Archetype 6
Outcome-Importance CLIOH
rank which outcomes a trial (or a condition's management) should measure — producing an interval-scale, patient-aligned endpoint hierarchy.
On the roadmap. Roadmap — needs a survey modality the lab has not built yet. We are not offering this archetype today — it is documented here for completeness.
- Evidence standard
- internal — the mixed clinician+patient consensus IS the answer (normative, patient-aligned; not outcome-validated)
- What you compare
- Candidate outcomes / trial endpoints
- Panel
- Mixed clinician + patient (the modality gap)
- Output
- A ranked endpoint set with interval-scale importance weights
- Validation
- Test-retest; panel-composition sensitivity; clinician–patient divergence reported; convergence with existing COS work
What it's for
A CLIOH study in which a mixed clinician + patient panel makes pairwise judgments over candidate outcomes, and a Bradley–Terry model produces an interval-scale importance ranking of trial endpoints (or outcomes to track in routine care).
"Of all the outcomes we could measure, which matter most — to clinicians and to patients — so the trial is powered on what counts?"
When to use it
(a) Designing a trial or core outcome set where endpoint choice is contested; (b) patient priorities should shape what's measured, not just clinician convention; (c) you want interval-scale importance weights rather than Delphi's coarse "retain/drop" vote; (d) the candidate outcome list is bounded and comparable; (e) the mixed-panel modality exists or is being built.
When not to
(a) A validated core outcome set already exists for the condition (adopt it); (b) you're ranking causal/predictor variables, not outcomes → Discovery §1 (note: pairing the two = Hybrid H2, the trial-design package); (c) you're eliciting treatment-attribute preferences for shared decisions → Patient-Centered §8; (d) the patient modality isn't built yet and patient input is essential (scope the modality first — that's the aspirational blocker).
What you get
A ranked endpoint set with interval-scale importance weights and explicit clinician–patient comparison, suitable as a trial's primary/secondary endpoint hierarchy or a core outcome set. Aggregate-only; no PHI; no unverified claims.
A real example
- Pediatric SCFE outcome-importance (candidate / aspirational): rank the candidate outcomes for a SCFE trial (e.g., AVN, chondrolysis, function, pain, return-to-activity) across clinicians and patients/families. Paired with a SCFE Discovery study, this is Hybrid H2 — a full trial-design package. Reusable artifact: a patient-weighted endpoint hierarchy.
Validation
Test-retest; drop-one-out and drop-one-constituency sensitivity; convergence with published COS for the condition; downstream check that trials adopting it measure what mattered to patients. Pre-register. The lay-language items themselves need cognitive-interview validation before fielding.
Common pitfalls
(a) Treating the ranking as predictive — it's normative. (b) Running it on the expert instrument without the patient modality — defeats the patient-alignment purpose (the aspirational blocker). (c) Averaging away clinician–patient disagreement. (d) Un-validated lay-language items that clinicians and patients interpret differently. (e) Strength-Spread failure — no outcome anchors → compressed importance scale. (f) Confusing importance-to-measure (here) with causal-importance (Discovery §1) or with treatment-attribute preference (Patient-Centered §8).
How it works — show me the method
Item selection. Candidate outcomes/endpoints, drawn from literature + existing COS work + patient input. Strength-Spread §1.1 applies (no natural endpoints to the importance scale) — include clearly-critical and clearly-peripheral outcome anchors so the ranking discriminates. The lay-language rendering of each outcome must be validated with patients before fielding.
What you compare. The elicited worth is importance for decision-making / trial-powering, not frequency or measurability. Outcomes must be phrased so both clinicians and patients judge the same construct — the central design challenge of the mixed modality.
Comparison structure. Bare-vote, binary forced-choice pairwise "which outcome matters more?", NO ties (ADR-CLIOH-03); BIBD allocation; ATRD Round 2 (ADR-CLIOH-04); single-step back button (ADR-CLIOH-02) — these locked UI/allocation rules carry over. Modality caveat: the expert-panel mechanics (the Unanimity Gate's blinded-opt-in unblinding; ATRD's expert-rationale framing) need re-design for a lay/patient population; that re-design is part of the modality-scoping work, not a settled mapping.
Panel. Mixed clinician + patient. The clinician sub-panel can run on the existing instrument; the patient sub-panel needs the lay-language, possibly larger-N modality. Model the two constituencies so their divergence is visible (see §10).
External ground truth. Internal. The consensus is the answer — this is a values/priorities judgment, not an outcome prediction, so there is no external gold standard. Validate by test-retest, panel-composition sensitivity, and convergence with existing COS/OMERACT work. The claim is normative/patient-aligned, not predictive.
Statistical backbone. Frequentist BT MLE via choix (bootstrap CIs, Firth) → an endpoint importance scale. Bayesian-hierarchical BT is well-motivated here for partial pooling across clinician vs patient sub-panels — and modeling the two separately is itself a finding (where do patients and clinicians disagree on what matters?). Report the divergence; do not average it away. Latent-class structure can surface patient subgroups with distinct priorities.
For researchers — reporting, IRB & grant language
Reporting standards. COS-STAR (core-outcome-set reporting); position against OMERACT Filter 2.1 (Boers et al. 2019, PMID 30770515) and the *. citation in the LitGuard sweep) and ICHOM standard sets — all panel-based outcome-prioritization frameworks that currently use Delphi, not BT. LitGuard + DLRP.
IRB / ethics. Human-subjects (patient participants); informed consent and accessibility obligations; the patient-facing modality carries its own ethics/UX bar. QI-vs-research determination up front. No PHI; aggregate-only. No unverified IRB claim in the UI.
Grant-application language.
"Core-outcome-set development — defining which outcomes a trial should measure — is dominated by the OMERACT Filter (Boers et al., J Rheumatol 2019, PMID 30770515), the. Outcome-Importance CLIOH applies Bradley–Terry modeling to a mixed clinician + patient panel's pairwise judgments of candidate endpoints, producing an interval-scale, patient-aligned importance hierarchy that distinguishes how patients and clinicians weight outcomes — a methodological advance over Delphi for trial-endpoint and core-outcome-set selection." **