What makes a 'good' decision with artificial intelligence? A grounded theory study in paediatric care

Area of Science:

• Clinical Decision-Making
• Machine Learning in Healthcare
• Artificial Intelligence in Medicine

Background:

• Clinical decisions increasingly involve patient-level data and advanced analytics.
• Integrating machine learning (ML) models into clinical practice necessitates a structured approach for safe and effective use.
• Existing frameworks may not fully address the complexities of ML-driven patient care.

Purpose of the Study:

• To develop a comprehensive framework for good clinical decision-making when using machine learning (ML) models for interventional, patient-level decisions.
• To provide clinicians with an actionable guide to navigate the integration of ML insights into patient care.
• To ensure ethical and medicolegal standards are maintained when employing AI in healthcare.

Main Methods:

• Grounded theory qualitative interview study with 16 healthcare professionals and an ML specialist.
• Semi-structured interviews simulated intensive care unit handovers with hypothetical patient cases and ML model visualizations.
• Framework development involved member checks and patient consultations for refinement.

Main Results:

• A framework was developed, grounded in medicolegal and ethical standards, for incorporating ML model inference into clinical decisions.
• Key decision-making factors include patient evidence, applicability of knowledge, patient/family/local context, model details, and model's representation of patient data.
• Participants exhibited automation bias, with some deferring to ML predictions and others prioritizing their medical knowledge; consistent desire for local performance data was noted.

Conclusions:

• Effective clinical decision-making with AI tools necessitates multi-domain reflection, balancing ML insights with clinical expertise.
• An actionable framework and question guide are provided to support clinicians in making sound decisions using AI.
• This work facilitates the responsible and ethical integration of ML into patient-level clinical practice.