Reimbursement agencies are increasingly using patient preference data to evaluate health technologies. Discrete choice experiments (DCE) are commonly used to elicit patient preferences, but they require large sample sizes to obtain meaningful results. For this reason, it is often not possible to use DCE to elicit patient preferences in rare diseases. This study assessed a swing weighting method for eliciting preferences from a small sample: patients with immunoglobulin A nephropathy (IgAN) in the United States (US) and China.

Attributes and levels were selected based on a review of clinical studies and qualitative research on patients. Computer-assisted, interview-based swing weighting exercises were piloted in a focus group with five participants each from the US and China. Preferences were then elicited in interviews with twenty-five patients in the US and fifteen patients in China. Consistency tests were used to assess internal validity. Qualitative data were collected on the reasons for patients’ preferences.

Preference consistency: The weights for one attribute were elicited twice. The difference between initial and consistency test weights was not statistically significant (p < 0.1), although this may partly reflect the small sample sizes. Trade-offs: Qualitative data were used to demonstrate the validity of interpreting participants’ ratings as trade-offs. Using the partial value function for end-stage renal disease as an example, qualitative data demonstrated that patients were able to provide face-valid reasons for different shaped, non-linear preference functions. Robustness of treatment evaluation: Three hypothetical treatment profiles (using the attribute swings) were constructed. Preferences for these treatment profiles were robust to variations in patients’ preferences; all patients preferred one specific profile. This finding was not sensitive to changes in weights.

This study supports the feasibility of collecting valid and robust preference data from small groups of patients using swing weighting. Further work could be done to test the performance of swing weighting in larger sample sizes.