Comparing Discrete Choice Experiment with Swing Weighting to Estimate Attribute Relative Importance: A Case Study in Lung Cancer Patient Preferences

Abstract

Introduction: Discrete choice experiments (DCE) are commonly used to elicit patient preferences and to determine the relative importance of attributes but can be complex and costly to administer. Simpler methods that measure relative importance exist, such as swing weighting with direct rating (SW-DR), but there is little empirical evidence comparing the two. This study aimed to directly compare attribute relative importance rankings and weights elicited using a DCE and SW-DR. Methods: A total of 307 patients with non–small-cell lung cancer in Italy and Belgium completed an online survey assessing preferences for cancer treatment using DCE and SW-DR. The relative importance of the attributes was determined using a random parameter logit model for the DCE and rank order centroid method (ROC) for SW-DR. Differences in relative importance ranking and weights between the methods were assessed using Cohen’s weighted kappa and Dirichlet regression. Feedback on ease of understanding and answering the 2 tasks was also collected. Results: Most respondents (>65%) found both tasks (very) easy to understand and answer. The same attribute, survival, was ranked most important irrespective of the methods applied. The overall ranking of the attributes on an aggregate level differed significantly between DCE and SW-ROC (P < 0.01). Greater differences in attribute weights between attributes were reported in DCE compared with SW-DR (P < 0.01). Agreement between the individual-level attribute ranking across methods was moderate (weighted Kappa 0.53–0.55). Conclusion: Significant differences in attribute importance between DCE and SW-DR were found. Respondents reported both methods being relatively easy to understand and answer. Further studies confirming these findings are warranted. Such studies will help to provide accurate guidance for methods selection when studying relative attribute importance across a wide array of preference-relevant decisions. Both DCEs and SW tasks can be used to determine attribute relative importance rankings and weights; however, little evidence exists empirically comparing these methods in terms of outcomes or respondent usability. Most respondents found the DCE and SW tasks very easy or easy to understand and answer. A direct comparison of DCE and SW found significant differences in attribute importance rankings and weights as well as a greater spread in the DCE-derived attribute relative importance weights.