patients and caregivers expressed concern about the model being in early stages of development and about how the delivery of ML-informed prognostic information could cause patients to lose hope or become confused about important prognostic or surgical information. Discussion: This study highlights the value of engaging multiple stakeholders and triangulating their perspectives when developing ML models to support clinical decision-making. While ML models show great promise in synthesizing large amounts of data and supporting decision-making, biases in the data used to train these models and over-reliance on their predictions risk negatively affecting patient outcomes.