Artificial intelligence (AI)-supported symptom self-assessment led to more appropriate patient care-seeking decisions
1. Cotte and colleagues evaluated changes in patients’ intentions and actual care after the use of an artificial intelligence (AI) symptom self-assessment platform.
2. AI-supported symptom assessment reduced patient uncertainty and promoted more appropriate use of healthcare resources.
Evidence Rating Level: 2 (Good)
Study Rundown: Healthcare systems globally face growing strains from rising care needs. However, patients’ self-triaging process can both overestimate urgency and miss potentially serious conditions. An AI-supported self-symptom assessment platform could improve care appropriateness. Cotte and colleagues evaluated Ada Health, an AI-supported diagnostic decision-support system, for its ability to influence patient behavior. Adult patients reported their care-seeking intention before and after the symptom self-assessment, and electronic records and follow-up surveys captured subsequent patient behavior. A panel of three senior physicians not involved in patient treatment reviewed the appropriate care level using clinical documentation. The study found that 33.0% of the participants changed their care plans following self-assessment, and uncertainty decreased from 12.6% to 5.0%. Additionally, primary care consultations increased from 16.3% to 42.1% following self-assessment. Of the non-emergency patients, appropriate care increased from 29.8% of initial plans to 64.4% of observed actions. This study demonstrated that AI-supported symptom self-assessment may lead to more appropriate care-seeking behavior among patients.
Click here to read the study in NEJM AI
Relevant Reading: Technology-supported self-triage decision making
In-Depth [prospective cohort]: 1,470 patients from Portugal’s largest private healthcare network were enrolled in the study. Care intentions were recorded before and immediately after the self-assessment. Care behavior was established for 721 participants by electronic health records and surveys. Three senior physicians, blinded to recommendations and intentions, retrospectively rated appropriate care levels for non-emergency patients. Emergency visits were reviewed separately. Among the 717 participants for whom both pre-assessment intentions and observed behavior were available, only 41% followed through with their initially planned care level. 17.2% escalated to a higher-urgency option, whereas 28.9% de-escalated to a lower-urgency one, indicating a net shift toward lower-acuity care. Primary care use increased from 16.3% to 42.1% (95% confidence interval [CI], 22.5% to 29.1%), while specialist visits decreased from 49.7% to 29.8% (95% CI, −23.0% to −16.6%). Proportions of participants who chose self-management remained stable, and emergency care use increased from 13.4% to 21.2% (95% CI, 5.2% to 10.4%). For the 382 non-emergency participants who were evaluated, care appropriateness increased by 34.55 percentage points (95% CI, 27.75-41.36). This study was limited by unknown behavior in 51.0% of the participants, inability to assess appropriateness of self-care, and inability to capture hospitalizations outside of the private healthcare network. Nonetheless, this study provided promising evidence that AI platforms can improve patients’ ability to seek more appropriate care.
Image: PD
©2026 2 Minute Medicine, Inc. All rights reserved. No works may be reproduced without expressed written consent from 2 Minute Medicine, Inc. Inquire about licensing here. No article should be construed as medical advice and is not intended as such by the authors or by 2 Minute Medicine, Inc.
