Discrete Choice Experiment Instrument (DCE)

Abstract

The Discrete Choice Experiment Instrument (DCE) developed by dosReis et al. (2023) is an advanced stated-preference quantitative measurement tool engineered to evaluate patient preferences, utility weights, and therapeutic trade-offs in the management of Major Depressive Disorder (MDD). Derived from exhaustive qualitative formative research, systematic literature synthesis, and iterative consultations with a multi-disciplinary Stakeholder Advisory Group (Xie et al., 2021), the instrument addresses a persistent limitation of traditional psychiatric rating scales by capturing how individuals weigh competing clinical, functional, and economic outcomes. Rather than assessing depressive symptom severity retrospectively, this instrument models prospective health decision-making behavior across distinct therapeutic attributes. The instrument comprises six systematically varied experimental choice tasks. Each choice task presents respondents with three distinct, multi-attribute treatment profiles alongside an overarching forced-choice elicitation query: ‘If you could only choose 1 option for treating depression, which do you most prefer?’ The evaluation profiles are defined by six core attributes established as central to patient decision-making: mode of treatment, time to treatment effect, days of hopefulness per week, effect on workplace and daily productivity, impact on interpersonal relations with others, and monthly out-of-pocket costs. Because the DCE operationalizes microeconomic stated-preference valuation rather than reflective psychometric scaling, standard Classical Test Theory metrics such as Cronbach’s alpha, exploratory factor analysis, and confirmatory factor analysis are fundamentally inapplicable; instead, psychometric validity is established through rigorous attribute-development qualitative validity, choice consistency testing, and structural behavioral modeling via random utility estimation. The instrument provides critical empirical data for personalized medicine, client-treatment matching, health technology assessment, and clinical trial design in neuropsychiatric care.

Keywords

Discrete Choice Experiment, Major Depressive Disorder, Patient Preferences, Treatment Attitudes, Stated-Preference Valuation, Random Utility Theory, Health Economics, Client-Treatment Matching, Shared Decision-Making, Treatment Planning, Mental Health Technology Assessment, Patient-Reported Outcomes

Authors

The Discrete Choice Experiment Instrument was developed through an interdisciplinary collaboration between academic health outcomes researchers and patient-centered health economics organizations:

• Susan dosReis, PhD — Department of Pharmaceutical Health Services Research, School of Pharmacy, University of Maryland Baltimore, Baltimore, Maryland, United States. (ORCID: 0003-2210-6650; Email: [email protected])
• Laura M. Bozzi, PhD — Department of Pharmaceutical Health Services Research, School of Pharmacy, University of Maryland Baltimore, Baltimore, Maryland, United States. (ORCID: 0001-5326-6521)
• Beverly Butler — PAVE Center, Department of Pharmaceutical Health Services Research, School of Pharmacy, University of Maryland Baltimore, Baltimore, Maryland, United States.
• Richard Z. Xie, PhD — Innovation and Value Initiative, Alexandria, Virginia, United States. (ORCID: 0002-2797-3019)
• Richard H. Chapman, PhD — Innovation and Value Initiative, Alexandria, Virginia, United States. (ORCID: 0002-1988-1368)
• Jennifer Bright, MPA — Innovation and Value Initiative, Alexandria, Virginia, United States. (ORCID: 0001-7442-3979)
• Erica Malik, MS — Innovation and Value Initiative, Alexandria, Virginia, United States. (ORCID: 0002-2092-4273)
• Julia F. Slejko, PhD — Department of Pharmaceutical Health Services Research, School of Pharmacy, University of Maryland Baltimore, Baltimore, Maryland, United States. (ORCID: 0001-6900-9220)

Purpose

Major Depressive Disorder is a highly heterogeneous psychiatric condition characterized by complex symptom presentations, variable treatment response rates, and substantial burdens on functional status and quality of life. Despite the clinical availability of diverse pharmacotherapies, psychotherapies, neuromodulation strategies, and integrative behavioral health interventions, non-adherence and treatment discontinuation rates remain exceptionally high, often exceeding 50% within the initial months of treatment initiation. A primary contributor to treatment non-adherence and sub-optimal therapeutic alliance is the disconnect between clinicians’ prescriptive priorities and patients’ multidimensional personal preferences. Traditional clinical assessment instruments, such as the Hamilton Depression Rating Scale (HDRS) or the Patient Health Questionnaire-9 (PHQ-9), are designed to measure symptom frequency and severity; they do not possess the methodological framework to evaluate how patients weigh competing benefits, administrative burdens, functional gains, interpersonal side effects, or financial toxicities.

The purpose of the Discrete Choice Experiment Instrument developed by dosReis et al. (2023) is to systematically elicit and quantify the latent decision architecture governing treatment choice among adults diagnosed with MDD. Rather than relying on simple rating or ranking scales, which are prone to ceiling effects (e.g., respondents rating all positive outcomes as maximally important), a discrete choice methodology forces participants to make realistic trade-offs between desirable and undesirable therapeutic attributes. For instance, a patient may be required to choose between a treatment modality offering rapid onset of symptom relief at high out-of-pocket costs versus a lower-cost option with slower onset or differential effects on relational dynamics.

In clinical practice and clinical decision support, the DCE provides a systematic basis for personalized client-treatment matching and shared decision-making. By identifying which attributes drive individual preference heterogeneity, clinicians can align treatment protocols with patient values, thereby enhancing treatment engagement, psychological buy-in, and long-term adherence. In health services research and pharmaceutical development, the instrument generates objective utility coefficients that parameterize cost-effectiveness analyses, inform value-based health technology assessments, and steer patient-centered drug development toward outcomes that matter most to individuals living with depression.

Psychological Construct

The psychological construct evaluated by this instrument is multidimensional treatment preference utility within the context of major depression. Within behavioral economics and quantitative psychology, treatment preference is conceptualized as a latent decision construct representing an individual’s subjective valuation of the overall utility of a healthcare intervention, decomposed into discrete, constituent attributes. Rather than existing as an immutable personality trait, treatment preference represents a goal-directed cognitive-affective evaluation shaped by lived clinical experience, symptom burden, functional aspirations, and socioeconomic constraints.

The DCE operationalizes this construct across six distinct conceptual dimensions:

• Mode of Treatment (Intervention Delivery Modality): Captures the psychological meaning, stigma, lifestyle compatibility, and perceived burden associated with the operational format of the therapeutic intervention. The levels evaluate preferences across three distinct therapeutic strategies: pharmacotherapy alone (medicine), combined treatment (medicine and psychotherapy), or multimodal comprehensive care (all modalities, incorporating behavioral, lifestyle, and supportive services). This dimension reflects attitudes toward pharmacological versus psychosocial mechanisms of healing.
• Time to Treatment Effect (Latency of Therapeutic Onset): Represents the temporal tolerance of patients enduring acute depressive states. Varying across 4 weeks, 6 weeks, and 9 weeks, this dimension captures temporal discounting, desperation for acute relief, and the psychological vulnerability associated with awaiting symptom remission while potentially enduring treatment-emergent side effects.
• Days of Hopefulness (Positive Affect Induction): Evaluates the emotional and affective efficacy of the intervention. Moving beyond the mere suppression of negative affect or vegetative depressive symptoms, this dimension explicitly quantifies restoration of hedonic capacity, optimism, and positive emotional valence across levels of 2 days/week, 4 days/week, and 6 days/week. It reflects recovery-oriented and humanistic dimensions of mental wellness.
• Effect on Productivity (Functional and Vocational Capacity): Measures the subjective importance of restoring cognitive, executive, and occupational performance. Operationalized as a 40%, 60%, or 90% increase in daily and workplace productivity, this attribute taps into the patient’s drive to re-engage with instrumental roles, regain economic self-sufficiency, and overcome depressive executive dysfunction.
• Relations with Others (Interpersonal and Social Functioning): Assesses the social impact of depressive treatments on interpersonal connection, family dynamics, and romantic relationships. Categorized as strained, improved, or no impact, this attribute acknowledges that pharmacotherapy and behavioral changes do not occur in an interpersonal vacuum; changes in emotional blunting, irritability, or communication patterns directly affect familial and social ecosystems.
• Monthly Out-of-Pocket Costs (Financial Burden / Resource Allocation): Quantifies the economic sacrifice required to sustain care. Operationalized across three realistic pricing tiers ($30, $90, and $270 per month), this dimension serves as the critical financial anchor, permitting researchers to estimate willingness-to-pay (WTP) and calculate marginal rates of substitution across clinical and functional benefits.

Theoretical Framework

The Discrete Choice Experiment Instrument is fundamentally grounded in Random Utility Theory (RUT), originally formulated by psychometrician Louis Leon Thurstone (1927) in his Law of Comparative Judgment and subsequently mathematically formalized within econometrics and mathematical psychology by Nobel laureate Daniel McFadden (1974). Random Utility Theory posits that an individual $n$, when faced with a set of mutually exclusive alternatives $j$ within a choice set $C$, chooses the alternative that yields the maximum perceived utility ($U_{nj}$).

According to RUT, this latent utility cannot be observed directly by the investigator. Instead, total utility is decomposed into two additive, orthogonal components: an explainable, systematic component ($V_{nj}$) and an unobservable, stochastic random error term (ε_nj):

U_nj = V_nj + ε_nj

The systematic component $V_{nj}$ is specified as a linear combination of the observable attributes ($X_{njk}$) of the depression treatment profiles, weighted by marginal utility coefficients (β_k):

V_nj = ∑ β_k X_njk

The stochastic error term ε_nj represents unobserved idiosyncrasies, situational fluctuations, measurement noise, and intra-individual cognitive variations. Depending on the distributional assumptions placed on ε_nj, specific probabilistic choice models are derived. When error terms are assumed to be independent and identically distributed (i.i.d.) following a Type I Extreme Value (Gumbel) distribution, the probability $P_{ni}$ that participant $n$ selects alternative $i$ over alternative $j$ takes the standard Multinomial Logit (MNL) formulation:

P_ni = exp(V_ni) / ∑_{j ∈ C} exp(V_nj)

Beyond McFadden’s economic choice framework, the instrument draws heavily from Lancaster’s Consumer Theory (Lancaster, 1966), which established that individuals do not derive utility from goods or services in their entirety, but rather from the specific constituent characteristics (attributes) those goods possess. In psychiatric care, a patient does not simply choose an “antidepressant regimen”; they choose a bundle of characteristics consisting of latency, interpersonal effects, emotional uplift, functional restoration, administration modality, and economic cost.

Validity

Because discrete choice experiments represent stated-preference choice tasks rooted in econometric decision modeling rather than reflective psychological trait inventories, standard psychometric validation procedures based on Classical Test Theory (such as bivariate Pearson correlations for convergent validity) are replaced by rigorous experimental design validity, qualitative content validation, and internal behavioral consistency checks.

Content and Face Validity

The content validity of the dosReis et al. (2023) DCE was established through an extensive, rigorous formative research process (Xie et al., 2021). The authors conducted comprehensive systematic literature reviews to delineate every patient-reported factor associated with depression care. Subsequently, a continuous Stakeholder Advisory Group (SAG) consisting of patients living with MDD, psychiatric clinicians, clinical psychologists, and health services researchers reviewed, refined, and prioritized the candidate attributes. Cognitive pre-testing interviews were executed with MDD patients to evaluate whether respondents understood the attribute definitions, could cognitively navigate the trade-offs without undue fatigue, and found the attribute levels realistic and representative of their real-world treatment struggles. This exhaustive qualitative pipeline guarantees that the six retained attributes represent the most salient decision criteria driving actual patient behavior.

Construct and Behavioral Validity

Construct validity in discrete choice methodology is demonstrated when the estimated model coefficients conform to theoretical economic and psychological expectations (often termed theoretical or directional validity):

• Price/Cost Sensitivity: The estimated marginal utility for monthly out-of-pocket costs is expected to be strictly negative (β_cost < 0), confirming downward-sloping demand curves where higher costs decrease selection probability, all else being equal.
• Clinical Efficacy Monotonicity: Utility coefficients for days of hopefulness (increasing from 2 to 6 days/week) and productivity improvements (increasing from 40% to 90%) must exhibit positive monotonic progressions (β_{6 days} > β_{4 days} > β_{2 days}).
• Latency Aversion: Greater delays in time to therapeutic onset (9 weeks vs. 4 weeks) should demonstrate negative utility weights, reflecting psychological impatience and acute symptom distress.

While dosReis et al. (2023) published the formative foundation and the finalized DCE design, classical convergent and divergent numerical coefficients against external scales (such as the MADRS or BDI-II) were neither relevant nor reported, as the instrument measures multi-attribute preference utility rather than depressive symptom severity.

Reliability

In classical psychometrics, reliability is typically evaluated via internal consistency metrics such as Cronbach’s alpha or McDonald’s omega. However, within stated-preference choice experiments, calculating Cronbach’s alpha across choice tasks is mathematically invalid. In a DCE, experimental attributes are intentionally made orthogonal or near-orthogonal using fractional factorial experimental design algorithms to eliminate inter-attribute correlation and permit independent parameter estimation. A high internal consistency coefficient would indicate severe multicollinearity in the design matrix, which violates basic econometric experimental design principles.

Instead, reliability in DCE instruments is appraised through choice consistency, stability of preference weights, and test-retest concordance:

• Internal Consistency Checks (Dominated Alternatives): Reliability can be formally tested by including a synthetic “dominated” choice task where one profile is unequivocally superior in every attribute (e.g., faster onset, higher days of hope, higher productivity, improved relations, lower cost, and preferred modality). Participants who fail such rationality checks may exhibit random responding, cognitive overload, or satisficing behaviors.
• Test-Retest Stability: Evaluated by presenting an identical choice scenario twice within the survey administration (separated by distractor tasks) and calculating the proportion of concordant choices (typically quantified using Cohen’s kappa, with target values > 0.70).
• Standard Errors of Utility Estimates: In random utility modeling (e.g., Mixed Logit or Latent Class Models), precision is demonstrated through narrow standard errors and statistically significant parameters ($p < 0.05$) across repeated attribute presentations across the respondent sample.

In the original primary study by dosReis et al. (2023), empirical test-retest reliability statistics were not reported for this initial release; investigators implementing the scale in field research are encouraged to embed duplicate choice tasks to establish sample-specific choice reliability.

Factor Analysis

Traditional Exploratory Factor Analysis (EFA) and Confirmatory Factor Analysis (CFA) are designed for continuous or ordinal reflective indicator variables that share common underlying variance. Because the DCE is an experimental task-based measure governed by Random Utility Theory, it does not possess a factor structure in the psychometric sense. No factor analytic results are available or methodologically appropriate for this instrument.

The statistical counterpart to factor analysis in discrete choice modeling is preference heterogeneity modeling, executed via Latent Class Analysis (LCA) or Random Parameters (Mixed) Logit (RPL/MIXL) modeling:

• Mixed Logit Modeling: Assumes that the preference weights β_k are continuous random variables distributed across the population according to specified probability density functions (typically normal or log-normal):
  β_nk = β_k + σ_k υ_nk
  where β_k is the mean marginal utility, σ_k is the standard deviation capturing preference dispersion, and υ_nk is individual-specific variation. Significant standard deviation estimates confirm that the population holds heterogeneous values toward that specific attribute.
• Latent Class Choice Models: Analytically analogous to mixture factor models, Latent Class Choice Analysis sorts respondents into discrete, unobserved market or clinical segments (classes). Each latent class shares homogeneous preference weights, while preferences vary substantially across classes. In MDD research, latent class modeling typically uncovers subgroups such as “Efficacy-Maximizers,” “Cost-Sensitive Pragmatists,” or “Side-Effect/Relationship-Focused” individuals.

Instrument / Measurement Tool

The operational specifications of the Discrete Choice Experiment Instrument are structured as follows:

• Test Type: Task-based Stated-Preference Discrete Choice Experiment (DCE); Multi-Attribute Conjoint Profile Instrument.
• Administration Format: Self-administered via computer, tablet, or web browser; electronic survey interface.
• Duration / Completion Time: Pre-test cognitive completion time is approximately 15 to 20 minutes.
• Number of Choice Tasks: 6 distinct experimental choice scenario questions.
• Profiles per Choice Task: 3 distinct, competing multi-attribute treatment profiles (e.g., Treatment Option A, Treatment Option B, Treatment Option C) per task.
• Overarching Elicitation Task: ‘If you could only choose 1 option for treating depression, which do you most prefer?’
• Core Evaluated Attributes and Levels:
  • Mode of treatment: 3 levels — (1) medicine; (2) medicine and psychotherapy; (3) all modalities.
  • Time to treatment effect: 3 levels — (1) 4 weeks; (2) 6 weeks; (3) 9 weeks.
  • Days of hopefulness: 3 levels — (1) 2 days/week; (2) 4 days/week; (3) 6 days/week.
  • Effect on productivity: 3 levels — (1) 40% increase; (2) 60% increase; (3) 90% increase.
  • Relations with others: 3 levels — (1) strained; (2) improved; (3) no impact.
  • Monthly out-of-pocket costs: 3 levels — (1) $30; (2) $90; (3) $270.
• Scoring and Statistical Estimation: Responses are binary selections across discrete alternative profiles per choice set. Data are coded into choice matrices and analyzed via Multinomial Logit (MNL), Mixed Logit (MIXL), or Latent Class Regression to derive: (a) part-worth utility coefficients (β), (b) relative attribute importance (RAI) percentages, and (c) marginal willingness-to-pay (WTP = -β_attribute / β_cost).

Permissions & Fee and Test Year

The Discrete Choice Experiment Instrument was developed and published in 2023 by Dr. Susan dosReis and colleagues. It is an open-access, non-commercial scientific research instrument designed to advance patient-centered outcome research and health economics in mental health. There are no licensing fees associated with academic or clinical non-profit use. Researchers and clinicians wishing to implement the experimental design matrix, survey software templates, or full choice vignettes should obtain permissions and design specifications by contacting the corresponding author:

Dr. Susan dosReis, PhD
Department of Pharmaceutical Health Services Research
School of Pharmacy, University of Maryland Baltimore
21201 Baltimore, MD, United States
Email: [email protected]

References

• dosReis, S., Bozzi, L. M., Butler, B., Xie, R. Z., Chapman, R. H., Bright, J., Malik, E., & Slejko, J. F. (2023). Preferences for treatments for major depressive disorder: Formative qualitative research using the patient experience. The Patient: Patient-Centered Outcomes Research, 16(1), 57–66. https://doi.org/10.1007/s40271-022-00596-6
• Lancaster, K. J. (1966). A new approach to consumer theory. Journal of Political Economy, 74(2), 132–157. https://doi.org/10.1086/259131
• McFadden, D. (1974). Conditional logit analysis of qualitative choice behavior. In P. Zarembka (Ed.), Frontiers in Econometrics (pp. 105–142). Academic Press.
• Thurstone, L. L. (1927). A law of comparative judgment. Psychological Review, 34(4), 273–286. https://doi.org/10.1037/h0070288
• Xie, R. Z., Bright, J., dosReis, S., Bozzi, L. M., Butler, B., Chapman, R. H., Malik, E., & Slejko, J. F. (2021). Engaging patients and stakeholders in defining treatment attributes for major depressive disorder: A qualitative study. Value in Health, 24(S1), S105–S106. https://doi.org/10.1016/j.jval.2021.04.526

Items of the Scale