Archetype 4
Triage / Priority CLIOH
map case features → an urgency score for live prioritization (OR booking, ED throughput, transport).
- Evidence standard
- external
- What you compare
- Real-time case profiles (structured intake, 12–25 features)
- Panel
- On-call attendings + senior residents who actually make the call
- Output
- Interval urgency score / dashboard widget (case features → score)
- Validation
- Prospective outcome correlation + drift monitoring
What it's for
A CLIOH study that produces a real-time decision-support tool mapping case features to an urgency score, for prioritizing competing cases.
"When two cases compete for the same slot, who goes first — and can a covering resident reproduce the attending's call?"
When to use it
(a) Resource constraints force prioritization; (b) trainees/covering staff make these calls inconsistently; (c) operational delay is measurable and consequential; (d) institutional buy-in for a dashboard / EHR widget; (e) prospective outcome data can be collected.
When not to
(a) Triage already handled by a validated tool (ESI, MeNTS) with good local performance; (b) the bottleneck is resources, not the decision; (c) you're ranking criteria, not scoring real cases → that's Discovery §1 (where OR/ER-Priority live today); (d) you need a categorical grade, not a rank → Classification §3.
What you get
A weighted-score formula or EHR/dashboard widget (two cases side-by-side → recommended order), with a confidence interval and an audit trail. Not a clinical decision tool until separately IRB-reviewed + prospectively validated + SaMD-grade QMS (DR caveat — binding).
A real example
- CORTICES OR-Priority (k=17 urgency factors today, Discovery §1) → deployed on real OR-board cases = OR Triage CLIOH.
- CORTICES ER-Priority (k=16, Discovery §1) → on-call real-case prioritization = ED/ER Triage CLIOH.
- Interurban 2026 OR-Triage — the by-design Strength-Spread slate; the demonstrator that bridges Discovery → Triage.
- Maturation produces Hybrid H4 (Triage × Risk-Factor): covariate-structured BT yields the implicit risk-factor weights as a second deliverable.
Validation
Prospective outcome correlation; head-to-head vs the incumbent triage protocol; downstream OR-throughput / LOS metrics; ongoing calibration monitoring (dynamic BT) so the score doesn't silently drift.
Common pitfalls
(a) Feature drift between training and deployment; (b) gaming the intake; (c) over-automation that overrides bedside judgment; (d) failing to monitor calibration over time (use dynamic BT); (e) confusing this with Discovery — if you're ranking factors not scoring cases, you're still in §1.
How it works — show me the method
Item selection. Real-time case profiles defined by a structured intake form (12–25 operationally-available features). The features are fixed; the cases vary. (Contrast Discovery, where the features themselves are the items being ranked.)
What you compare. Operationally available features only — vitals, anatomic location, mechanism, time-since-injury, NV status, comorbidities. If a feature isn't available at the moment of the call, it can't be in the score.
Comparison structure. Covariate-augmented / structured BT (Springall 1973; Cattelan 2012): worth(case) = f(case features). Pairwise "which case should go first?", bare-vote, no ties (ADR-CLIOH-03). BIBD over case pairs; ATRD on contested pairs. Longitudinal → dynamic / time-varying BT (Cattelan, Varin & Firth 2013) to monitor non-stationarity (practice drift). (DR's active-learning pair selection = future enhancement.)
Panel. On-call attendings + senior residents who actually make the call; 10–20 raters. Including the eventual users improves face validity and adoption.
External ground truth. Required. Validate the score against post-hoc outcomes — Clavien-Dindo, mortality, LOS, time-to-OR adherence. Precedent: Buehler et al. Langenbeck's Arch Surg 2025 (triage-level protocol violation → higher complication rate, 64% vs 43%).
Statistical backbone. Structured/regression BT with item covariates → output is a logistic prediction model interpretable as a worth score (bedside-usable). Frequentist primary; dynamic BT for drift; Bayesian where covariate count is high relative to data. Concurrent-validity precedent: Tu et al. 2020 (COVID surgical prioritization, inter-rater Spearman > 0.80).
For researchers — reporting, IRB & grant language
Reporting standards. TRIPOD / TRIPOD-AI; SQUIRE (if framed as QI). Make the QI-vs-research distinction early (changes IRB path). LitGuard + DLRP.
IRB / ethics. Human-subjects (prospective patient data). QI-vs-research determination up front. No PHI in the elicitation instrument; deployment-phase data governance is separate and heavier. No unverified IRB claim in the UI.
Grant-application language.
"Building on validated COVID-era surgical prioritization scoring (Tu et al. 2020, J Minim Invasive Gynecol, inter-rater Spearman > 0.80) and emergency-surgery triage outcome data (Buehler et al. 2025, Langenbeck's Arch Surg), Triage CLIOH produces a pediatric-orthopaedic-specific interval urgency score via covariate-structured Bradley–Terry modeling of expert pairwise prioritizations of real case profiles, validated prospectively against length-of-stay and complication outcomes and monitored for drift with time-varying Bradley–Terry."