Measuring the Preferences of Patients With Type II Diabetes Using Best-worst Scaling and Discrete Choice Experiment: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Measuring the Preferences of Patients With Type II Diabetes Using Best-worst Scaling and Discrete Choice Experiment (NCT02637622), a registered phase not stated study. The primary outcome is Patient Medication Preference as Estimated by a Choice Model.

2. Clinical Question

The broader purpose is methodological: to study ways of designing preference studies, analyzing variation in preferences, and assessing the relevance of stated-preference methods to patients and stakeholders. Type II diabetes is the case study through which the investigators pursued those aims. The registry also lists self-reported difficulty understanding and answering survey questions as a secondary outcome. It does not report results for that outcome.

3. Trial Design

This is a completed survey-based study registered as NCT02637622. Its title describes measuring the preferences of patients with type II diabetes using best-worst scaling and a discrete choice experiment. The sponsor is Johns Hopkins Bloomberg School of Public Health. The registry lists the study start as 2013-07 and primary completion as 2016-10, with enrollment of 1103.

The two listed arms are Best-Worst Scaling (Case 2) and Discrete Choice Experiment. The arm descriptions characterize each as a way of measuring preference. These labels describe the methods used to elicit preferences; they do not describe medication treatment groups. Allocation is not stated. The registry cautions that the study is not a randomized comparison and that the arms are not a like-for-like control comparison. It is therefore important to read the arms as distinct preference-measurement approaches, not as randomized alternatives whose outcomes establish a causal effect of one method over the other.

The primary outcome is a mean preference estimate for medication attribute levels, with a 95% confidence interval, estimated by conditional logit regression. For each arm, the registry describes six medication attributes, each with 3 levels. The estimates within an attribute are constrained to sum to 0; a positive or negative estimate indicates a level valued higher or lower than the average for that attribute. The timeframe is a one-time survey, and the primary population is respondents who completed it.

The registry also lists self-reported difficulty understanding and answering survey questions as a secondary outcome, and reports 3 outcome measures in total. It does not report a sample-size justification, allocation procedure, or protocol details for the survey tasks. Specific recruitment methods and the number of completed responses by arm are not reported in the registry record.

4. Participant Flow and Accounting

The registry reports how participants moved through each stage of the study. All counts below are exactly as posted; missing counts are not inferred from other stages.

Before assignment: 638 respondents (34%) did not respond to the survey and 139 people among those who responded (11%) were excluded based on the screening questions.

Overall Study

Analysis population for the primary outcome, as defined in the registry: Respondents who completed the survey.

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

Population: Respondents who completed the survey.

6. Primary Endpoint

The registry lists the primary outcome as Patient Medication Preference as Estimated by a Choice Model, measured over One-time survey.

The primary endpoint is patient medication preference as estimated by a choice model. It is a continuous outcome expressed as utility estimates for medication attribute levels. The registry states that the estimates come from conditional logit regression and are reported as means with 95% confidence intervals. For each arm, the survey covered six attributes with 3 levels each. Within an attribute, the level estimates sum to 0. A positive estimate means that a level's parameter is higher than the average for that attribute; a negative estimate means it is lower.

These estimates are used to represent how respondents value alternative medication characteristics in the choices posed by a stated-preference survey. In general, a choice model translates patterns of selections among alternatives into parameter estimates associated with their characteristics. Such estimates can describe relative preference within the model and survey context; they are not direct measures of clinical benefit, safety, adherence, or treatment effectiveness.

The registry does not give the names or definitions of the medication attributes and levels, the actual parameter estimates, or the confidence interval bounds. Consequently, the direction and magnitude of respondents' preferences for any particular level cannot be interpreted from the registry record. The one-time survey timeframe also means the endpoint describes stated preferences at the survey assessment, not change over time.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The primary outcome uses a conditional logit regression to estimate preferences for levels of medication attributes. The registry identifies the model and the structure of the estimates, but it does not give the survey choice tasks, the exact model specification, or the coefficient results. The method explanation below distinguishes the information recorded for this study from general statistical interpretation.

What does a choice model estimate?

In general, a stated-preference choice model represents choices as arising from the characteristics of the alternatives presented to respondents. The model estimates how the listed characteristics relate to the probability of selecting an alternative, under the model's assumptions. Here, the registry says that conditional logit regression produced a preference estimate for each medication attribute level. It does not report how choice tasks were constructed, how many choices each respondent made, or what additional terms, if any, appeared in the model.

In general, model coefficients are expressed on a statistical utility scale. The scale is a model-based representation of relative preference, not a natural unit that can be directly equated with clinical outcomes or a respondent's personal rating. A coefficient can be interpreted only in relation to the model, attributes, levels, and coding used. Because the specific attributes and level descriptions are not reported in the registry record, the substantive meaning of any particular coefficient cannot be assessed from the record.

How should the level estimates be read?

The registry states that there are six medication attributes and 3 levels per attribute in each arm, and that level estimates within an attribute sum to 0. This constraint sets the attribute's levels in relation to one another. Under the registry's description, an estimate above zero indicates a parameter higher than that attribute's average, while one below zero indicates a parameter lower than its average. A value near zero would be near the attribute average, but the actual values are not reported.

That within-attribute comparison does not mean that estimates from different attributes are directly comparable in importance. Nor does a positive estimate mean that the associated attribute level has a beneficial clinical effect. It describes a preference parameter in the fitted choice model. In general, coefficient magnitudes depend on the scale and specification of the model as well as on how the alternatives are framed.

What does the confidence interval add?

The primary outcome is described as a mean with a 95% confidence interval. In general, a confidence interval gives a range of values compatible with the data and the interval's statistical procedure, subject to the assumptions used. Its location conveys the estimated direction and its width conveys precision. The confidence level is a property of the method across repeated applications; it is not a statement that a particular fixed parameter has a 95% probability of falling within an already calculated interval.

The registry does not report confidence interval bounds, standard errors, or other numerical uncertainty summaries. As a result, it is not possible to evaluate precision, assess whether intervals overlap across arms, or determine whether any estimate is close to zero. Even with such numbers, a formal comparison between arms would require attention to the study's allocation and analysis design; the registry explicitly warns that the arms are not a like-for-like randomized comparison.

Why can over-dispersion matter?

In general, over-dispersion occurs when observed variability exceeds what a statistical model assumes. If it is present and not accounted for, standard errors can be too small and confidence intervals can be too narrow, making estimates appear more precise than warranted. In repeated-choice data, responses from the same person may be related, and model assumptions about the response process matter for uncertainty estimation. This is general statistical context, not a claim that over-dispersion was found in this study. The registry does not report diagnostic checks, respondent-level correlation adjustments, or other variance procedures.

What can be said about sample size?

The registry lists enrollment of 1103, but does not report the completed-response count or analysis count. Enrollment alone is not the effective sample size for an endpoint when some enrolled people may not complete a survey or may have unusable responses. The record does not report attrition or the number analyzed in either arm.

Any sample-size figures based on fractions of an observed effect are planning illustrations, not observed power and not the study's original sample-size justification. The stated planning assumptions use a two-sided significance level of 0.05, equal allocation, and a normal approximation. Those assumptions do not establish that allocation in the actual study was equal: the registry says allocation was not stated. The registry does not report a protocol-based sample-size calculation or its justification. Such planning scenarios should be read only as conditional illustrations, not as proof that the study had a particular probability of detecting an effect.

Overall, the statistical description is enough to identify the endpoint type, model family, and centering interpretation, but not enough to reproduce the analysis or evaluate model adequacy. The choice-task design, model specification, missing-response handling, standard-error procedure, diagnostics, and numerical estimates are not reported in the registry record.

10. Multiplicity, Missing Data and Analysis Populations

The primary outcome involves a set of preference estimates across multiple medication attributes and levels, and the registry also lists a secondary outcome about difficulty understanding and answering the survey questions. When many estimates are examined, the chance of observing apparently notable results by chance can increase. In general, multiplicity can be addressed through a prespecified hierarchy, adjusted inference, or an interpretation focused on estimation and uncertainty rather than isolated significance tests. The registry does not report a multiplicity strategy, a hierarchy among attribute-level estimates, or a plan for handling multiple comparisons.

The record identifies respondents who completed the survey as the primary outcome population, but it does not state how many respondents completed it or how many contributed to each arm's model. It does not describe whether partially completed questionnaires were retained, how unanswered items or choice tasks were handled, or whether exclusions were applied. Missing-data assumptions and methods are not reported in the registry record. Without the counts and handling rules, readers cannot assess whether the analysis population differs from the enrolled population or whether missing responses might affect the preference estimates.

The registry lists 3 outcome measures and no analyses. It describes the primary outcome and names a secondary outcome, but does not report analysis populations, model specifications beyond conditional logit regression, or numerical results. It also does not report whether the analysis followed a prespecified protocol or statistical analysis plan. These absences limit assessment of selective reporting and the relationship between planned and reported analyses. The number of outcome measures alone does not indicate which analyses were conducted or how multiplicity was handled.

11. What the Result Does, and Does Not, Mean

The study addresses how to measure stated preferences among people with type II diabetes, not whether one medication or treatment improves health. Its primary endpoint is a model-based estimate of preference for medication attribute levels. This kind of endpoint can help characterize what respondents value in the alternatives they consider, but it is not a clinical outcome and does not establish effectiveness, safety, adherence, or actual behavior in routine care.

The registry provides a definition for interpreting positive and negative preference parameters relative to the average for each attribute. However, the estimated values and their confidence intervals are not reported. Accordingly, no specific preference direction, magnitude, ranking, or degree of precision can be attributed to the respondents on the basis of the registry record. The listed enrollment does not fill that gap because the number of completed surveys and analyzed respondents is not stated.

The two arms use different preference-elicitation methods. Because allocation is not stated and the registry cautions that the arms are not a randomized, like-for-like control comparison, the design does not support attributing any between-arm contrast solely to the method. The record also does not describe whether characteristics of respondents or survey procedures differed by arm.

The study's broader methodological aims are relevant to patient-centered outcome research: it sought to advance engagement practices and examine design and analysis approaches for preference studies. Those aims describe the study's purpose, not a demonstrated result. The registry does not report outcomes that would show which method performed better, whether respondents found one method easier, or whether a specific engagement practice changed research quality. The listed secondary outcome on survey difficulty is not accompanied by results.

Thus, the record documents a completed preference-measurement study and describes its intended statistical endpoint, but it does not provide the estimates needed to characterize the preferences observed. Interpretation should remain at the level of study design and outcome definition rather than extending to claims about medication choices or patient care.

12. Why This Trial Matters Statistically

Preference data raise statistical questions that are distinct from those in a conventional treatment comparison. The outcome is not a direct clinical event, but a model-derived representation of choices among alternatives. Researchers must therefore consider how attributes and levels are defined, how tasks are presented, and what assumptions connect observed choices to preference estimates. The registry identifies two methods intended to elicit such preferences and a conditional logit model for estimating level parameters.

The outcome's centering rule is also important. Estimates sum to 0 within each attribute, so each level is interpreted relative to the attribute average rather than as an absolute score. This makes clear why model estimates should not be treated as clinical benefits or compared across attributes without understanding their scaling and specification. Reporting the actual estimates and confidence intervals would be necessary to judge both the pattern and its precision; those results are not reported in the registry record.

Comparisons between preference methods require careful design. The registry states that allocation was not stated and that the arms are not a like-for-like control comparison. That caveat matters because apparent differences may reflect group composition or other design features, not only the elicitation approach. A statistical contrast is most informative when the allocation and analysis structure support the intended interpretation.

The study is also a useful reminder that enrollment does not by itself establish the analysis sample size. Survey completion, response quality, and missing choices can affect the information available for estimating preferences. The registry specifies completed respondents as the primary population but does not report their count or missing-data procedures. Together, these issues make transparent reporting of design, model assumptions, uncertainty, and analysis populations central to interpreting stated-preference research.

13. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

The registry record leaves important details of the study design and analysis unspecified. It does not state how participants were allocated to Best-Worst Scaling (Case 2) or the Discrete Choice Experiment, and it cautions that the arms are not a randomized, like-for-like control comparison. This limits the interpretation of any contrast between the methods and prevents a straightforward causal reading.

Although enrollment is listed as 1103, the number of survey completers and the number analyzed in each arm are not reported. The record does not describe recruitment, survey completion, exclusions, or the handling of partially completed responses. It also does not report the names and descriptions of the medication attributes and levels, the choice-task design, or respondent characteristics relevant to interpreting the estimates.

The endpoint description identifies conditional logit regression and a sum-to-zero constraint within each attribute, but further model details are not reported. The registry does not specify the model formula, variance estimation, diagnostics, or treatment of dependence among responses from the same respondent. No numerical primary estimates or confidence interval bounds are reported, so the direction, magnitude, and precision of preferences cannot be evaluated. Results for the secondary survey-difficulty outcome are also not reported.

The record does not report a sample-size justification, a multiplicity plan, missing-data assumptions, or a statistical analysis plan. These limitations concern what is documented in the registry and should not be taken as proof that the investigators did not address these issues elsewhere. They do mean that the registry record alone cannot establish how those decisions were made or assess their implications.

14. Related Tutorials

15. Related Statistical Calculators

16. Sources

• ClinicalTrials.gov: Measuring the Preferences of Patients With Type II Diabetes Using Best-worst Scaling and Discrete Choice Experiment — NCT02637622.
• Registry record last updated 2024-11-29; status last verified 2024-10.
• Results first posted 2024-11-29.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

17. Record Summary

NCT02637622 is a completed study sponsored by Johns Hopkins Bloomberg School of Public Health, with enrollment of 1103. The registry lists a start of 2013-07 and primary completion of 2016-10. It concerns type II diabetes and evaluates two preference-measurement approaches: Best-Worst Scaling (Case 2) and a Discrete Choice Experiment.

The primary outcome is mean patient medication preference estimated using conditional logit regression, with a 95% confidence interval. The outcome description covers six medication attributes, each with 3 levels per arm, and defines estimates relative to the average within each attribute. A secondary outcome concerns self-reported difficulty understanding and answering survey questions.

The registry does not report the primary estimates, confidence interval bounds, completed survey counts, or secondary outcome results. Allocation is not stated, and the record cautions that the arms are not a randomized, like-for-like control comparison. It lists 3 outcome measures and no analyses.