Archetype 8
Patient-Centered Preference CLIOH
measure population-level patient/family preference weights over treatment attributes and burdens, to inform shared decision-making and labeling.
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 — patient values ARE the answer (normative; no external standard — preferences are not 'right' or 'wrong')
- What you compare
- Treatment attributes, burdens, outcomes-as-experienced
- Panel
- Patients / families (population sample, not an expert panel)
- Output
- Population (and subgroup) preference weights
- Validation
- Test-retest; subgroup/latent-class structure; internal consistency — not outcome-validation
What it's for
A CLIOH-family study that elicits patient/family preferences over treatment attributes and burdens — via a population-scale choice instrument — and produces interval-scale preference weights for shared decision-making, decision aids, or patient-preference information.
"What do patients and families actually value and trade off when choosing among treatments — measured at population scale, with subgroups, not assumed by clinicians?"
When to use it
(a) A genuine preference-sensitive decision (multiple reasonable options with different attribute/burden profiles); (b) patient values should drive the weights, not clinician proxies; (c) you need population-level, generalizable estimates (and subgroups); (d) a regulatory or shared-decision use case justifies the larger-N machine; (e) the population-scale modality exists or is being built.
When not to
(a) You want clinician judgment, not patient values (most other archetypes); (b) you're ranking trial endpoints with a mixed panel → Outcome-Importance §6; (c) a small expert panel is the right source (this is the wrong archetype — §8 is patient-population by definition); (d) the population modality isn't built and patient-preference data is essential — scope the modality first (the aspirational blocker).
What you get
Population and subgroup preference weights (part-worth utilities or BWS scores), usable in a shared-decision aid, a patient-preference-information submission, or device labeling/IFU. Aggregate-only display; no PHI; no unverified regulatory claim.
A real example
- Pediatric scoliosis: surgery vs bracing family preferences (candidate / aspirational): elicit how families trade off correction, recovery, activity restriction, scar, and recurrence risk between operative and bracing pathways. Reusable artifact: a validated attribute set + population preference weights feeding a shared-decision aid.
- Adolescent assent (BWS Case 2): an age-appropriate preference instrument bringing the adolescent's own voice into assent decisions — a distinctive, high-value sub-case.
Validation
Test-retest; internal consistency / dominated-choice checks; latent-class stability; external validity via demographic representativeness of the sample. Pre-register attributes, levels, and the analysis plan. Not outcome-validation — there is no outcome to validate preferences against.
Common pitfalls
(a) Using an expert panel — §8 is patient-population by definition; a 20-clinician panel answers a different question. (b) Forcing it onto the pairwise expert instrument instead of building the DCE/BWS modality (the aspirational blocker). (c) Averaging away preference heterogeneity instead of modeling latent classes. (d) Treating preferences as right/wrong or trying to outcome-validate them. (e) Un-validated lay attributes or a dominant attribute that collapses the choice. (f) Underpowered sampling that misses subgroups. (g) Confusing it with Outcome-Importance §6 — §6 ranks trial endpoints with a mixed panel; §8 estimates treatment-attribute preferences with patients.
How it works — show me the method
Item selection. Items are treatment attributes/levels (efficacy, recovery time, scar, recurrence risk, activity restriction, cost/burden), defined as in a DCE attribute-and-levels design or a BWS Case 2 profile. Lay-language and cognitive-interview validation are mandatory before fielding. Coverage of the realistic attribute space matters more than "anchoring" in the expert-panel sense.
What you compare. The construct is preference / value, explicitly not clinical importance or outcome likelihood. Attributes must be mutually intelligible to lay respondents and span the real trade-offs of the decision. Avoid dominant attributes that collapse the choice.
Comparison structure. Field standard is DCE or BWS Case 2/3, not the lab's bare-vote pairwise instrument — this is the central modality difference. Where a pairwise rendering is used, the locked single-step back button (ADR-CLIOH-02) and no-ties discipline (ADR-CLIOH-03) carry over to the UI, but the expert-panel mechanics (Unanimity Gate, ATRD targeted-rationale) do not map to an anonymous population sample and are out of scope for this modality. Defining the right CLIOH-compatible population instrument is the core aspirational task.
Panel. Patients / families, sampled to represent the relevant population, with planned subgroups (age, severity, prior experience). N≈100–500 (DR range) — a recruitment machine the lab has never run. For pediatric decisions, adolescent assent is a distinct, important sub-case (BWS Case 2 for assent).
External ground truth. Internal — and absolutely so. Patient preferences are not "right" or "wrong"; there is no external standard to validate against. Validate by test-retest, internal consistency, and the stability/interpretability of subgroup (latent-class) structure. The claim is purely normative/descriptive of what patients value.
Statistical backbone. The estimation backbone is preference modeling — conditional/mixed logit for DCE, or BWS scaling — which is outside the lab's current choix-based pairwise pipeline and part of what the modality build must add. Latent-class / mixture models are central here (distinct patient preference segments are the typical, important finding). Bayesian hierarchical estimation is natural at population scale. Report heterogeneity as the headline.
For researchers — reporting, IRB & grant language
Reporting standards. ISPOR good-research-practices for conjoint/DCE and for BWS; PREFER-style patient-preference reporting; CHERRIES for the web survey. Position against the broader Best–Worst Scaling healthcare literature (Schuster, Crossnohere, Campoamor, Hollin & Bridges, J Choice Modelling 2024 — 195 health applications; Hollin et al., Pharmacoeconomics 2022 systematic review, PROSPERO CRD42020209745). LitGuard + DLRP.
IRB / ethics. Human-subjects (patient/family participants), with heightened obligations: informed consent, accessibility, and — for pediatric assent — age-appropriate design and assent/consent structure. QI-vs-research determination up front. No PHI; aggregate-only. The patient-facing instrument carries a real UX/ethics bar. No unverified IRB claim in the UI.
Grant-application language.
"Patient-preference information is increasingly central to device evaluation and shared decision-making, with Best–Worst Scaling now applied across 195 documented health contexts (Schuster, Crossnohere, Campoamor, Hollin & Bridges, Journal of Choice Modelling 2024; Hollin et al., Pharmacoeconomics 2022). Patient-Centered Preference CLIOH extends the framework to a population-scale, lay-language choice instrument (discrete-choice experiment or Best–Worst Scaling Case 2) that estimates interval-scale patient and family preference weights — including for pediatric assent — over treatment attributes and burdens, with latent-class modeling to surface distinct preference subgroups, informing shared-decision aids and patient-preference submissions." (Aspirational pending the population-scale modality build.)