Authors
Joseph W Hogan
Published in
NEJM AI. Pages AIe2601363. Nov 06, 2026. Epub Nov 06, 2026.
Abstract
AbstractThe use of risk prediction models to guide clinical decision-making is becoming more common in practice. Evaluating their effectiveness is often done after implementation and without the benefit of an experimental design. Although evaluations based on observational data can yield useful information, they are prone to multiple sources of potential bias. Randomized trials, such as the one conducted by Barnado and colleagues and published in this issue, represent the highest standard of evidence for establishing causal effects of interventions. Their study provides evidence that model-guided referral to early rheumatology evaluation can expedite the diagnosis of autoimmune disease. The study also demonstrates the feasibility of implementing trials in care settings to generate rigorous, real-world assessments of model- and AI-guided clinical workflows on patient-level end points. Using predictive models impacts patient outcomes by prompting a sequence of decisions or interventions. In the AUtoimmune Risk Assessment (AURA) study, for example, risk classification is the first step; the second step is selective referral for expedited rheumatology evaluation. Generating an earlier diagnosis depends on whether the model is flagging the right patients for early referral, and whether early referral itself yields more diagnoses. Building on the contribution of Barnado and colleagues, this Editorial describes a design that can isolate the separate effects of each sequential step of the overall intervention. More generally, it highlights the fundamental importance of trial design for understanding not just whether a model-guided intervention works, but also how.
PMID:
42993021
Bibliographic data and abstract were imported from PubMed on 06 Nov 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 8
- Comments 0