Evaluation and Comparison of Latent Health Risk Prediction Models for Clinical Triage: Protocol for a Mixed Methods Study

Clinical Snapshot

Key Findings

• P Value: Not yet available

• Effect Size: Not yet available

• Primary Outcome: Protocol paper — no results available. Planned primary outcome: Spearman rank correlation between FI-lab/ETHOS-ARES severity rankings and clinician consensus judgments across 20 ED case vignettes reviewed by ≥30 clinicians

• Nnt Or Sensitivity: Not applicable at protocol stage; no diagnostic accuracy or therapeutic outcome metrics are planned in the current protocol scope

• Confidence Interval: Not yet available; protocol does not pre-specify CI reporting for rank correlations

Clinical Application

The study design is feasible within a tertiary UK hospital setting. The 20-case vignette format is practical for busy clinicians. However, scalability of the evaluation framework to other centres, specialties, or real-time EHR-integrated environments remains undemonstrated. The ETHOS-ARES model requires EHR integration infrastructure that may not be available in many hospitals. Australian relevance is limited at this stage. Australian EDs operate under the Australasian Triage Scale (ATS, 5-category), which differs from UK triage frameworks. The MIMIC-IV-ED dataset reflects US clinical practice patterns, laboratory reference ranges, and documentation styles that may not translate to Australian EHR systems (e.g., those using Cerner, Epic, or state-based systems such as ieMR in Queensland). The FI-lab tool's laboratory deficit thresholds would require local validation against Australian reference intervals. Neither tool is currently TGA-listed as a Software as a Medical Device (SaMD), and PBS considerations are not applicable. RACGP and ACEM (Australasian College for Emergency Medicine) guidelines do not currently reference either tool. Any future Australian implementation would require prospective local validation and ACEM endorsement processes. Emergency department clinicians considering adoption of latent health risk scoring tools (FI-lab or transformer-based foundation models) as adjuncts to clinical triage decision-making in adult ED presentations

Abstract

BACKGROUND: Clinical triage requires integrating multiple information sources to identify patients at risk of deterioration. Tools capturing global health assessments beyond disease-specific scores are being developed using either bottom-up aggregation of simple indicators or top-down machine learning from large datasets. Their alignment with expert clinical judgment remains poorly characterized. OBJECTIVE: This study evaluates 2 latent health measurement approaches: Frailty Index-laboratory, a transparent bottom-up tool aggregating laboratory abnormalities via deficit accumulation theory, and ETHOS-ARES (Enhanced Transformer for Health Outcome Simulation-Adaptive Risk Estimation System), a transformer-based foundation model generating multidimensional patient representations from electronic health records. We assess whether each tool's severity rankings align with clinical consensus and whether they offer utility in triage decisions. METHODS: In this 3-phase mixed methods study, at least 30 clinicians across hospital specialties reviewed 20 emergency department presentations derived from Medical Information Mart for Intensive Care IV-Emergency Department. Phase 1 compared unaided clinician severity and urgency judgments against model outputs using Spearman rank correlation, with a Turing-inspired indistinguishability test assessing whether model rankings fell within the distribution of clinician assessments. Phase 2 allocated clinicians to receive Frailty Index-laboratory or ETHOS-ARES outputs, measuring anchoring effects via within-person pre-post comparisons and exploring clinical utility through semistructured interviews analyzed using the Framework Method. RESULTS: Ethics approval was granted in June 2025 (KCL Research Ethics Office; MRSP-24/25-48707). Recruitment began in October 2025 (32 clinicians recruited as of manuscript submission), with data collection expected to be completed in January 2026 and analysis planned for March or April 2026. CONCLUSIONS: This study will quantify model-clinician agreement, measure anchoring effects, and generate qualitative insights on utility, trust, and adoption. The findings will inform the implementation of latent health measurement tools in clinical practice and provide a framework for the early-stage evaluation of artificial intelligence-based clinical decision support systems.

References

1. 1.Roberts, M., Pelkonen, O., Shamsutdinova, D., Sanderson, S. C., Renc, P., & Logan Ellis, H. (2026). Evaluation and comparison of latent health risk prediction models for clinical triage: Protocol for a mixed methods study. JMIR Research Protocols. https://doi.org/10.2196/85437