From Advice to Action: Real-World Behaviour of Patients Using an Integrated Clinical AI for Navigating the Healthcare System
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
AI-powered digital front-door (DFD) tools are increasingly used to guide patients to appropriate care and alleviate healthcare system pressure. Yet, most evaluations narrowly assess triage accuracy or stated intent, offering limited insight into real-world behaviour and care appropriateness.
Methods
The ESSENCE study was a prospective, real-world quality improvement evaluation embedded in Portugal’s largest private healthcare network (CUF). Adults using the Ada Health clinical AI (cAI) via the myCUF app reported their care intent before and after symptom assessment. Actual behaviour was tracked through electronic health records and follow-up surveys. Physician panels retrospectively assessed the appropriateness of intended and observed care.
Results
Of 1,470 adult (≥18 years) participants (mean age 38.5 years; 57.7% female), 33% revised their planned choice of care immediately after assessment. Uncertainty dropped from 13% to 5%. Among 721 participants with follow-up, 59% changed their behaviour: 29% de-escalated, 17% escalated, and 13% resolved prior uncertainty. General practitioner (GP) visits more than doubled (42% vs. 19%), while specialist visits halved (30% vs. 57%). Overall, the proportion of appropriate care increased from 30% pre-assessment to 64% post-behaviour (p□<□0.001). Of those initially planning to visit the emergency department, 39% opted for alternative care. 93% of those followed up were later confirmed to have avoided unnecessary emergency department (ED) visits.
Conclusion
Embedding AI symptom assessment into a DFD shifted both patient intentions and behaviours, reducing uncertainty and promoting more appropriate healthcare use. These findings highlight the value of AI not only in generating accurate recommendations, but also in shaping real-world decision-making. Future research should identify specific factors that influence behaviour and evaluate impact across diverse healthcare settings.
