BACKGROUND: Increasingly, researchers are interested in perspectives from preference data that can be gained through latent-class analysis (LCA). Typical model-selection approaches for choosing the ideal class size include the Akaike (A1C) and Bayesian information criteria (BIC). Results presented typically include a single model with a fixed number of classes, usually 2 or 3. However, there are practical benefits for clinical and commercial strategies in exploring and reporting on a wider range of sample segmentation.
OBJECTIVES: LCA analysis explored preference information gained or lost while incrementally moving from 3 to 5 classes.
METHODS: In an online discrete-choice experiment survey, preferences were elicited for an interception treatment for type 1 diabetes (T1D) among 1,501 parents in the United States. Respondents were told to assume their child would develop T1D in the future and were offered choices between two hypothetical treatments and a monitoring-only option. Data segmentation was explored with 3-, 4-, and 5-class latent class models.
RESULTS: The 3-class model revealed a segment that prioritized avoiding risks (24% class membership probability), a segment that valued efficacy (57%), and a segment with disordered preferences (20%). Adding a fourth class revealed two types of efficacy segments: those who valued short-term efficacy (32%) and another segment that valued long-term efficacy most (23%). A fifth class revealed, for the first time, a small segment of respondents (8%) who preferred monitoring only over any interception treatment.
CONCLUSIONS: Understanding how preferences vary across segments of the sample is important, and research would benefit from presenting LCA results with varying numbers of classes. Comparing results across models aids in understanding how preference results evolved over 3, 4, and 5 classes with important implications for clinical and commercial decision making.
