CCM 和 KLAS Research 报告：63%的卫生系统缺乏先进的 AI 战略框架
CCM and KLAS Research Report: 63% of Health Systems Lack Advanced AI Strategy Frameworks
您应了解的内容A report by the Center for Connected Medicine (CCM) at UPMC and KLAS Research reveals that while 93% of surveyed health system leaders have deployed third-party AI, strategy and testing infrastructure lag far behind.由 UPMC 的 Center for Connected Medicine (CCM) 和 KLAS Research 发布的一份报告显示，尽管接受调查的卫生系统领导者中有93%已经部署了第三方 AI，但战略和测试基础设施远远落后。
临床文档和环境式记录以52%的部署率位居首位，其次是收入周期和编码（36%）、医学影像（32%）以及嵌入 EHR 的临床决策支持（32%）。
尽管92%的组织会开展部署前测试，但不到一半（44%）的组织拥有专用数据平台或沙盒环境，用于验证模型的准确性、安全性和漂移。
63%的卫生系统 AI 战略仍处于“发展中”或临时制定状态，仅有4%被描述为“先进”。
主要运营痛点包括依赖手动电子表格（17名受访者）、团队之间的数据定义不一致（14名受访者），以及严重的人才或资源限制（11名受访者）。
CCM 和 KLAS Research 揭示医疗 IT 的治理与测试缺口企业医疗 IT、临床转型和医疗保健分析领域面临一项结构性运营挑战：卫生系统对 AI 的采购速度显著超过了安全验证算法所需的内部治理、数据架构和测试流程的建设速度。
尽管高管董事会面临采用临床和行政 AI 的巨大压力，但在没有专用测试环境的情况下部署算法，会使卫生系统面临算法漂移、未经审查的偏见、安全漏洞以及未经验证的投资回报率等风险。
为量化这一运营缺口，UPMC 的 Center for Connected Medicine (CCM) 和 KLAS Research 发布了一份题为“Validation and Trust: How Health Systems Are Testing and Governing Analytics and AI Solutions”的综合报告。
该研究对27个卫生系统的 C-suite、高级临床信息学人员和 IT 高管进行了调查，表明医疗保健 AI 的主要瓶颈已经从工具采购转向运营验证和基础设施。
该研究概述了当前卫生系统部署中的关键技术和结构性现实：
广泛部署与基础设施缺口：
93%的卫生系统已经部署第三方 AI，但只有44%的卫生系统拥有专用数据环境（例如云数据湖仓、生产环境克隆或真实世界数据平台），用于在临床集成前测试模型。
主要用例分布：
环境式临床文档记录的采用率最高（52%），其次是收入周期管理和编码（36%）、诊断影像（32%）以及嵌入 EHR 的临床决策支持（32%）。
数据质量摩擦：
依赖手动变通方法/电子表格（17名受访者）、各部门之间指标不一致（14名受访者）以及非结构化临床数据（10名受访者）是影响模型可靠性的最大障碍。
计算机模拟沙盒验证：
先进组织正在部署专用平台，例如 UPMC 的真实世界数据引擎 Ahavi，以针对去标识化患者群体评估第三方算法，同时不影响实际医疗服务。
“医疗保健行业已经从讨论 AI 的潜力迅速转向在整个企业中积极部署解决方案，”UPMC 首席医疗信息官 Rob Bart 医生表示。“卫生系统目前正专注于构建必要的治理结构、测试能力和组织战略，以确保 AI 能够带来有意义且可衡量的价值。”
What You Should Know A report by the Center for Connected Medicine (CCM) at UPMC and KLAS Research reveals that while 93% of surveyed health system leaders have deployed third-party AI, strategy and testing infrastructure lag far behind.
Clinical documentation and ambient scribing lead deployment at 52%, followed by revenue cycle and coding (36%), medical imaging (32%), and EHR-embedded clinical decision support (32%).
While 92% of organizations conduct pre-deployment testing, less than half (44%) possess a dedicated data platform or sandbox environment to validate model accuracy, safety, and drift.
63% of health system AI strategies remain “developing” or ad hoc, with only 4% described as “advanced”.
Top operational pain points include reliance on manual spreadsheets (17 respondents), inconsistent data definitions across teams (14 respondents), and severe talent or resource constraints (11 respondents).
CCM and KLAS Research Expose Health IT’s Governance and Testing Deficit
The enterprise health IT, clinical transformation, and healthcare analytics sectors face a structural operational challenge: health system AI procurement is significantly outpacing the internal governance, data architecture, and testing pipelines required to validate algorithms safely.
While executive boards face immense pressure to adopt clinical and administrative AI, deploying algorithms without dedicated testing environments exposes health systems to algorithmic drift, unvetted bias, security vulnerabilities, and unvalidated ROI.
To quantify this operational gap, the
Center for Connected Medicine (CCM) at UPMC and KLAS Research released a comprehensive report titled Validation and Trust: How Health Systems Are Testing and Governing Analytics and AI Solutions Surveying C-suite, clinical informatics, and IT executives across 27 health systems, the research demonstrates that the primary bottleneck in healthcare AI has shifted from tool procurement to operational validation and infrastructure.
The research outlines key technical and structural realities across current health system deployments:
Widespread Deployment vs. Infrastructure Deficit:
93% of health systems have deployed third-party AI, yet only 44% maintain a dedicated data environment (such as cloud lakehouses, production clones, or real-world data platforms) to test models before clinical integration.
Primary Use Case Distribution:
Ambient clinical documentation leads adoption (52%), followed by revenue cycle management and coding (36%), diagnostic imaging (32%), and EHR-embedded clinical decision support (32%).
Data Quality Friction:
Reliance on manual workarounds/spreadsheets (17 respondents), inconsistent metrics across departments (14 respondents), and unstructured clinical data (10 respondents) represent the greatest obstacles to model reliability.
In Silico Sandbox Validation:
Advanced organizations are deploying dedicated platforms—such as UPMC’s real-world data engine, Ahavi —to evaluate third-party algorithms against de-identified patient populations without disrupting live care.
“The health care industry has moved remarkably quickly from discussing the potential of AI to actively deploying solutions across the enterprise,” stated Dr. Rob Bart, Chief Medical Information Officer at UPMC. “Health systems are now focused on building the governance structures, testing capabilities and organizational strategies necessary to ensure AI delivers meaningful and measurable value.” Reader Interactions