Physician Decision-Making in Acute Ischemic Stroke with and without a Guideline-Encoded Clinical Decision Support System
- 1UPMC
Abstract
Background: Applying extensive, frequently revised guidelines at the point of care is difficult. We evaluated whether acute ischemic stroke decisions made with LUCID, a clinical decision support system encoding the 2026 American Heart Association/American Stroke Association guideline, were more often guideline-concordant than decisions made without it. Methods: In this pilot vignette study, clinicians completed 8 stroke vignettes (24 questions) alone (Arm A) or with access to LUCID (Arm B). Questions bundling recommendations of different classes were split into nodes (29 decision points). The primary outcome, full guideline concordance, was compared between decision points made with and without LUCID using cluster-bootstrap risk differences (RDs). Decision confidence and workload (NASA Task Load Index; missing scores multiply imputed) were compared between arms. Results: Fifteen clinicians (Arm A, 5; Arm B, 10) contributed 351 decision points: 151 with LUCID and 200 without. Full concordance was 99.3% (150/151) with LUCID versus 71.0% (142/200) without (RD, 28.3 [95% CI, 21.7 to 34.9] percentage points; P<0.001), and at least partial concordance was 100% versus 93.5% (RD, 6.5 [2.3 to 10.2]; P=0.001). Concordance with versus without LUCID was 100% versus 67.9% for Class 1, 100% versus 79.3% for Class 3: Harm, and 100% versus 74.3% for Class 3: No Benefit recommendations. Commission errors occurred at 0 of 66 versus 10 of 64 Class 3 decision points (P<0.001). Decision confidence was numerically higher in Arm B than in Arm A (median, 5 versus 4; P=0.24). Workload was lower in Arm B for mental demand (P=0.045), with borderline reductions in temporal demand (P=0.090) and composite workload (P=0.084). Conclusions: In this pilot, decisions made with a guideline-encoded decision support system were almost uniformly guideline-concordant, including for recommendations against harmful interventions, and workload tended to be lower with system access. Confirmation in a larger study with blinded outcome assessment is warranted.
Related
- Development and evaluation of Retrieval-Augmented Generation (RAG) and GraphRAG for guideline-based clinical decision support in chronic kidney disease
- EM-BENCH: A Reproducible Framework for Benchmarking Guideline-Concordant Clinical Decision Making in Large Language Models
- Effect of guideline-assisted prompting on large language model performance in medication-related osteonecrosis of the jaw–related endodontic decision-making
- Enhancing Cardiac Rehabilitation Referral at Discharge Using EHR-Based Decision Support for Guideline-Eligible Patients
- Artificial intelligence versus physician decision-making in acute heart failure: a comparative evaluation of four contemporary AI systems
BibTeX
@misc{doheim2026physician,
title = {{Physician Decision-Making in Acute Ischemic Stroke with and without a Guideline-Encoded Clinical Decision Support System}},
author = {M. F. F. Doheim and R. G. Nogueira},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.26.26366524},
url = {https://www.medrxiv.org/content/10.64898/2026.10.26.26366524v1},
note = {CC BY-ND},
}
RIS
TY - UNPB
TI - Physician Decision-Making in Acute Ischemic Stroke with and without a Guideline-Encoded Clinical Decision Support System
AU - M. F. F. Doheim
AU - R. G. Nogueira
PY - 2026
DA - 2026/10/26
PB - medRxiv
DO - 10.64898/2026.10.26.26366524
UR - https://www.medrxiv.org/content/10.64898/2026.10.26.26366524v1
AB - Background: Applying extensive, frequently revised guidelines at the point of care is difficult. We evaluated whether acute ischemic stroke decisions made with LUCID, a clinical decision support system encoding the 2026 American Heart Association/American Stroke Association guideline, were more often guideline-concordant than decisions made without it. Methods: In this pilot vignette study, clinicians completed 8 stroke vignettes (24 questions) alone (Arm A) or with access to LUCID (Arm B). Questions bundling recommendations of different classes were split into nodes (29 decision points). The primary outcome, full guideline concordance, was compared between decision points made with and without LUCID using cluster-bootstrap risk differences (RDs). Decision confidence and workload (NASA Task Load Index; missing scores multiply imputed) were compared between arms. Results: Fifteen clinicians (Arm A, 5; Arm B, 10) contributed 351 decision points: 151 with LUCID and 200 without. Full concordance was 99.3% (150/151) with LUCID versus 71.0% (142/200) without (RD, 28.3 [95% CI, 21.7 to 34.9] percentage points; P<0.001), and at least partial concordance was 100% versus 93.5% (RD, 6.5 [2.3 to 10.2]; P=0.001). Concordance with versus without LUCID was 100% versus 67.9% for Class 1, 100% versus 79.3% for Class 3: Harm, and 100% versus 74.3% for Class 3: No Benefit recommendations. Commission errors occurred at 0 of 66 versus 10 of 64 Class 3 decision points (P<0.001). Decision confidence was numerically higher in Arm B than in Arm A (median, 5 versus 4; P=0.24). Workload was lower in Arm B for mental demand (P=0.045), with borderline reductions in temporal demand (P=0.090) and composite workload (P=0.084). Conclusions: In this pilot, decisions made with a guideline-encoded decision support system were almost uniformly guideline-concordant, including for recommendations against harmful interventions, and workload tended to be lower with system access. Confirmation in a larger study with blinded outcome assessment is warranted.
ER -