Carta Healthcare调查：81%的医疗IT领导者表示临床医生必须控制AI
Carta Healthcare Survey: 81% of Healthcare IT Leaders Say Clinicians Must Control AI
你应该了解的内容
根据一项Carta Healthcare调查，81%的受访CHIME成员IT高管认为临床医生参与AI的设计和部署“非常重要或至关重要”，没有任何受访者认为这不重要。
77%的技术领导者表示，当AI工具与临床医生协作运行时，他们看到了更好的临床和运营结果；没有任何人表示这种协作未能带来帮助。
高管们将AI幻觉认定为无人监督模型的首要风险，其排名高于模型漂移、算法偏见、文档错误、责任风险以及受保护健康信息（PHI）暴露。
一线临床医生的采用取决于及早参与软件设计、与现有电子健康记录（EHR）工作流程的无缝集成、透明的推理过程以及正式治理。
调查结果支持Carta Healthcare的“Hybrid Intelligence”模型。该模型将先进的AI自动化与持有资质的临床数据抽取人员相结合，以提供准确、可审计的临床数据管理。
幻觉缓解、Hybrid Intelligence与工作流程适配
Carta Healthcare发布的一项针对医疗保健信息管理高管学院（CHIME）技术领导者的市场调查，凸显了医疗系统C-suite中的行业共识：人工智能必须作为由临床医生治理的力量倍增器运行，而不是取代人类临床判断的自主系统。
该调查说明了在医疗系统中部署企业级AI的具体运营前提：
临床医生控制要求：
81%的IT高管认为临床医生参与AI工具的设计和部署“非常重要或至关重要”，没有任何受访者将临床医生监督归类为不重要。
协作绩效影响：
77%的领导者表示，与AI独立运行相比，当AI与临床医生协同工作时，临床和运营结果更优，这确立了人机AI协作作为实现可衡量价值的主要框架。
无监督AI的主要风险：
IT领导者认为，在没有人工监督的情况下部署AI，模型幻觉是最大的单一风险，其排名高于算法偏见、模型漂移、法律责任和受保护健康信息（PHI）暴露。
一线采用的驱动因素：
要在临床工作流程中成功采用，必须让一线医疗团队尽早参与工具开发，将软件直接嵌入既有EHR流程，确保模型输出透明，并在执行临床或运营决策前保持人工验证。
“医疗系统IT领导者明确告诉我们，只有当临床医生保持控制时，AI才能赢得信任，”Carta Healthcare CEO Brent Dover表示。“这套模式经得起审计、认证审查，也经得起必须为某项结果进行辩护的关键时刻。由临床医生掌舵、AI作为力量倍增器，是唯一能够实现规模化的模式。”
What You Should Know 81% of surveyed CHIME member IT executives consider clinician involvement in the design and deployment of AI “mostly or vitally important,” with zero respondents rating it unimportant, according to a Carta Healthcare survey 77% of technology leaders report seeing better clinical and operational outcomes when AI tools operate collaboratively alongside clinicians, and none reported that collaboration failed to help.
Executives identify AI hallucination as the primary risk of unsupervised models, ranking it ahead of model drift, algorithmic bias, documentation errors, liability, and protected health information (PHI) exposure.
Frontline clinician adoption relies on early involvement in software design, seamless integration into existing electronic health record (EHR) workflows, transparent reasoning, and formal governance.
The survey findings support Carta Healthcare’s “Hybrid Intelligence” model, which pairs advanced AI automation with credentialed clinical abstractors to deliver accurate, auditable clinical data management.
Hallucination Mitigation, Hybrid Intelligence, and Workflow Fit A market survey of College of Healthcare Information Management Executives (CHIME) technology leaders, released by Carta Healthcare, highlights an industry consensus across health system C-suites: artificial intelligence must operate as a clinician-governed force multiplier rather than an autonomous replacement for human clinical judgment.
The survey illustrates specific operational prerequisites for enterprise AI deployment across health systems:
Clinician Control Mandate:
81% of IT executives rate clinician involvement in the design and deployment of AI tools as mostly or vitally important, with zero respondents classifying clinician oversight as unimportant.
Collaborative Performance Impact:
77% of leaders report superior clinical and operational outcomes when AI operates alongside clinicians rather than on its own, establishing human-AI collaboration as a primary framework for measurable value.
Top Unsupervised AI Risks:
IT leaders identified model hallucinations as the single greatest risk of deploying AI without human oversight, outranking algorithmic bias, model drift, legal liability, and protected health information (PHI) exposure.
Drivers of Frontline Adoption:
Successful adoption across clinical workflows requires involving frontline care teams early in tool development, embedding software directly into established EHR routines, ensuring transparent model outputs, and maintaining human validation before executing clinical or operational decisions.
“Health system IT leaders are telling us plainly that AI earns trust when clinicians stay in control,” said Brent Dover, CEO of Carta Healthcare. “That is what holds up under audit, under accreditation review, and in the moment a result has to be defended. Clinicians at the helm, with AI as the force multiplier, is the only model that scales.” Reader Interactions