Hybrid H2
Discovery × Outcome-Importance CLIOH
combine variable discovery (Discovery §1) with endpoint ranking (Outcome-Importance §6) to produce a complete multicenter-trial-design package — which variables to measure and which outcomes to power on.
- Evidence standard
- internal (both deliverables) — expert/mixed-panel consensus is the standard for the variables and the endpoints
- Panel
- Discovery: expert; Outcome-Importance: mixed clinician + patient (the modality gap)
- Validation
- Test-retest; panel-composition sensitivity; convergence with later trial data / existing COS work
What it's for
A paired CLIOH exercise: a Discovery study ranks the candidate variables/predictors for a condition, and an Outcome-Importance study ranks the candidate trial endpoints, together specifying both the measurement set and the primary/secondary endpoint hierarchy for a future trial.
Designing a trial requires answering two questions at once — what to measure and what to optimize. Discovery answers the first; Outcome-Importance the second. Delivered together, the output is a ready-to-protocol trial-design package rather than two disconnected consensus papers. Because core-outcome-set initiatives all run this as Delphi, a BT-based H2 offers interval-scale endpoint weights they currently lack.
"For a trial in this condition, which variables should we measure (and which predictors matter), and which outcomes should it be powered on — ranked, with the patient voice included on the endpoints?"
When to use it
Use when designing a (multicenter) trial from a still-murky evidence base and you need both the variable set and the endpoint hierarchy, with patients weighing in on outcomes. Don't when only one half is needed (use the parent), when a validated core outcome set already exists for the condition, or — for now — when you need the Outcome-Importance half executed before the mixed-panel modality is built (scope that modality first; Taxonomy §7 item 5).
What you get
- Discovery: ranked variable/predictor hierarchy (the trial's measurement set).
- Outcome-Importance: ranked endpoint set with interval-scale importance weights (the trial's primary/secondary endpoint hierarchy), patient-weighted.
A real example
Pediatric SCFE trial-design package (candidate / aspirational). Discovery ranks the severity/causal variables for SCFE (buildable now); Outcome-Importance ranks SCFE trial endpoints with a clinician+patient panel (awaits the modality). Together: a complete design package for a multicenter SCFE trial. Reusable artifact: the paired variable + endpoint hierarchies. (The DR's worked example pairs a Discovery on SCFE with "CLIOH-Pediatric-SCFE outcome importance.")
Validation
Test-retest + panel-composition sensitivity on each scale; convergence with existing COS/OMERACT work where available; downstream check that trials adopting the package measure what mattered. Pre-register both elicitations.
Common pitfalls
(a) Treating the endpoint ranking as predictive — it's normative. (b) Conflating "causally important" (Discovery) with "important to measure as an endpoint" (Outcome-Importance) — they're different questions; keep the two scales distinct. (c) Running the §6 half on the expert instrument without the patient modality, which would defeat the patient-alignment purpose. (d) Averaging away clinician–patient disagreement instead of reporting it. (e) Strength-Spread failure on the variable half. (f) Forgetting this is two linked studies with a single design purpose.
How it works — show me the method
Comparison structure. Both halves use bare-vote, binary forced-choice pairwise, NO ties (ADR-CLIOH-03), BIBD allocation, ATRD Round 2 (ADR-CLIOH-04), single-step back button (ADR-CLIOH-02). The Discovery half runs on the expert HTML instrument now. The Outcome-Importance half needs the mixed clinician+patient instrument (lay-language, possibly population-scale) — the modality to be scoped before execution. Strength-Spread §1.1 binds hard on the Discovery (variable) half (no natural anchors).
Panel. Discovery: expert panel, 15–30. Outcome-Importance: mixed clinician + patient — the composition that makes the endpoint ranking patient-aligned and the reason a new recruitment/instrument modality is required.
Ground truth. Internal for both. The panel consensus is the standard; the claim is normative ("these are the variables/endpoints the field and patients prioritize"), not predictive. Validate by test-retest, panel-composition sensitivity, and convergence with later empirical trial data.
Statistical backbone. Frequentist BT MLE via choix (bootstrap CIs, Firth) for each half → a variable worth scale and an endpoint worth scale. Bayesian-hierarchical BT is conditional (subgroups; partial pooling across clinician vs patient sub-panels in the §6 half — where modeling the two constituencies separately is itself a finding). Report clinician–patient divergence on endpoints rather than averaging it away.