Key points are not available for this paper at this time.
Human-centered artificial intelligence (AI) is increasingly recognized as important to advancing quality, safety, and trust in healthcare analytics, yet most validation frameworks continue to prioritize technical metrics over interpretability and stakeholder experience. Quality Function Deployment (QFD) offers a systematic method for translating customer requirements into technical design, but its application within healthcare AI remains limited. This study adapts QFD methodology to systematically align user feedback with prioritized technical requirements in healthcare AI systems. The analysis encompassed 14,938 patient reviews from 53 hospitals, from which 1,279 negative reviews were extracted for thematic analysis using large language model-driven coding (Cohen's Kappa = 0.81) and empirical factor structure, mapping multidimensional patient needs to technical specifications through a House of Quality matrix. Sensitivity analysis revealed that Granular Categorization demonstrated the highest improvement potential, achieving 21.9% advantage over LLM Coding Accuracy. This framework offers a potential approach for integrating technical validation and human-centered quality assessment, and may provide guidance for developing trustworthy, interpretable, and equitable digital medicine. While validation is limited to Malaysian private hospitals, the methodology offers a potentially scalable approach for healthcare AI development that warrants further validation. Future directions include real-world deployment across diverse populations and dynamic regulatory contexts.
Muda et al. (Mon,) studied this question.