Patient perspectives on AI-based decision support in surgery.

Abstract

Background: Predictive machine learning in healthcare, especially in surgical decisions, is advancing swiftly. Yet, literature on patient views regarding predictive machine learning, specifically its use throughout the clinical course, is scarce. Views among patients who underwent colorectal surgery (CRS) on the use of intra-operative predictive machine learning (IPML) by surgeons, particularly those aiming to predict colorectal anastomotic leakage (CAL), were explored in this study.

Objective: This study investigated the views of patients who previously underwent CRS on the implementation of IPML models. Domains of interest were perceptions of IPML, perceived role in decision-making and information provided in the clinical encounter.

Methods: A qualitative research design was employed, using focus groups and semi-structured interviews with patients who had undergone CRS. Descriptive thematic analysis was used to analyse data and identify prevailing themes and attitudes. The associations in the code tree were established based on a co-occurrence table. The patient sample size was determined using a saturation analysis.

Results: A study with n=19 participants across four focus groups and seven interviews found a generally positive perception regarding the use of IPML models in CRS. Participants recognised their potential to enhance surgical decision-making but stressed the surgeon's role as the primary decision-maker, suggesting IPML models act as advisory tools, with surgeons able to override recommendations. Personalised communication and consideration of quality of life were emphasised, highlighting the need for a balanced integration of IPML models to support clinical judgement and the construction of patient preferences.

Conclusion: IPML in CRS is well-received by participants, provided that surgeons retain the ability to override model recommendations and document their decisions transparently. Trust in the surgeon remains a key factor in patient acceptance of IPML, reinforcing the need for clear explanations during consultation sessions. Regardless of the use of IPML, tailoring patient communication and addressing the quality-of-life impacts of anastomosis vs stoma are also critical.