Published on
In a study published in JAMA Network Open, researchers developed an artificial intelligence (AI) model and tested its ability to assist in triaging referrals from primary care to specialized care. Evaluated against a reference standard, the AI tool achieved good overall accuracy of 0.716 (95% confidence interval [CI], 0.694-0.737) of referrals with high specificity (0.801, 95%CI, 0.777-0.822) but lower sensitivity (0.542, 95%CI, 0.501-0.582). In other words, the AI model correctly classified 71.6% of referrals while demonstrating high specificity. Compared to human triage, the AI model had a higher false-negative rate, however, a negative decision in this situation would translate to the need for further human evaluation as opposed to missing a clinical diagnosis, for example. Of the 45,039 total cases studied, 63% required additional information on the first evaluation. Specialty-specific performance varied with rheumatology showing the best results. According to the authors of this Brazilian study, “the AI model performance was moderate, with clear gains in specificity compared with the current approach when both were compared with a consensus reference standard.” The reference standard was the consensus of 2 physicians with extensive experience, according to the study.
Deep dives: As with most AI testing, the results suggest AI can be leveraged to supplement and support human clinical judgment. The promise is that AI can reduce workloads for clinicians by combing through patient data and standards of care, then quickly suggesting diagnoses or next steps, which the clinicians can take under consideration.
Read More
- Can Artificial Intelligence Provide Real Value in Triaging Urgent Care Patients?
- Embracing AI In Medicine The Role Of Large Language Models In Healthcare