Presentation (CTI)

OBJECTIVES: Evaluation of digital health interventions (DHIs) often remains fragmented, although HTA decisions require multidimensional evidence. This study applied a preference framework covering user value, interaction quality, system integration, and societal impact. It aimed to identify latent preference classes within these value dimensions and examine how class-specific decision patterns can inform more differentiated DHI assessment.
METHODS: A large-scale preference study was conducted in Germany among the general population (N=5,170) using four value dimension-specific discrete choice experiment modules. Participants were randomly assigned to one module and evaluated hypothetical DHIs described by dimension-specific goal-attainment attributes, out-of-pocket costs, and daily time investment. Latent class models were estimated separately for each module. Model selection was guided by information criteria and interpretability of class-specific utility patterns. Because recent evidence suggests that inconsistent or weakly differentiated response patterns may indicate low engagement, poor data quality, or fraudulent participation, a prespecified response-quality framework with 16 flags was applied.
RESULTS: Five-class models were selected across all modules. Despite different attribute sets, the analyses revealed recurring class profiles. A large class in each module showed limited differentiation across attributes. Further recurring profiles included a highly cost-sensitive class, a time-critical class, and a comprehensively value-oriented class. Smaller dimension-specific classes showed strong focal preferences for clinical benefit, interaction quality, technical system quality, or environmental and societal impact. The response-quality assessment tested whether the least differentiated classes reflect substantive indifference, heterogeneous low-involvement preferences, or systematic response-quality concerns.
CONCLUSIONS: Latent class analysis adds critical value to preference-based evaluation by revealing distinct decision logics that aggregate models may obscure. The repeated emergence of comparable class profiles across four value dimensions suggests that stakeholders differ not only in attribute weights but also in how they process value trade-offs. Incorporating latent preference structures and response-quality diagnostics improve the interpretation, credibility, and policy relevance of preference evidence for DHI evaluation.