customer health scoring

本文提出了一种服务健康工程方法，通过结合遥测、工作流完成情况等手段来检测分布式系统中的静默故障和异步工作停滞问题。

Ensuring sustained safety, performance stability, and clinical value of deployed medical AI systems remains a critical governance challenge. Method: This study proposes the first post-deployment monitoring framework integrating three dimensions—system integrity, dynamic performance stability, and real-world clinical impact—grounded in three synergistic principles: quantifiable metrics, clear accountability assignment, and closed-loop response orchestration. The framework is compatible with both traditional and generative AI systems and technically integrates runtime error detection, input distribution shift analysis, clinical workflow-embedded feedback collection, and multi-tiered dashboards. Contribution/Results: Validated at Stanford Health Care, the framework reduced mean AI system fault response time by 62%, achieved an 89% early detection rate for critical performance degradation, and produced a reusable, standardized monitoring plan template.

This study addresses the challenge of healthcare professionals (HCPs) inefficiently interpreting and leveraging patient-generated health data (PGHD)—such as wearable sensor outputs and symptom logs—to support cardiac risk reduction. We designed and implemented INSIGHT, an interactive dashboard co-developed with clinicians that integrates multimodal PGHD and introduces a large language model (LLM)-powered natural language interface. This interface enables dynamic querying, automated summarization, and generation of personalized clinical insights, overcoming semantic and cognitive limitations of conventional visualization tools. Validated in real-world clinical settings, INSIGHT significantly improved HCPs’ efficiency and depth of PGHD interpretation, enhanced the precision and personalization of physical activity interventions, and established the first LLM-augmented, HCP-centered paradigm for AI-enhanced clinical data cognition.

To address insufficient transparency, accountability gaps, and uncontrolled bias in AI dataset development, this paper proposes the first structured scoring card framework aligned with the full data lifecycle. The framework operationalizes the System Card concept into an auditable assessment system spanning five dimensions: data dictionary, collection methodology, composition, motivation, and preprocessing. It integrates a standardized intake form, multidimensional weighted scoring criteria, cross-dataset consistency evaluation, and a metadata governance model. Uniquely combining technical specifications with ethical review, it ensures end-to-end transparency and traceable accountability. Empirical evaluation across four heterogeneous datasets demonstrates that the framework accurately identifies dataset deficiencies, generates actionable, dataset-specific improvement recommendations, and significantly enhances documentation completeness and trustworthiness. This work establishes a novel data governance paradigm for building fair and interpretable AI systems.

This paper addresses modeling challenges associated with Complex Performance Indicators (CPIs)—multidimensional composite metrics such as customer satisfaction and sustainability indices—in their design, interpretation, and dynamic evolution. Following the PRISMA-ScR framework, we conduct a systematic scoping review employing thematic coding, modeling-feature extraction, and cross-framework comparative analysis. We propose, for the first time, a comprehensive taxonomy of CPI modeling features and quantitatively assess the coverage of these features by mainstream Model-Driven Engineering (MDE) frameworks. The study bridges a critical gap in the literature: the absence of a systematic MDE-oriented review for CPI modeling. It identifies essential modeling capabilities required to ensure CPI understandability, maintainability, and evolvability. As a key outcome, we deliver a reusable MDE-CPI modeling guideline, providing both theoretical foundations and methodological support for academic research and industrial practice.

This work proposes a large language model (LLM)-based framework for adaptive health dashboards that overcomes the limitations of traditional rule-based systems or data-intensive approaches, which struggle to deliver real-time personalization from sparse and heterogeneous user interactions. By employing structured, hierarchical prompt engineering, the framework integrates three distinct behavioral signals—explicit preference feedback, spatial drag-and-drop rearrangements, and hover dwell duration—while decoupling temporal context modeling, behavior interpretation, preference constraints, and user profile synthesis. This design enables immediate fusion of multimodal inputs and supports high-level layout decisions. Evaluated in a scenario comprising 14 health metrics and 7 visualization components, the approach demonstrates significant improvements in both interface personalization and interpretability.

This study addresses a critical gap in geriatric care research, which has predominantly emphasized operational efficiency while lacking robust connections to clinical health outcomes. Through a systematic review of 30 interdisciplinary studies at the intersection of industrial engineering and operations research applied to elderly care, the work categorizes existing literature into three thematic domains: home-based medical care, polypharmacy management, and chronotherapeutic clinical scheduling. It proposes a novel conceptual framework that explicitly links operational optimization with clinical outcomes, advocating a paradigm shift from isolated task-level improvements toward integrated, multi-layer decision-making. By incorporating human-centered design, systems analysis, and emerging technologies such as digital twins and large language models, the study highlights the field’s current overemphasis on workforce scheduling at the expense of health impact, thereby laying a theoretical and technical foundation for intelligent care decision systems spanning hospital and community settings.

This study addresses the lack of standardization in emergency department (ED) workflows, where patient boarding is frequently attributed to demand based on subjective assumptions rather than empirical evidence. Leveraging end-to-end process mining techniques on ED event logs, this work employs inductive process discovery, token-based conformance checking, and data preprocessing to quantify flow performance and assess structural consistency. The analysis identifies 884 control-flow variants, revealing that a model exhibiting perfect fitness yet low precision indicates the absence of normative pathways. Notably, it uncovers a previously undocumented clinical priority inversion wherein urgent patients experience longer lengths of stay than critical patients. Based on these findings, two process improvement strategies are proposed, providing an evidence-based foundation for optimizing ED management.

This study addresses the rare-event validation bias in electronic health record (EHR) algorithm evaluation caused by prohibitive manual review costs. We propose a sampling-based statistical validation framework that exhaustively verifies positive samples while randomly sampling negative ones. The method incorporates Horvitz-Thompson estimators with finite population variance corrections to estimate performance metrics, and integrates Neyman optimal allocation with adaptive sampling techniques to support confidence interval planning and risk-guided sample allocation. Experimental results demonstrate that this framework substantially reduces validation bias, achieves near-nominal coverage rates, and proves effective in a breast cancer recurrence prediction task.

This work addresses the suboptimality of conventional approaches that rely solely on prediction accuracy to set intervention thresholds and select algorithms in real-world settings characterized by constrained service capacity and stochastic individual compliance. To overcome this limitation, we propose a decision framework for AI-assisted intervention deployment that jointly optimizes the choice of predictive algorithm and intervention threshold, balancing resource utilization with coverage of high-value individuals. We formalize the shortcomings of standard strategies, introduce Operational AUC (OpAUC)—a novel evaluation metric aligned with operational objectives—and rigorously characterize the theoretical properties of optimal thresholds. Empirical evaluation on early sepsis detection data demonstrates that our approach significantly improves intervention effectiveness under limited resources.