ISPOR Conjoint Analysis / Discrete-Choice Experiment Good Research Practices

What it is

— The ISPOR Conjoint Analysis / Discrete-Choice Experiment (DCE) Good Research Practices is a three-report series issued by task forces of ISPOR (the Professional Society for Health Economics and Outcomes Research). It is the field's de facto standard for designing, conducting, analyzing, and reporting stated-preference studies in health. The series comprises: (1) Bridges et al. 2011 — a 10-item conjoint-analysis-applications checklist spanning the research question, attribute and level identification, choice-task and experimental design, preference-elicitation format, instrument design, data-collection plan, statistical analysis, and reporting of results and limitations; (2) Johnson et al. 2013 — good practices for constructing experimental designs (full vs partial profiles, opt-out/status-quo alternatives, D-efficiency, fractional-factorial and Bayesian designs, blocking, and the number of choice tasks); and (3) Hauber et al. 2016 — good practices for the statistical analysis of DCE data (random-utility theory, conditional/multinomial logit, mixed/random-parameters logit, latent-class models, hierarchical Bayes, and tests of dominance, transitivity, and monotonicity). ISPOR maintains the series among its good-practices reports for outcomes research. It is a design-and-reporting good-practice standard — not a regulatory mandate.

When to use

— Apply this series whenever the deliverable is a quantitative measurement of how patients, caregivers, clinicians, or the public trade off attributes of a health intervention — a DCE, best-worst scaling (object/profile/multiprofile), ranking, or rating-based conjoint task. Typical decision contexts: generating patient-preference information for an FDA benefit-risk / Patient Preference Information submission; supplying preference weights to a multi-criteria decision analysis (MCDA) or benefit-risk framework in an HTA/payer dossier; quantifying acceptable risk or willingness-to-pay/willingness-to-accept for a value argument; and peer-reviewed publication of a stated-preference study. Decision rule for which report governs which step: use Bridges 2011 as the overarching checklist and for attribute/instrument development and reporting; reach for Johnson 2013 when specifying the experimental design and choice-set generation; and apply Hauber 2016 when choosing and reporting the choice model. This series governs stated-preference work; it is the wrong instrument for revealed-preference analyses of real-world choices captured in claims or EHR data, which fall under observational-study reporting guidelines (STROBE/RECORD-PE) and causal-inference designs instead.

What it requires

— The substantive domains the checklist enforces, framed for a defensible preference study: (1) a clear research question and decision context that the elicited preferences must inform; (2) attribute identification grounded in qualitative formative work (literature review, patient/clinician interviews, focus groups) so the attribute set is ecologically valid and not analyst-imposed; (3) attribute-level selection that is plausible, non-dominated, and spans a clinically meaningful range; (4) choice-task construction — alternatives per task, full vs partial profiles, and an explicit decision on opt-out/status-quo options; (5) experimental design with a stated efficiency criterion (D-efficiency, fractional-factorial or Bayesian designs), blocking, and a justified number of choice tasks (Johnson 2013); (6) preference-elicitation format (DCE vs BWS vs ranking/rating) matched to the cognitive demands of respondents; (7) instrument design with cognitive pretesting, framing checks, and pilot testing; (8) a data-collection plan specifying mode, sampling frame, and sample-size/power justification; (9) statistical analysis using a random-utility-consistent model, explicit testing for preference heterogeneity (mixed logit, latent-class), and diagnostic checks for dominance, monotonicity, and attribute non-attendance (Hauber 2016); and (10) transparent reporting of results, internal/external validity, and limitations.

When NOT to use — limitations and common misapplications

— This is a reporting and design good-practice series, not a risk-of-bias instrument and not a quality score: a fully completed Bridges checklist does not certify that a DCE is unbiased or that its preference weights are valid. Concrete failure modes: (a) treating stated preferences as if they were causal descriptions of real-world behavior — the stated-vs-revealed-preference gap means a DCE predicts hypothetical choices, not market or adherence behavior; (b) skipping qualitative attribute development and imposing an analyst-chosen attribute set, yielding ecologically invalid results no design efficiency can rescue; (c) reporting only pooled marginal utilities or a single conditional-logit model without testing for preference heterogeneity (mixed logit or latent class), masking clinically important subgroups; (d) deriving willingness-to-pay from a cost attribute that was poorly specified or non-linear, producing unstable monetary estimates; (e) "checklist-as-theater" — ticking items in an appendix while the underlying design (dominated alternatives, implausible levels, no pretesting) is weak; and (f) using the wrong guideline family — applying STROBE, RECORD-PE, or a causal observational-design framework to a stated-preference study, or conversely using this DCE series to govern a revealed-preference claims/EHR analysis.

How it maps to this catalog

— The implementing concept in this repository is preference-study, which carries the operational detail for DCE/BWS design and analysis that this guideline requires. Attribute identification and instrument development map to the patient-reported-outcome and qualitative concepts: pro-development and pro-validation (formative development and psychometric/instrument testing), qualitative-interview and qualitative-synthesis (the formative work that grounds the attribute set), and pro-rwe for downstream real-world preference evidence. The data-collection-plan and sample-size requirements map to sample-size-power-precision-rwe. Preference weights that feed value and benefit-risk arguments connect to hrqol (utility-adjacent valuation) and to the health-economic concepts when DCE outputs populate an MCDA or benefit-risk model. Applied note: in an HTA/payer dossier, DCE-derived attribute importance weights are most often consumed by an MCDA or structured benefit-risk framework rather than by a claims/EHR fitness-for-use assessment — so the chain of evidence runs from qualitative attribute development (pro-development, qualitative-interview) through an efficient experimental design and a heterogeneity-aware choice model (preference-study) to a transparent benefit-risk or value narrative. For an FDA Patient Preference Information submission, the same chain must additionally document the maximum acceptable risk or risk-tolerance estimate and its uncertainty, which depends directly on the experimental-design and statistical-analysis good practices in Johnson 2013 and Hauber 2016.