Symptoms and causes of obstacles in AI-based telemonitoring: strategies for system development through a CRISP-DM...

Chronic diseases, particularly heart failure (HF), impose a substantial clinical and socioeconomic burden worldwide. AI-based telemonitoring holds significant promise for improving chronic disease management through continuous monitoring and early detection of deterioration. However, translation into routine clinical practice remains limited due to challenges regarding data quality, transparency, regulation, and integration into established clinical workflows. Furthermore, the underlying end-to-end development processes are often inadequately documented and exhibit a positive publication bias, thereby limiting the opportunity to learn from prior efforts.

This study employs a holistic, process-oriented approach to examining obstacles across the full development lifecycle of AI-based telemonitoring systems. Existing Cross Industry Standard Process for Data Mining (CRISP-DM) variants are synthesized into a structuring framework tailored to telemonitoring contexts and used to guide a qualitative case study of the KardioInterakt project, a German research initiative developing an AI-based telemonitoring system for HF management.

By systematically reconstructing decisions, constraints, and mitigation strategies and synthesizing them with prior work, we offer practice-oriented recommendations with potential transferability for the development of AI-based telemonitoring systems. These pertain to scoping and feasibility assessment, stakeholder involvement, data strategy, regulatory compliance, user-centered design, and transparent publication.

Our findings offer actionable guidance for the development of viable, trustworthy, robust, and regulatory-compliant AI-based telemonitoring solutions. The process-oriented recommendations address prevailing challenges and demonstrate how these can be managed across the development lifecycle, thereby promoting transparency and supporting successful integration into routine clinical practice.