尽管人工智能（AI）和机器人技术能够解决当今的许多医疗挑战，但是这些技术必须克服若干障碍，才能大大改善我们的医疗服务系统。

人工智能是指在复杂的医疗数据分析中使用近似人类认知的算法，目的是确定预防或治疗技术与治疗结果之间的关系。机器人技术主要涉及机器人的设计，开发和操作，将其用于患者监控、评估以及手术辅助等活动。AI辅助的机器人技术有可能完全改变医疗方式，因为机器有可能变得足够强大和聪明，足以替代医师和医务人员。

虽然完全自主的机器人系统仍然很遥远，但是AI和机器人技术已经在临床环境和医师培训中提供了许多优点，能够提高医疗效果。此外，机器人/AI技术的临床应用正在不断增长，越来越多的大型提供商携新兴技术进入了这一领域。这些包括能够进行机器人辅助手术的达芬奇系统；使用图像和计算机控制的机器人技术辐射肿瘤的射波刀（Cyberknife）技术；Babylon Health公司推出的AI辅助的远程诊疗APP；纪念斯隆-凯特琳癌症中心使用Watson Oncology来解释癌症患者的临床资料。

尽管AI和机器人技术潜力无限，但是有几个挑战阻碍了它们的迅速应用。

技术和数据限制

在电子病历（EHR）整合方面，最大的障碍之一是提供完整和全面的患者数据。许多AI技术依赖于输入大量数据来执行高级分析，但是如果不能全面了解患者的病史，AI系统就无法运作。更糟糕的是，这可能会威胁患者的健康。

不幸的是，当前医疗提供商收集和维护患者数据的方法是不够全面的，互操作性的固有复杂性又加剧了这一问题。提供商面临数百种EHR，每种都具有不同的数据架构，他们难以为患者维护单一、全面的病例。无法与其它EHR集成以了解病人的完整病史，对于任何机器学习技术都是一个挑战。

要成功采用AI技术，提供商必须首先克服全面整合患者数据的技术短板。虽然完全的互操作性可能还需要几年才能实现，但是提供商仍然可以在手术或诊断之前人工收集所有相关的患者数据，以此利用AI技术。

经济问题

除了技术限制之外，AI和机器人技术对于中小型提供商来说也是一个重大的经济负担。高昂的采购和实施成本，以及年度维护和合同成本，会对组织造成重大的经济负担。此外，提供商还需要考虑机器人仪器和配件的成本，这些仪器和配件都是根据每个手术采购和更换的。最后，许多技术需要广泛的医师和员工培训，昂贵的工作流程更改，在某些情况下还需要更改EHR。

此外，有研究表明，机器人手术的费用大约是传统手术的10倍，没有证据表明其结果明显优于传统手术。更糟糕的是，在一些案例中，机器人故障导致患者受伤甚至死亡，从而导致法律费用和其它不可预见的费用。

最后，在当下的健康保险系统中，保险公司只愿意支付最低限度的保险金，提供商该如何在这种情况下为患者提供最好的医疗服务呢？像Mayo Clinic，MD Anderson和Partners Healthcare这样的大型医疗机构来有大量资金用于创新和探索AI项目；然而，较小的机构可能无法负担引入AI和机器人技术的高昂成本。

法律，法规和伦理风险

采用AI和机器人技术的提供商面临的最大问题是法律、法规和伦理方面的问题。由于机器人和AI技术的有效性具有敏感性和相对不确定性，所以这方面的风险更大，特别是当医生可以成功实施手术时。

在手术室中使用完全自主的AI机器人需要机器足够聪明，能够计算风险并做出“类似人类”的决策。但是，为了“计算”风险并做出临床决策，我们应该采用什么标准来对这些机器进行道德规划？想象一个未来世界，我们引入了AI/机器人接生系统；当宝宝和母亲的心率都开始下降，两者都命悬一线时，机器人如何选择呢？同样重要的是，病人、家属和法律系统会如何应对这种问题？

此外，获得病人的完整医疗记录，是许多AI技术的必要条件，这也为患者的隐私和安全带来了重大挑战。这引发了对医院IT监管问题的考虑：如何保护患者数据？如何控制数据访问？

对于准备并愿意采用机器人和AI技术的组织来说，深入了解各种风险以及各种技术的局限性，制定强有力的治理模式，实施非常严格的协议和工作流程将使他们能够保护自己免受这些风险和诉讼，成功采用这些技术。

社会抵制

可以说，全面采用这些技术的最大障碍是社会抵制。即使组织愿意承担与临床机器人和AI相关的固有风险，他们该如何确保医生和护士能够舒心地采用机器人治疗？如何确保保险公司愿意为此付款？如何确保患者愿意接受？

医生是否愿意承认机器人能够执行（或者更好地执行）他们培训多年的技能？历史上，机器人通常执行低工资的常规工作；然而，AI正在改变这一格局；创造性的白领工作也将受到影响，医生也不例外。例如，能够读取患者医学图像的机器可以迅速替代放射科医生。或者，能够计算和管理IV药物的机器可以替代麻醉师。

同样，我们也不能忽视患者的反应。随着病人越来越多的作为消费者，患者对他们的医疗方法能够接受何种程度的控制，被谁控制？即使事实证明机器人做的手术具有较高的质量和较低的价格，但是患者可否不选择机器人？此外，保险公司是否愿意支付这些治疗费用？

当然，患者、提供商、医疗机构和政府必须转变思想，才能使机器人真正成为医疗市场的主流。组织必须将机器人和AI作为为患者争取更大利益的补充手段。

在不久的将来，机器人很可能比人类更有能力做人类的任务。然而，为了加快变革的步伐，想要采用这些技术的提供商必须提前思考，了解他们该如何适应未来的医疗世界，以及如何减轻他们面临的潜在挑战。

通过了解技术和数据差距，法律和伦理限制，成本和社会问题，提供商可以适当地做准备，以采用AI和机器人技术。   

《Why robotics and AI still face an uphill battle in healthcare》

https://www.information-management.com/opinion/why-robotics-and-ai-still-face-an-uphill-battle-in-healthcare

Ashley Amaral

Vashist Krishna

August 03 2017, 6:30am EDT

While artificial intelligence and robotics have the potential to solve many of today’s healthcare challenges, several obstacles must be overcome for these technologies to dramatically improve our care delivery system.

Artificial intelligence (AI) refers to the use of algorithms that approximate human cognition in the analysis of complex medical data, with the aim of determining relationships between prevention or treatment techniques and patient outcomes. Robotics primarily deals with the design, development and operation of robots for activities such as patient monitoring and evaluation as well as surgical assistance. Together, AI-powered robotics have the potential to completely change the way healthcare is delivered, as machines may someday become powerful and smart enough to replace physicians and medical staff.

While fully autonomous robotic systems are still far away, there are a number of advantages that AI and robotics already offer in clinical settings and physician training to advance care and improve outcomes. Moreover, the clinical application of robotic/AI technologies is growing, with more big-name providers coming into the scene alongside new and emerging technologies. These include the da Vinci system that enables robot-assisted surgery; the Cyberknife, which uses images and computer-controlled robotics to radiate tumors; Babylon Health’s AI-powered chatbot to assist in remote patient diagnosis; and Memorial Sloan Kettering’s use of Watson Oncology to interpret cancer patients’ clinical data.

Despite great promise from AI and robots, several challenges are still preventing rapid adoption.

Technology and data constraints

Given the current challenges of EHR integration, one of the largest hurdles would be the availability of complete and comprehensive patient data. Many AI technologies rely on the ability to digest large sets to perform advanced analytics, but without a complete understanding of the patient’s medical history, the system fundamentally doesn’t work. Worse, it may put the patient’s health at risk.

Unfortunately, the means by which healthcare providers collect and maintain patient data today is insufficient. Exacerbating this issue are the inherent complications around interoperability. With hundreds of EHRs, each with different data architectures, it’s extremely difficult for providers to keep a single, comprehensive health record for a patient. The inability to integrate with other EHRs to understand the full patient history remains a challenge for any machine learning technology.

Providers that want to successfully adopt AI-related technologies must first overcome technological deficiencies surrounding comprehensive patient data. While full interoperability may be years away, providers can still leverage AI technologies by manually collecting all relevant patient data, performing the appropriate diligence before a procedure or diagnosis.

Financial and economic concerns

In addition to technological limitations, AI and robotics present significant financial concerns for smaller or mid-sized providers. The high acquisition and implementation costs, along with annual maintenance and contract costs, places a significant financial burden on organizations. In addition, providers also need to consider the costs of robotic instruments and accessories, which are procured and replaced on a per-procedure basis. Finally, many of these technologies require extensive physician and staff training, costly workflow changes, and in some cases, changes to the EHR.

Further, there have been studies showing that robotic surgeries can cost roughly 10 times more than traditional surgeries, with no evidence that outcomes are significantly better than traditional methods. Worse, there have been instances where robotic malfunction has resulted in injuries and even death, resulting in the potential for legal fees and other unforeseen costs.

Finally, in today’s health insurance system, how do providers justify the cost of delivering the very best care to patients when insurers today are only reimbursing for good enough care? For larger healthcare institutions such as Mayo Clinic, MD Anderson, and Partners Healthcare, there is plenty of funding to take on innovative and exploratory AI programs from top vendors; however, smaller institutions may struggle to justify the high cost of introducing AI and robotic technologies until they are more proven.

Legal, regulatory and ethical risk

Among the biggest concerns facing providers adopting AI and robotics are the legal, regulatory and ethical concerns. Because of the sensitivity and relative uncertainty around effectiveness of robotic and AI technologies, the risk is greater, particularly when the procedure could have been successfully performed by a physician. Navigating these challenges may require providers to adopt a new approach to governance and legal arbitrage.

Using fully autonomous, AI-powered robots in operating rooms requires machines to be smart enough to calculate risks and make “human-like” decisions. However, what criteria should we ethically program these machines to use in order to “calculate” the risks and make clinical decisions? Consider a future world where we’ve introduced an AI/robotic delivery system; how does the robot choose one life over another when both baby and mother’s heart rate begins to drop and both are entering a critical state? Equally important, how would the patient, families and legal system respond?

Moreover, having access to a patient’s complete health history, a requisite for many AI technologies, also presents some major challenges to patient privacy and security. This raises regulatory concerns for hospital IT—how patient data be securely protected, and how will access to data be controlled?

For organizations that are ready and willing to adopt robotic and AI technologies, gaining a thorough understanding the various risks as well as the limitations of each technology, developing a strong governance model, and implementing very rigid protocols and workflows will enable them to protect themselves against such risks and litigation, demonstrate successes, and help shape the delivery of these technologies to improve adoption.

Social resistance

Arguably the biggest barrier to full-scale adoption is social resistance. Even if organizations are willing to take on the inherent risk associated with clinical robotics and AI, how do organizations ensure that the doctors and nurses are comfortable adopting robotic care; that insurers are willing to pay for it; and that patients are willing to receive it?

Will doctors ever be willing to concede that their training-based skills could be performed, perhaps better, by a robot? Historically robotic-enabled job replacement has typically impacted low-wage, routine jobs; however, AI is changing the game; creative and intelligent white-collar jobs will also be impacted, and doctors are not excluded. For example, a machine that reads patient’s medical images could quickly replace radiologists. Or one that can calculate and administer IV medications could replace an anesthesiologist.

Similarly, we cannot ignore how patients will respond. With the rise of the patient as a consumer, what control will patients have over the method of care they receive, and by whom? Even if the facts demonstrate higher quality care and better overall financial value for a robotic-operated surgery, does the patient have a choice? Also, will payers be willing to reimburse for these treatments?

Certainly, a cultural shift among patients, providers, institutions and governments must take place in order for robotics to truly become mainstream in the healthcare market. Organizations must think of robotics and AI as a means of supplementing care for the greater good of the patient.

In the near future, it’s likely that robots can be even more capable than humans at doing very human-like tasks. However, to accelerate the path to change, providers considering the adoption of these technologies must think ahead to understand how they will fit in the future healthcare world and how to mitigate some of the potential challenges they will face.

By understanding the technology and data gaps, legal and ethical limitations, costs, and social concerns, providers can properly prepare their organization for the adoption and growth of AI and robotic technologies.

This article originally appeared in Health Data Management.

  附件：

《Why robotics and AI still face an uphill battle in healthcare》--原文.pdf

《Why robotics and AI still face an uphill battle in healthcare》--译文.pdf