Explainable AI decision support improves accuracy during telehealth strep throat screening

Area of Science:

• Telehealth and digital health innovations.
• Clinical decision support systems (CDSS) and artificial intelligence (AI).
• Infectious disease diagnostics and management.

Background:

• Smartphone-acquired images offer new avenues for remote diagnosis in telehealth.
• Clinician trust and understanding are critical for adopting AI-based clinical decision support systems (CDSS).
• This study investigates human-AI interaction paradigms for an AI CDSS detecting streptococcal pharyngitis (strep throat) from smartphone images.

Purpose of the Study:

• To examine how different human-AI interaction paradigms influence clinician responses to an AI CDSS for strep throat detection.
• To assess the impact of explainable AI on clinician accuracy, confirmatory testing, trust, and understanding.

Main Methods:

• A randomized experiment was conducted with 121 telehealth providers using an online survey.
• Participants evaluated clinical vignettes with either a Modified Centor Score or an explainable AI prototype.
• Linear models analyzed the effect of explainable AI on diagnostic accuracy, confirmatory testing, trust, and understanding.

Main Results:

• AI-based CDSS improved clinician predictions compared to the Centor Score.
• Despite higher agreement with AI predictions, clinicians reported lower trust in AI than in the Centor Score.
• Lower trust in AI led to increased requests for in-person confirmatory testing.

Conclusions:

• AI-based CDSS can enhance the accuracy of remote strep throat screening in telehealth.
• Improving human-machine collaboration, particularly trust and intelligibility, is essential for AI integration.
• Effective AI integration ensures optimal utilization of smartphones and AI for virtual healthcare and prevents antibiotic over-prescription.