On November 30, 2022, OpenAI released ChatGPT, the first chatbot and virtual assistant powered by large language models (LLMs). In just five days, ChatGPT attracted over 1 million users and reached 200 million monthly active users worldwide within fifteen months. This sudden surge of interest in artificial intelligence (AI) has expanded its potential from a niche concept to a mainstream obsession.

AI and machine learning were already making strides in medicine and healthcare, but with the advent of prescriptive and generative AI, new opportunities emerged to redefine how healthcare professionals diagnose, treat, and monitor patients [1]. AI has the potential to enhance diagnostic precision and provide personalized care by bridging the gap between digitalized medical data, clinical decisions, and optimized healthcare delivery.

The term “Augmented Intelligence” may be more fitting than “Artificial Intelligence,” as it emphasizes AI’s role as a collaborator that enhances human intelligence rather than replacing it. As large language models become more advanced, it is important to address the technical, ethical, social, and practical challenges they present.

AI’s role is evolving from a mere tool to an assistant and potentially to a colleague. Just as human colleagues are expected to adhere to strict ethical and professional guidelines, AI systems must also be designed with similar standards in mind to support healthcare professionals and maintain integrity and trust in clinical settings.

Establishing clear guidelines and regulations for augmented intelligence will be vital for integrating AI into healthcare teams [2]. This ensures that AI enhances care delivery in a safe, reliable, and trustworthy manner without compromising patient safety and autonomy and that it benefits all communities, including those in low-resource settings and minority groups.

This insight is derived from the collaborative perspectives of 22 experts from a 3-day international AI roundtable at the ISICEM conference in Brussels in March 2024. It sheds light on the current situation and challenges regarding AI in acute medicine and urges stakeholders to work together to leverage AI-enabled care and expand acute medicine’s reach.

Several papers have demonstrated that predictive models could recognize patterns or identify early warning signs of critical conditions, potentially leading to more timely interventions and improved patient outcomes [3].

AI systems can combine data from diverse sources such as imaging, electronic health records, and wearable devices to offer a holistic view of a patient’s condition. They can also help extract usable information from the current data overload that everyone in healthcare is exposed to. AI systems can then help make clinical decisions that align with real-world complexities and patient-specific needs, providing healthcare professionals with a comprehensive understanding that improves care delivery.

AI systems could also streamline note-taking, documentation, and correspondence between healthcare providers and patients [4]. Additionally, AI could help in research. Current research in acute settings has several limitations. Populations of acutely ill patients are highly heterogeneous, and diseases are exceptionally dynamic. Unsurprisingly, randomized controlled trials have often failed to show positive results. AI could significantly improve trial design and execution by offering new ways to address these challenges. AI could identify precise patient phenotypes for accurate inclusion criteria, ensuring that trials enroll the most suitable participants [5]. It can also assist in real-time monitoring of trial participants, providing early signals of efficacy or adverse effects. Additionally, predictive models can help adapt trial designs dynamically, allowing investigators to adjust interventions based on emerging data.

Creating digital twins of patients and healthcare systems will enable researchers and clinicians to simulate potential outcomes, optimize resource allocation, and effectively guide care delivery. By developing accurate, data-driven digital twins, healthcare professionals can conduct controlled experiments and identify the best strategies to deliver truly personalized precision medicine. This digital “dry run” could reduce the risks and costs of testing novel treatments in vulnerable patient populations. Before jumping to these implementations, however, further research must prove how AI models can truly discriminate association from causality or how they can help investigators reduce uncertainty in their models, make trial design more efficient, and, ultimately, improve clinical outcomes [6].

Research has shown that AI can predict clinical trajectories in research settings but moving towards actionable AI or AI-enabled care, where insights directly guide clinical decisions in real-time, remains a significant challenge.

Establishing standardized data frameworks and promoting their adoption is vital to facilitating the seamless exchange of healthcare data across systems. Data fragmentation obstructs the development of robust AI models and hinders their smooth integration into clinical workflows.

In ICUs, for instance, patients often present a wide range of conditions, making it challenging to classify their medical phenotypes without detailed patient data. Real-time data gathering and analysis are crucial to effectively identify individual patient phenotypes, a practice that is not commonly implemented. The establishment of collaborative real-time data networks is essential, as no single ICU can independently gather all necessary information.

AI-based clinical decision support systems often lack situational awareness due to limited training in replicating real-world clinical decision-making processes. This gap can hinder AI systems from understanding clinical context and providing valuable input for clinical decision-making.

Concerns about privacy, data security, and transparency can be alarming for patients, families, healthcare organizations, and governments.

Clinician acceptance is hindered by the ‘black box’ problem, where models are not easily interpretable. Deep learning models may require more transparency, leading to skepticism among clinicians who need help understanding how a system arrives at its conclusions.”

Overcoming these challenges necessitates a comprehensive framework prioritizing the following core elements:

1. 1.

   Social Contract for AI: Develop a social contract with input from clinicians, data experts, policymakers, patients, and families to ensure that AI tools respect patient rights and autonomy while upholding ethical standards.

2. 2.

   Human-Centric AI Development: Empower rather than replace healthcare professionals. Systems must be designed to enhance clinical decision-making while maintaining the clinician-patient relationship [7]. Do this most inclusively, driving improvements for everyone.

3. 3.

   Data Standardization and Infrastructure: Establish unified standards and infrastructure to enable seamless data sharing and foster collaboration. For instance, OMOP, FHIR, and i2b2 can play pivotal roles in creating robust data structures that support AI integration.

4. 4.

   Federated Real-Time Networks: Create real-time clinical research networks to enhance collaboration and enable data aggregation to study rare events. This will also improve phenotyping and allow clinicians to tailor treatments based on precise patient subtypes, moving toward a personalized and actionable AI model.

5. 5.

   Education and Training: Provide healthcare professionals with the training to utilize AI tools effectively, understand their strengths and limitations, understand and accept uncertainty, and interpret probabilistic information. At the same time, build a “learn while doing” culture with AI-augmented human systems continuously improving their models as they analyze new data and adapt to changing clinical landscapes.

6. 6.

   Collaborative Research and Development: Encourage partnerships between the public and private sectors to drive research that addresses critical needs in acute and critical care. This should be inclusive, with a specific focus on not leaving behind minorities and low-resource settings.

Integrating AI in medicine can transform healthcare delivery, but achieving this vision requires a concerted effort—Fig. 1. Stakeholders must embrace a unified approach to AI integration, advocating for robust data infrastructures, ethical frameworks, and collaborative networks that can harness the full potential of AI. By focusing on data standardization, real-time ICU networks, education, and working on a new “social contract for AI between all stakeholders,” we can move toward a future where AI-enabled care brings acute medicine where and when it is needed, ultimately improving patient outcomes, and enhancing the clinician-patient relationship.

DALL-E interpretation of the viewpoint. DALLE Open AI 2024

Full size image

Maurizio Cecconi, Massimiliano Greco, Benjamin Shickel, Derek Angus, Heatherlee Bailey, Elena Bignami, Thierry Calandra, Leo Anthony Celi, Sharon Einav, Paul Elbers, Ali Ercole, Hernando Gomez, Michelle NG Gong, Matthieu Komorowski, Vincent Liu, Soojin Park Aarti Sarwal, Christopher Seymour, Fernando Zampieri, Fabio Silvio Taccone, Jean-Louis Vincent, Azra Bihorac.

No datasets were generated or analysed during the current study.

1. The AI doctor. will see you… eventually. The Economist; 2024.

2. Pinsky MR, Bedoya A, Bihorac A, Celi L, Churpek M, Economou-Zavlanos NJ, Elbers P, Saria S, Liu V, Lyons PG, et al. Use of artificial intelligence in critical care: opportunities and obstacles. Crit Care. 2024;28(1):113.

   Article PubMed PubMed Central Google Scholar

3. Komorowski M, Celi LA, Badawi O, Gordon AC, Faisal AA. The Artificial Intelligence Clinician learns optimal treatment strategies for sepsis in intensive care. Nat Med. 2018;24(11):1716–20.

   Article CAS PubMed Google Scholar

4. Komorowski M, del Pilar Arias López M, Chang AC. How could ChatGPT impact my practice as an intensivist? An overview of potential applications, risks and limitations. Intensive Care Med. 2023;49(7):844–7.

   Article PubMed Google Scholar

5. Angus DC. Randomized clinical trials of Artificial Intelligence. JAMA. 2020;323(11):1043–5.

   Article PubMed Google Scholar

6. Messeri L, Crockett MJ. Artificial intelligence and illusions of understanding in scientific research. Nature. 2024;627(8002):49–58.

   Article CAS PubMed Google Scholar

7. Cecconi M. Reflections of an intensivist in 2050: three decades of clinical practice, research, and human connection. Crit Care. 2023;27(1):391.

   Article PubMed PubMed Central Google Scholar

Download references

The authors did not receive support from any organization for the submitted work.

Authors and Affiliations

1. Humanitas University, Milan, Italy

   Maurizio Cecconi & Massimiliano Greco

2. IRCCS Humanitas Research Hospital, Milan, Italy

   Maurizio Cecconi & Massimiliano Greco

3. University of Florida, Gainesville, USA

   Benjamin Shickel & Azra Bihorac

4. Erasme University Hospital, HUB, Université Libre de Bruxelles, Brussels, Belgium

   Jean-Louis Vincent

1. Maurizio CecconiView author publications

   You can also search for this author in PubMed Google Scholar

2. Massimiliano GrecoView author publications

   You can also search for this author in PubMed Google Scholar

3. Benjamin ShickelView author publications

   You can also search for this author in PubMed Google Scholar

4. Jean-Louis VincentView author publications

   You can also search for this author in PubMed Google Scholar

5. Azra BihoracView author publications

   You can also search for this author in PubMed Google Scholar

Contributions

Every author participated to the roundtable discussion in Brussels, and contributed to the writing of the manuscript.

Corresponding author

Correspondence to Maurizio Cecconi.

Ethics approval and consent to participate

Not applicable.

Competing interests

The authors declare no competing interests.

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

Cite this article

Cecconi, M., Greco, M., Shickel, B. et al. Artificial intelligence in acute medicine: a call to action. Crit Care 28, 258 (2024). https://doi.org/10.1186/s13054-024-05034-7

Download citation

• Received: 10 July 2024

• Accepted: 12 July 2024

• Published: 29 July 2024

• DOI: https://doi.org/10.1186/s13054-024-05034-7

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

2022 年 11 月 30 日，OpenAI 发布了 ChatGPT，这是第一个由大型语言模型支持的聊天机器人和虚拟助手（LLMs ）。短短五天内，ChatGPT 就吸引了超过 100 万用户，并在 15 个月内达到了 2 亿全球月活跃用户。人们对人工智能 (AI) 的兴趣突然高涨，其潜力已从一个小众概念扩展为主流的痴迷。

人工智能和机器学习已经在医学和医疗保健领域取得了长足的进步，但随着规范性和生成式人工智能的出现，出现了重新定义医疗保健专业人员如何诊断、治疗和监测患者的新机会[1]。人工智能有潜力通过弥合数字化医疗数据、临床决策和优化医疗服务之间的差距来提高诊断精度并提供个性化护理。

“增强智能”一词可能比“人工智能”更合适，因为它强调人工智能作为协作者的作用，增强而不是取代人类智能。随着大型语言模型变得更加先进，解决它们带来的技术、道德、社会和实践挑战非常重要。

人工智能的角色正在从单纯的工具演变为助手，甚至可能成为同事。正如人类同事应遵守严格的道德和专业准则一样，人工智能系统的设计也必须考虑到类似的标准，以支持医疗保健专业人员并保持临床环境中的完整性和信任。

建立明确的增强智能指南和法规对于将人工智能整合到医疗团队中至关重要[2]。这确保人工智能以安全、可靠和值得信赖的方式增强护理服务，而不损害患者的安全和自主权，并惠及所有社区，包括资源匮乏地区和少数群体。

这一见解源自 2024 年 3 月在布鲁塞尔举行的 ISICEM 会议上为期 3 天的国际人工智能圆桌会议上 22 名专家的合作观点。它揭示了急性医学中人工智能的现状和挑战，并敦促利益相关者共同努力，利用人工智能支持的护理并扩大急性医学的覆盖范围。

几篇论文表明，预测模型可以识别模式或识别危急情况的早期预警信号，从而有可能导致更及时的干预措施并改善患者的治疗结果 [3]。

人工智能系统可以结合成像、电子健康记录和可穿戴设备等不同来源的数据，以提供患者病情的整体视图。它们还可以帮助从医疗保健中每个人都接触到的当前数据过载中提取可用信息。然后，人工智能系统可以帮助做出符合现实世界复杂性和患者特定需求的临床决策，为医疗保健专业人员提供全面的了解，从而改善护理服务。

人工智能系统还可以简化医疗保健提供者和患者之间的笔记、文档和通信[4]。此外，人工智能可以帮助研究。目前在急性环境下的研究存在一些局限性。急症患者群体具有高度异质性，并且疾病异常动态。毫不奇怪，随机对照试验常常未能显示出积极的结果。人工智能可以通过提供应对这些挑战的新方法来显着改善试验设计和执行。人工智能可以识别精确的患者表型，以实现准确的纳入标准，确保试验招募最合适的参与者 [5]。它还可以帮助实时监测试验参与者，提供疗效或不良反应的早期信号。此外，预测模型可以帮助动态调整试验设计，使研究人员能够根据新出现的数据调整干预措施。

创建患者和医疗保健系统的数字双胞胎将使研究人员和临床医生能够模拟潜在结果、优化资源分配并有效指导护理服务。通过开发准确的、数据驱动的数字双胞胎，医疗保健专业人员可以进行受控实验并确定提供真正个性化精准医疗的最佳策略。这种数字化“试运行”可以降低在弱势患者群体中测试新疗法的风险和成本。然而，在进行这些实施之前，进一步的研究必须证明人工智能模型如何能够真正区分关联与因果关系，或者它们如何帮助研究人员减少模型中的不确定性，使试验设计更加高效，并最终改善临床结果[6]。

研究表明，人工智能可以预测研究环境中的临床轨迹，但转向可操作的人工智能或人工智能支持的护理（其中见解直接实时指导临床决策）仍然是一个重大挑战。

建立标准化数据框架并促进其采用对于促进跨系统医疗数据的无缝交换至关重要。数据碎片阻碍了强大的人工智能模型的开发，并阻碍其顺利融入临床工作流程。

例如，在 ICU 中，患者经常表现出各种各样的病症，因此在没有详细患者数据的情况下很难对他们的医疗表型进行分类。实时数据收集和分析对于有效识别个体患者表型至关重要，但这种做法并不常见。建立协作实时数据网络至关重要，因为没有哪个 ICU 能够独立收集所有必要的信息。

由于复制现实临床决策过程的培训有限，基于人工智能的临床决策支持系统通常缺乏态势感知。这种差距可能会阻碍人工智能系统理解临床背景并为临床决策提供有价值的输入。

对隐私、数据安全和透明度的担忧可能会让患者、家庭、医疗机构和政府感到震惊。

“黑匣子”问题阻碍了临床医生的接受，因为模型不容易解释。深度学习模型可能需要更高的透明度，这会导致需要帮助了解系统如何得出结论的临床医生的怀疑。”

克服这些挑战需要一个优先考虑以下核心要素的综合框架：

1. 1.

   
   人工智能社会契约：根据临床医生、数据专家、政策制定者、患者和家庭的意见制定社会契约，以确保人工智能工具尊重患者权利和自主权，同时维护道德标准。

2. 2.

   
   以人为本的人工智能开发：增强而不是取代医疗保健专业人员。系统的设计必须能够增强临床决策，同时维持临床医患关系[7]。以最包容的方式做到这一点，推动每个人的进步。

3. 3.

   
   数据标准化和基础设施：建立统一的标准和基础设施，以实现无缝数据共享并促进协作。例如，OMOP、FHIR 和 i2b2 在创建支持 AI 集成的强大数据结构方面可以发挥关键作用。

4. 4.

   
   联合实时网络：创建实时临床研究网络以加强协作并实现数据聚合以研究罕见事件。这也将改善表型分析，并允许临床医生根据精确的患者亚型定制治疗方案，迈向个性化且可操作的人工智能模型。

5. 5.

   
   教育和培训：为医疗保健专业人员提供有效利用人工智能工具、了解其优势和局限性、理解和接受不确定性以及解释概率信息的培训。与此同时，利用人工智能增强的人类系统建立“边做边学”的文化，在分析新数据和适应不断变化的临床环境时不断改进其模型。

6. 6.

   
   合作研究与开发：鼓励公共和私营部门之间的伙伴关系，推动满足急症和重症护理关键需求的研究。这应该是包容性的，特别注重不留下少数群体和资源匮乏的环境。

将人工智能融入医学可以改变医疗保健服务，但实现这一愿景需要共同努力——图 1。 1. 利益相关者必须采用统一的人工智能集成方法，倡导强大的数据基础设施、道德框架和协作网络，以充分利用人工智能的潜力。通过专注于数据标准化、实时 ICU 网络、教育以及制定新的“所有利益相关者之间的人工智能社会契约”，我们可以迈向这样一个未来：人工智能支持的护理在需要的地方和时间提供急症医学，最终改善患者的治疗效果，并加强临床医患关系。

DALL-E观点的解释。从开放到 2024 年

 全尺寸图像

毛里齐奥·切科尼、马西米利亚诺·格雷科、本杰明·希克尔、德里克·安格斯、希瑟利·贝利、埃琳娜·比格纳米、蒂埃里·卡兰德拉、利奥·安东尼·塞利、莎朗·艾纳夫、保罗·埃尔伯斯、阿里·埃尔科莱、埃尔南多·戈麦斯、米歇尔·吴·宫、马蒂厄·科莫罗夫斯基、文森特·刘、Soojin Park Aarti萨瓦尔、克里斯托弗·西摩、费尔南多·赞皮耶里、法比奥·西尔维奥·塔科内、让-路易斯·文森特、阿兹拉·比霍拉克。

当前研究期间没有生成或分析数据集。

1. 
   人工智能医生。最终会见到你。经济学人； 2024 年。

2. 
   Pinsky MR、Bedoya A、Bihorac A、Celi L、Churpek M、Economou-Zavlanos NJ、Elbers P、Saria S、Liu V、Lyons PG 等。人工智能在重症监护中的应用：机遇和障碍。危重护理。 2024；28(1):113。

   
   文章 PubMed PubMed Central Google Scholar

3. 
   科莫罗夫斯基 M、塞利 LA、巴达维 O、戈登 AC、费萨尔 AA。人工智能临床医生学习重症监护中脓毒症的最佳治疗策略。纳特医学。 2018；24(11):1716–20。

   
   文章 CAS PubMed 谷歌学术

4. 
   Komorowski M、del Pilar Arias López M、Chang AC。 ChatGPT 对我作为重症监护医师的实践有何影响？潜在应用、风险和限制的概述。重症监护医学。 2023；49(7)：844–7。

   
   文章 PubMed 谷歌学术

5. 
   安格斯特区。人工智能的随机临床试验。贾马。 2020；323(11):1043–5。

   
   文章 PubMed 谷歌学术

6. 
   梅塞里·L，克罗克特·MJ。人工智能和科学研究中的理解错觉。自然。 2024；627(8002)：49–58。

   
   文章 CAS PubMed 谷歌学术

7. 
   Cecconi M. 2050 年重症监护医师的反思：三十年的临床实践、研究和人际关系。危重护理。 2023；27(1):391。

   
   文章 PubMed PubMed Central Google Scholar

 下载参考资料

作者提交的工作没有得到任何组织的支持。

 作者和单位

1. 
   人文大学，米兰，意大利

   
   毛里齐奥·切科尼 & 马西米利亚诺·格雷科

2. 
   IRCCS Humanitas 研究医院，意大利米兰

   
   毛里齐奥·切科尼 & 马西米利亚诺·格雷科

3. 
   佛罗里达大学，盖恩斯维尔，美国

   
   本杰明·希克尔和阿兹拉·比霍拉克

4. 
   伊拉斯姆大学医院，中心，布鲁塞尔自由大学，布鲁塞尔，比利时

    让·路易·文森特

1. 
   Maurizio Cecconi查看作者出版物

   
   您也可以在PubMed中搜索该作者 谷歌学术

2. 
   Massimiliano Greco查看作者出版物

   
   您也可以在PubMed中搜索该作者 谷歌学术

3. 
   本杰明·希克尔查看作者出版物

   
   您也可以在PubMed中搜索该作者 谷歌学术

4. 
   让-路易斯·文森特查看作者出版物

   
   您也可以在PubMed中搜索该作者 谷歌学术

5. 
   Azra Bihorac查看作者出版物

   
   您也可以在PubMed中搜索该作者 谷歌学术

 贡献

每位作者都参加了布鲁塞尔的圆桌讨论，并为手稿的写作做出了贡献。

 通讯作者

通讯作者：毛里齐奥·切科尼。

道德批准并同意参与

 不适用。

 利益竞争

作者声明没有竞争利益。

 出版商备注

施普林格·自然对于已出版的地图和机构隶属关系中的管辖权主张保持中立。

开放获取本文根据知识共享署名-非商业性-禁止衍生品 4.0 国际许可证获得许可，该许可证允许以任何媒介或格式进行任何非商业使用、共享、分发和复制，只要您给予原作者适当的署名（ s) 和来源，提供知识共享许可证的链接，并指出您是否修改了许可材料。根据本许可，您无权共享源自本文或其部分内容的改编材料。本文中的图像或其他第三方材料包含在文章的知识共享许可中，除非材料的出处中另有说明。如果文章的知识共享许可中未包含材料，并且您的预期用途不受法律法规允许或超出了允许的用途，则您需要直接获得版权所有者的许可。要查看此许可证的副本，请访问 http://creativecommons.org/licenses/by-nc-nd/4.0/。

 转载和许可

 引用这篇文章

 下载引文

• 
  收件日期： 2024 年 7 月 10 日

• 
  接受日期： 2024 年 7 月 12 日

• 
  发布日期： 2024 年 7 月 29 日

 分享这篇文章

您与之分享以下链接的任何人都可以阅读此内容：

抱歉，本文目前没有可共享的链接。

由 Springer Nature SharedIt 内容共享计划提供