Sepsis and acute respiratory distress syndrome (ARDS) are common in intensive care units. Managing sepsis and ARDS is inherently complex and requires making numerous decisions under uncertainty. Artificial intelligence (AI) clinical decision support systems (CDSSs) offer a promising approach to support care management for sepsis and ARDS. The goal of this randomized, survey-based study is to compare treatment recommendations enacted by clinicians to those generated by an AI CDSS. The study will investigate whether an AI CDSS can generate treatment recommendations that are safe, appropriate, and indistinguishable to those provided by real clinicians. In this study, participants (i.e., critical care clinicians) will review a series of critical care cases (vignettes) in an electronic survey. Each vignette will contain a de-identified case of a patient with sepsis and ARDS as well as treatment recommendations for the case. Participants will assess the safety and appropriateness of each treatment recommendations and answer whether they think the treatment recommendations came from the clinician or an AI CDSS.
Study Type
INTERVENTIONAL
Allocation
RANDOMIZED
Purpose
HEALTH_SERVICES_RESEARCH
Masking
SINGLE
Enrollment
350
The clinical vignette will contain treatment recommendations which were generated by an artificial intelligence-based clinical decision support system.
University of Pennsylvania
Philadelphia, Pennsylvania, United States
Accuracy of Predicting the Source of Treatment Recommendation
Participants will answer if they think the treatment recommendations came from artificial intelligence (AI) or a clinician for each clinical vignette. Accuracy will be measured by participants correctly identifying the source of treatment recommendation.
Time frame: From enrollment to the end of the survey, an average of 45 minutes
Confidence of Predicting the Source of Treatment Recommendation
Participants will respond to their confidence in their prediction in whether the treatment recommendations of a vignette came from artificial intelligence or from a clinician. Confidence will measured on a Likert scale ranging from 0 (Not at all confident) to 7 (Extremely confident).
Time frame: From enrollment to the end of the survey, an average of 45 minutes
Appropriateness of Treatment Recommendations
Appropriateness will be measured by participants' assessments of the clinical appropriateness of the treatment recommendations in the vignettes via Yes-No and free-text responses.
Time frame: From enrollment to the end of the survey, an average of 45 minutes
Safety of Treatment Recommendations
Safety will be measured by participants' assessments of the overall safety of the treatment recommendations in the vignettes via Yes-No and free-text responses.
Time frame: From enrollment to the end of the survey, an average of 45 minutes
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.