Evaluating AI-based comprehensive clinical decision support for sepsis and ARDS: protocol for a Clinician Turing Test

Area of Science:

• Critical Care Medicine
• Artificial Intelligence in Healthcare
• Clinical Decision Support Systems

Background:

• Evaluating AI clinical decision support systems (CDSSs) in practice is challenging due to limited early-stage data.
• High-stakes decisions in the Intensive Care Unit (ICU) for conditions like sepsis and Acute Respiratory Distress Syndrome (ARDS) necessitate robust AI validation.
• The AI Ventilator Assistant (AVA) was developed for sepsis ARDS patients on mechanical ventilation, but predictive performance alone is insufficient for safety assessment.

Purpose of the Study:

• To introduce and validate a novel Clinician Turing Test for assessing the safety and appropriateness of the AI Ventilator Assistant (AVA).
• To determine if critical care clinicians can differentiate between AI-generated and human-generated treatment recommendations for sepsis ARDS patients.
• To provide a preclinical signal of AVA's safety and appropriateness before large-scale clinical deployment.

Main Methods:

• A multisite, randomized, electronic, vignette-based Phase 1b study employing a Clinician Turing Test design.
• Recruitment of 350 critical care clinicians (physicians, advanced practice providers) across six US hospitals.
• Participants reviewed clinical vignettes with AI-generated or human-enacted treatment plans, randomly assigned (1:1), to identify the source. Primary endpoint: accuracy in source identification via mixed-effects logistic regression.

Main Results:

• The study is designed to assess the primary endpoint of clinician accuracy in distinguishing AI-generated from human-generated treatment recommendations.
• Secondary endpoints include evaluating clinicians' perceptions of safety, appropriateness, confidence, and interest in AI CDSSs.
• The study aims to provide preliminary data on AI CDSS clinical appropriateness without the risks of actual deployment.

Conclusions:

• The Clinician Turing Test offers a novel approach to evaluating AI CDSS safety and appropriateness in a preclinical setting.
• Successful 'passing' of the Turing test by AVA would indicate a strong preclinical signal of its safety and appropriateness.
• This validation method informs decisions regarding future clinical implementation and evaluation of AI CDSSs in real-world environments.