EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Trust-by-Design Analytics for Consumer Health Wearables: A Cost-Sensitive Framework for Deploying Generative AI in Continuous Cardiac Monitoring

Haoran Zhou, Yufeng Tao, Miaomiao Zhang

Journal of Business and Data Analytics · 30.12.2023 · 1 (1) · 1-22

DOI: 10.63646/bekl6302 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Business and Data Analytics, online erschienen am 30.12.2023. Verfasst von Haoran Zhou und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Continuous cardiac monitoring through photoplethysmography (PPG) on consumer wearables has migrated from a fitness convenience to a consequential clinical signal, particularly for screening atrial fibrillation (AF) at population scale. Yet the operating environment of such devices — motion artefacts, perfusion variability, and ambulatory noise — degrades the very deep-learning classifiers that vendors and payers rely on to convert raw waveforms into business and clinical decisions. This article develops a Trust-by-Design Analytics framework for deploying generative artificial intelligence (GenAI) within consumer health wearables, with a specific focus on the AF-screening pipeline. The framework couples a generative adversarial denoiser, a calibrated probabilistic classifier, and a decision-theoretic uncertainty-quantification (DTUQ) layer into a single, cost-sensitive deployment unit. Using a controlled numerical study calibrated to published wearable AF cohorts, we show that DTUQ-gated decisions recover area-under-curve performance from 0.69 under realistic noise to 0.85 after generative reconstruction, while reducing expected misclassification cost by approximately 41% relative to a cost-sensitive noisy baseline at a clinically realistic false-negative-to-false-positive cost ratio of 5:1. We articulate three managerial implications. First, calibration error — not raw accuracy — is the binding constraint on safe scaling of consumer wearables into reimbursable digital-health pathways. Second, the marginal value of a generative denoiser is monotone in the underlying cost asymmetry, which means investment cases for GenAI in this domain must be built on cost-sensitive, rather than accuracy-based, evaluation. Third, an explicit abstain-and-refer option closes the gap between regulatory expectations for explainable medical AI and the operational reality of always-on consumer sensing. The framework is intended as a deployment template for analytics teams working at the boundary of regulated clinical decision support and at-scale consumer engagement.

Zitieren

APA

Zhou, H., Tao, Y., & Zhang, M. (2023). Trust-by-Design Analytics for Consumer Health Wearables: A Cost-Sensitive Framework for Deploying Generative AI in Continuous Cardiac Monitoring. Journal of Business and Data Analytics, 1(1), 1-22. https://doi.org/10.63646/bekl6302

Vancouver

Zhou H, Tao Y, Zhang M. Trust-by-Design Analytics for Consumer Health Wearables: A Cost-Sensitive Framework for Deploying Generative AI in Continuous Cardiac Monitoring. Journal of Business and Data Analytics. 2023;1(1):1-22. doi:10.63646/bekl6302

BibTeX
@article{Zhou2023,
  author = {Zhou, Haoran and Tao, Yufeng and Zhang, Miaomiao},
  title = {Trust-by-Design Analytics for Consumer Health Wearables: A Cost-Sensitive Framework for Deploying Generative AI in Continuous Cardiac Monitoring},
  journal = {Journal of Business and Data Analytics},
  year = {2023},
  volume = {1},
  number = {1},
  pages = {1--22},
  doi = {10.63646/bekl6302}
}
Kostenlos starten →