matlab .fda文件_A. FDA现有的上市后角色

matlab .fda文件

There is an elephant in the room. Tension in the air that we need to discuss. FDA is trying to convince artificial intelligence developers in healthcare to let the agency have much broader power and authority over marketed products than the agency presently does in exchange for, FDA suggests, faster premarket review times. My advice to industry? Don’t do it. Don’t agree to regulatory changes that would give FDA expanded post-market powers. It won’t be worth it.

Ť这里是房间里的大象。 我们需要讨论的紧张气氛。 FDA试图说服医疗保健领域的人工智能开发人员，使该机构对市售产品的权力和权威比该机构目前要广得多，以换取FDA建议加快上市前审查时间。 我对行业的建议？ 不要这样 不要同意将使FDA扩大上市后权力的法规变更。 这是不值得的。

To be clear, I’m not giving this advice simply because FDA has failed to quantify any premarket advantage, whether in terms of reduced evidence required for submission to the agency or speedier reviews. Instead, this is my advice simply because the cost of giving FDA expanded postmarket authority would be too much to pay. While I agree expanded vigilance is indeed necessary when commercializing medical AI that can adapt to new data, I urge the medical AI industry to be proactive in accepting responsibility for their own postmarket monitoring and continuous improvement, in lieu of increased FDA oversight.

明确地说，我并不是仅仅因为FDA无法量化任何上市前优势而给出建议，无论是减少提交给该机构所需的证据还是加快审查的速度。 相反，这只是我的建议，因为给予FDA扩大后上市授权的成本实在太大了。 虽然我同意在商业化能够适应新数据的医疗AI时确实确实需要提高警惕，但我还是敦促医疗AI业界积极主动地为自己的上市后监测和持续改进承担责任，以代替对FDA的监督。

This is a big topic, so I’m dividing it into three different parts that will be published separately. What follows in this post is part one, where I provide the background on FDA’s plans to ask Congress for more legal authority to regulate the marketing of medical AI. Among other things, for context, I will describe FDA’s existing authority, and I will address the standard by which we should evaluate whether expanded regulatory oversight is best for the patient. The second part will focus on what medical AI developers should do now to take responsibility for marketing medical AI. In other words, I’ll talk about best practices that it’s my hope industry will adopt. Then third, in my final post I will make the case why Congress should not give FDA expanded oversight of medical AI commercialization.

这是一个很大的主题，因此我将其分为三个不同的部分，将分别发布。 这篇文章的后续内容是第一部分，在此我提供了FDA要求国会要求更多法律权力来规范医疗AI营销的计划的背景。 在上下文中，除其他事项外，我将描述FDA的现有权限，并讨论标准，通过该标准我们应该评估扩大的监管监督是否最适合患者。 第二部分将重点介绍医疗AI开发人员现在应该做什么，以负责营销医疗AI。 换句话说，我将讨论我希望业界采用的最佳实践。 第三，在我的最后一篇文章中，我将说明为什么国会不应该给予FDA对医疗AI商业化的扩大监督。

A. FDA现有的上市后角色 (A. FDA’s Existing Postmarket Role)

For context, it’s important to understand that FDA already has a substantial role overseeing the commercialization of medical devices, but it is a legally constrained one. Here is a very high-level summary of some of FDA’s more significant, relevant legal authorities over medical products in the marketplace:

就上下文而言，重要的是要了解FDA已经在监督医疗器械的商业化方面发挥了重要作用，但是它在法律上受到限制。 以下是有关市场上医疗产品的FDA的一些更重要的相关法律授权的高级摘要：

· If a company wants to change a marketed medical device in a material way, the company typically needs to get FDA clearance or approval for the change before making it. There are exceptions for small changes in either the product or its labeling. There are also intermediate reports that can be filed unless significant updates, including for Class II devices, “letters to file” and those that can be aggregated and reported later for Class III devices in annual reports changes.

·如果公司希望以实质性方式更改市场上出售的医疗器械，则该公司通常需要获得FDA许可或批准才能进行更改。 产品或其标签的细微变化也有例外。 除非有重大更新，否则还有一些中间报告可以提交，包括对II类设备的“文件归档”，以及可以在以后的年度报告更改中对III类设备进行汇总和报告的更新。

· Manufacturers are required to keep track of complaints they receive, and the investigations they perform under their quality system to assess the significance of those complaints. Manufacturers maintain all of these records, and the FDA has the power to make site inspections that allow them access to all of the quality information including these records.

·制造商必须跟踪收到的投诉以及他们在质量体系下进行的调查，以评估这些投诉的重要性。 制造商保留所有这些记录，而FDA有权进行现场检查，以使他们能够访问包括这些记录在内的所有质量信息。

· Manufacturers have an ongoing obligation to report to the FDA problems with their products aftermarket introduction. These problems are either instances where the product has hurt someone in a material way, or where the product has malfunctioned, and even though it didn’t hurt anyone, if the malfunction were to recur, it’s possible that the product would hurt someone. FDA can punish companies for failing to report these events.

·制造商有持续的义务向其产品上市后向FDA报告问题。 这些问题可能是产品严重伤害某人或该产品出现故障的情况，即使它并未伤害任何人，但如果故障再次发生，则该产品很可能会伤害某人。 FDA可以惩罚未能报告这些事件的公司。

· FDA has the power to order a recall, although it’s more typical for FDA merely to cajole the manufacturer into initiating its own recall.

·FDA有权下令召回，尽管对于FDA而言，更典型的是仅仅诱使制造商发起自己的召回。

· FDA can require, if it goes through a legal process, a particular manufacturer to undertake postmarket surveillance of its product.

·如果经过合法程序，FDA可以要求特定制造商对其产品进行上市后监督。

· FDA can initiate any number of legal proceedings such as a seizure action to pursue a manufacturer it has reason to believe is producing products that don’t comply with the law. FDA can also bring punishing legal actions such as enormous civil and criminal fines after the fact for bad behavior.

·FDA可以发起许多法律程序，例如扣押行动，以起诉有理由相信其生产的产品不符合法律的制造商。 对于不良行为，FDA也可以提起惩罚性法律行动，例如对民事和刑事处以巨额罚款。

· FDA has several other legal authorities such as banning certain products that it considers unsafe, and administratively — without a court — detaining products over which it has concerns.

·FDA还有其他几个法律机构，例如禁止某些其认为不安全的产品，并在行政上(无需法院批准)扣留其担心的产品。

But that, it would seem, in the agency’s opinion is not enough power for FDA when it comes to the commercialization of AI.

但是，就机构而言，在AI的商业化方面，FDA似乎还没有足够的权力。

B.房间里的大象 (B. The Elephant in the Room)

Every time FDA raises the topic of its new vision for regulating AI, in addition to discussing changes to the premarket review process and the quality system requirements, FDA throws in at the end of the discussion its desire to expand its postmarket authorities. I’ll share two examples.

每次FDA提出其监管AI的新愿景主题时，除了讨论对上市前审查流程和质量体系要求的变更之外，FDA在讨论结束时都提出了扩大其上市后授权的愿望。 我将分享两个示例。

1.软件预认证程序 (1. Precertification Program for Software)

FDA appears to have modeled its proposed precertification program after the TSA’s Pre-Check program for US airline travelers, suggesting that it will offer a quicker route to market for manufacturers that are willing to undergo rigorous precertification to assure that they have a robust culture of quality that is likely to produce safe and effective medical software. It’s a big topic, and frankly there are substantial pluses and minuses to what FDA has proposed. The program is still under development, and is in the pilot stage.

FDA似乎是在TSA针对美国航空旅客的Pre-Check计划之后，对拟议的预认证计划进行了建模，表明FDA将为愿意进行严格的预认证以确保他们拥有健全的质量文化的制造商提供更快的上市途径。可能会产生安全有效的医疗软件。 这是一个很大的话题，坦率地说，FDA的建议有很多利弊。 该程序仍在开发中，并且处于试验阶段。

But specifically for purposes of this article, FDA has made it clear that it wants expanded postmarket authority in connection with that proposed program. In the last public iteration of the program, released in January 2019,[1] FDA observes:

但特别是出于本文的目的，FDA已明确表示希望与该拟议计划相关的扩展上市后授权。 在2019年1月发布的该程序的最后一次公开迭代中， [1] FDA观察到：

We believe that excellent organizations not only focus on quality while developing [software as a medical device] products, but they also grow and evolve based on lessons learned from real-world usage of their products after they launch. While specific [Real-World Performance, or RWP] data elements and analytic methodologies may differ across organizations and product categories, excellent organizations consistently collect and analyze post-launch data from diverse sources to inform their operations and decision making, from quality control to product development for new market segments.

我们相信，优秀的组织不仅会在开发[作为医疗设备的软件]产品时关注质量，而且会在产品发布后从实际使用中汲取的经验教训中成长和发展。 尽管不同组织和产品类别的特定[现实世界绩效或RWP]数据元素和分析方法可能有所不同，但优秀的组织始终如一地收集和分析来自各种来源的发布后数据，以告知其从质量控制到产品的运营和决策。开发新的细分市场。

On that we agree. But after observing that all companies in the program should collect such data, FDA then says such company should:

对此我们表示同意。 但是在观察到程序中的所有公司都应收集此类数据之后，FDA随后说该公司应：

allow FDA access to analytics on data elements relevant to organizational excellence and product-level safety and effectiveness.

允许FDA访问与组织卓越性以及产品级安全性和有效性有关的数据元素的分析。

At least during the pilot phase of this program, FDA expects to receive the collected postmarket information on an ongoing basis.

至少在该计划的试验阶段，FDA希望能够持续接收所收集的上市后信息。

Following an interactive FDA review and mutual agreement that the proposed [Real-World Performance Analytics, or RWPA] plan is adequate for monitoring safety and effectiveness of the [software as a medical device, or SaMD] product, Pilot Participants will collect and analyze RWP data elements through the processes and methodologies defined by the organizations and share the selected RWPA with the FDA as agreed.

经过FDA的交互式审查并相互同意，拟议的[Real-World Performance Analytics，或RWPA]计划足以监视[软件，作为医疗设备或SaMD]产品的安全性和有效性，试点参与者将收集和分析RWP通过组织定义的过程和方法来收集数据元素，并按协议与FDA共享选定的RWPA。

FDA then justifies their added requirements of sharing postmarket data by explaining why the agency wants access to the company’s postmarket data:

FDA随后解释了为何要访问该公司的上市后数据，从而证明了其共享上市后数据的附加要求：

Robust reporting and increased transparency in postmarket data collection on the part of precertified organizations may also enable FDA to explore opportunities to optimize existing postmarket obligations for precertified organizations, including reporting and inspection requirements. …Selected RWPA tracking real-world health analytics, user experience analytics, and product performance analytics could be submitted to FDA on a periodic basis….Regular access to RWPA would increase FDA familiarity with the types of data being collected and any potential signals being monitored by precertified organizations. The Agency believes that this increased familiarity would facilitate collaborative engagement between FDA and precertified organizations when products require modifications or updates. Collaborative engagement aimed at rapidly addressing any postmarket safety or security concerns, in turn, may reduce the need for compliance actions and streamline review of any required product modifications….While access to RWPA would allow FDA to work with individual manufacturers to maintain product quality and ensure patient safety, it additionally might enable FDA to identify potential or emerging issues across product classes and to notify manufacturers before product quality is affected.

预先认证组织的稳健报告和提高上市后数据收集的透明度也可能使FDA能够探索机会，以优化对预先认证组织的现有上市后义务，包括报告和检查要求。 …可以实时将选定的RWPA跟踪现实世界的健康分析，用户体验分析和产品性能分析提交给FDA。定期访问RWPA可以使FDA更加熟悉所收集的数据类型和所监测的任何潜在信号由预先认证的组织。 FDA认为，当产品需要修改或更新时，这种增加的熟悉度将促进FDA与预先认证的组织之间的协作参与。 反过来，旨在Swift解决任何售后安全或安保问题的协作参与可能会减少合规行动的需要并简化对任何必需的产品修改的审查…。尽管可以使用RWPA，但FDA可以与各个制造商合作以保持产品质量并为了确保患者安全，它还可以使FDA识别产品类别中潜在或正在出现的问题，并在产品质量受到影响之前通知制造商。

Notice that FDA is trying very hard to sell the idea as an advantage to industry. I could keep quoting the document, but you get the point. FDA really wants to see the postmarket data. Plus they want to see it on a periodic basis, and we don’t know what periodic means. It seems clear from the context, however, that it must be reasonably frequent or those so-called benefits would not be realized.

请注意，FDA正在努力将这一想法作为对行业的优势进行销售。 我可以继续引用该文档，但是您明白了。 FDA确实希望查看上市后数据。 另外，他们希望定期查看它，而我们不知道定期意味着什么。 但是，从上下文来看，显然必须经常这样做，否则那些所谓的好处将无法实现。

2. 2019 AI概念文件 (2. 2019 AI Concept Paper)

While the precertification program would apply to all software used as a medical device, FDA was much more specific as to artificial intelligence in its April 2019 AI/ML Concept paper.[2] First, here are two definitions that are important to understand the basic structure of this AI/ML regulatory approach.

虽然预认证计划将适用于用作医疗设备的所有软件，但FDA在2019年4月的AI / ML概念文件中对人工智能的定义更为明确。 [2]首先，这里有两个定义对于理解此AI / ML监管方法的基本结构很重要。

SaMD Pre-Specifications (SPS): A SaMD manufacturer’s anticipated modifications to “performance” or “inputs,” or changes related to the “intended use” of AI/ML-based SaMD….

SaMD预先规格说明(SPS)：SaMD制造商对“性能”或“输入”的预期修改，或与基于AI / ML的SaMD的“预期用途”相关的更改。

Algorithm Change Protocol (ACP): Specific methods that a manufacturer has in place to achieve and appropriately control the risks of the anticipated types of modifications delineated in the SPS….

算法更改协议(ACP)：制造商已采用的特定方法来实现并适当地控制SPS中描述的预期修改类型的风险。

Using that framework, the agency then states its expectations with regard to postmarket reporting.

然后，该机构使用该框架陈述其对上市后报告的期望。

Through this framework, manufacturers would be expected to commit to the principles of transparency and real-world performance monitoring for AI/ML-based SaMD. FDA would also expect the manufacturer to provide periodic reporting to FDA on updates that were implemented as part of the approved SPS and ACP, as well as performance metrics for those SaMD. This commitment could be achieved through a variety of mechanisms.

通过这个框架，制造商将致力于基于AI / ML的SaMD的透明性和真实性能监控的原则。 FDA还希望制造商就批准的SPS和ACP的一部分实施的更新以及这些SaMD的性能指标向FDA进行定期报告。 可以通过多种机制来实现这一承诺。

Transparency may include updates to FDA, device companies and collaborators of the manufacturer, and the public, such as clinicians, patients, and general users. …[M]anufacturers may consider unique mechanisms for how to be transparent — they may wish to establish communication procedures that could describe how users will be notified of updates (e.g., letters, email, software notifications) and what information could be provided (e.g., how to appropriately describe performance changes between the current and previous version).

透明度可能包括对FDA，设备公司和制造商的合作者以及公众(例如临床医生，患者和一般用户)的更新。 …[M]制造商可能会考虑采用独特的机制来提高透明度-他们可能希望建立一种沟通程序，该程序可以描述如何向用户通知更新(例如，信件，电子邮件，软件通知)以及可以提供哪些信息(例如， ，如何恰当地描述当前版本与先前版本之间的性能变化)。

Real-world performance monitoring may also be achieved in a variety of suggested mechanisms that are currently employed or under pilot at FDA, such as … real-world performance analytics via the Pre-Cert Program. Reporting type and frequency may be tailored based on the risk of the device, number and types of modifications, and maturity of the algorithm (i.e., quarterly reports are unlikely to be useful if the algorithm is at a mature stage with minimal changes in performance over the quarter).

现实世界的性能监视也可以通过FDA目前采用或正在试用的各种建议机制来实现，例如…通过Pre-Cert程序进行现实世界的性能分析。 报告类型和频率可以根据设备的风险，修改的次数和类型以及算法的成熟度进行调整(例如，如果算法处于成熟阶段且性能变化最小，则季度报告不太可能有用)季度)。

The agency seems to have a larger vision of its own role than many outside the agency appreciate. The bottom line is FDA wants to become your business partner. They want to look over your shoulder through the entire marketing of the product to make their own independent judgments about what should happen next.

该机构似乎比其外部的许多人对自己的角色抱有更大的视野。 底线是FDA希望成为您的业务合作伙伴。 他们希望在产品的整个营销过程中都盯着您的肩膀，对接下来应该做什么做出自己的独立判断。

C.了解FDA的行为 (C. Understanding FDA’s Behavior)

During Part III of this article I’m going to make the case against expanded FDA postmarket oversight, and I want to be clear why I think it’s such a bad idea. Many of my arguments in part three will be grounded in organizational behavior,[3] a branch of psychology that says we all, to some extent, behave in response to organizational structure, purpose and culture in which we work.

在本文的第三部分中，我将提出反对扩大FDA上市后监督的理由，我想弄清楚为什么我认为这是个坏主意。 我在第三部分中的许多论点都将基于组织行为， [3]心理学的一个分支，它表示我们在一定程度上是对我们工作的组织结构，目标和文化做出React的。

Scholars have long suggested that government agencies use the “rational-legal authority” leadership framework to organize their work.[4] Almost 80 years ago, Max Weber enumerated a number of principles of governmental organization including: a formal organizational hierarchy, management by rules, organization by functional specialty, selecting people based on their skills and technical qualifications, and a purposefully impersonal environment (e.g., applying the same rules and structures to all members of the organization). As a federal agency with about 15,000 employees, based on FDA’s purpose and organizational structure we can predict certain styles of behavior for the organization as a whole.

长期以来，学者一直在建议政府机构使用“理性-法律权威”领导框架来组织其工作。 [4]大约80年前，马克斯·韦伯(Max Weber)列举了许多政府组织的原则，包括：正式的组织等级，按规则进行管理，按职能专业进行组织，根据其技能和技术资格选择人员，以及有目的的非人格环境(例如，对组织的所有成员应用相同的规则和结构)。 作为拥有约15,000名员工的联邦机构，根据FDA的宗旨和组织结构，我们可以预测整个组织的某些行为方式。

To understand the culture at FDA, we need to appreciate and frankly be sympathetic to the environment in which FDA must operate. FDA gets political and public applause for problems averted, but much more typically gets political and public criticism for injuries that result from medical products. Agency leaders routinely are called to testify before Congress about problems that Congress asserts FDA missed during its reviews.

要了解FDA的文化，我们需要赞赏并坦率地同情FDA必须在其中运行的环境。 FDA因避免了问题而获得了政治和公众的掌声，但更常见的是由于医疗产品造成的伤害而受到了政治和公众的批评。 通常要求机构负责人就国会断言FDA在审查过程中遗漏的问题向国会作证。

More than that, FDA faces an enormous challenge in recruiting talented scientists to work for a federal agency where salaries cannot keep pace with industry. While this causes tremendous turnover, more fundamentally it means that the agency must adopt a different recruiting tactic than dangling paychecks. The recruiting strategy at FDA is quite simple: it is mission driven. Come to FDA and make a difference. That appeals to many, but it has particularly appealed to those who want to make a big difference. The way people at FDA make a big difference is by preventing harmful products from reaching the market. The agency’s whole orientation is around its regulatory mission.

不仅如此，FDA在招募有才能的科学家到联邦机构工作时面临着巨大的挑战，因为那里的薪水无法与行业保持同步。 虽然这会导致大量人员流动，但从根本上讲，这意味着该机构必须采用不同于悬而未决的薪水的招聘策略。 FDA的招聘策略非常简单：它是任务驱动的。 来FDA有所作为。 这吸引了许多人，但它尤其吸引了那些希望发挥重要作用的人。 FDA的人们与众不同的方法是防止有害产品进入市场。 该机构的整体定位围绕其监管使命。

Low to mid-level employees such as Division Directors have tremendous power to make decisions based on their own personal judgments, and those judgments are frequently quite conservative. FDA decisions should be science-driven, but science is often not definitive and judgment is required. While reasonable people can disagree, when senior officials attempt to overrule lower ones, the lower ranking FDA employees sometimes believe that FDA managers are acting for the wrong reasons, for example following broad policy statements that might come from the White House. FDA employees have complained to members of Congress that they think management at FDA follows political whims rather than science[5] and, as a result, leaders at FDA have become reluctant to overrule their employees. This gives those frontline employees tremendous power to make decisions they think are right.

中低层员工(例如部门主管)具有巨大的权力，可以根据自己的个人判断做出决定，而这些判断通常非常保守。 FDA的决定应以科学为依据，但科学通常不是确定性的，需要作出判断。 尽管有理智的人可能会不同意，但是当高级官员试图推翻低级官员时，低级FDA雇员有时会认为FDA经理的行为是出于错误的原因，例如遵循白宫可能发布的广泛政策声明。 FDA的雇员向国会议员抱怨说，他们认为FDA的管理遵循的是政治狂想，而不是科学[5] ，因此，FDA的领导人变得不愿推翻其雇员。 这使一线员工拥有强大的权力来做出他们认为正确的决定。

As such, for a company trying to innovate AI in the context of medicine, FDA as an institution would make about the worst possible business partner. While I explain more below, what I want to be clear about is that this is because of organizational behavior and psychology; it is not about individual people at FDA. As a matter of fact, many of the folks at FDA are among my favorite people in the whole world. They are typically smart, hardworking, and passionately dedicated. Most are very good at their jobs. Plus a huge number of FDA decisions get made every day that are well-grounded in fact, science and law. But not all.

因此，对于一家试图在医学领域创新AI的公司而言，FDA作为一个机构将成为最糟糕的商业伙伴。 尽管我在下面进行了更多解释，但我想清楚的是这是由于组织行为和心理所致； 这与FDA的个人无关。 事实上，FDA的许多人都是我最喜欢的人之一。 他们通常很聪明，勤奋，热情奉献。 大多数人都很擅长工作。 加上每天都会有大量的FDA决策，这些决策实际上是科学，法律依据。 但不是所有的。

FDA has a job to do, as a federal law enforcement agency that includes special agents who actually carry guns.[6] It’s important to remember that FDA is not the NIH or the CDC for that matter. FDA catches criminals. They bring wayward, and sometimes dishonest, industry members back into line through threats, formal agency enforcement actions and judicial enforcement. On behalf of consumers everywhere, I am glad.

FDA要做的工作是作为联邦执法机构，其中包括实际携带枪支的特工。 [6]重要的是要记住，FDA并非NIH或CDC。 FDA抓获罪犯。 他们通过威胁，正式的机构强制措施和司法强制措施使行业成员任性而有时甚至是不诚实。 我代表世界各地的消费者感到高兴。

However, sometimes FDA uses other less formal coercion such as adverse media, perhaps when they feel they have a weak legal case or they simply don’t want to expend the resources. This is actually a common strategy at FDA. Take for example in 1991, when FDA Commissioner David Kessler ordered federal agents to seize more than 12,000 gallons of Citrus Hill orange juice amidst charges that the juice was falsely labeled as ‘fresh’. “Today’s action will send a clear message that the FDA will not tolerate such violations of the law,’ FDA Commissioner David Kessler told a conference in Palm Beach Gardens, Fla.”[7] Frankly using the media means that FDA doesn’t have to prove its case in court. They get their effect by trying to disgrace publicly the company. In the orange juice case, Procter & Gamble adamantly believed its use of the word “fresh” was lawful. Given their responsibilities, FDA plays hardball. I certainly don’t agree with some of the tactics the agency uses.

但是，有时FDA使用其他不那么正式的强制手段，例如不良媒体，也许是当他们觉得自己的法律案件不力或根本不想花费资源时。 这实际上是FDA的常见策略。 以1991年为例，当时FDA专员大卫·凯斯勒(David Kessler)下令联邦人员没收12,000加仑的柑橘山橙汁，理由是该汁被错误地标记为“新鲜”。 “今天的行动将清楚地表明，FDA将不会容忍这种违法行为，” FDA专员大卫·凯斯勒(David Kessler)在佛罗里达州棕榈滩花园的一次会议上说。 [7]坦率地使用媒体意味着FDA不必在法庭上证明其案件。 他们试图通过公开羞辱公司来取得效果。 在橙汁案中，宝洁公司坚信其使用“新鲜”一词是合法的。 鉴于他们的责任，FDA采取强硬态度。 我当然不同意该机构使用的一些策略。

None of those things makes the institution of FDA a good business partner. They are first and foremost a regulatory law enforcement agency, as they should be.

这些都没有使FDA成为一个好的商业伙伴。 他们首先是应有的监管执法机构。

D.判断最佳政策方法的标准 (D. Standard for Judging the Best Policy Approach)

In this series of articles, I’m proposing a policy approach to managing the risks of AI in healthcare while such products are on the market. In thinking about the best policy approach, our focus needs to be on patients. Whether looking at it from the perspective of industry, government, or healthcare providers, the common ground is that we all exist to serve patients. We have to put patients first.

在本系列文章中，我提出了一种策略方法来管理此类产品投放市场时在医疗保健中使用AI的风险。 在考虑最佳政策方法时，我们的重点应放在患者身上。 无论是从行业，政府还是医疗保健提供者的角度来看，其共同点都是我们都存在为患者提供服务。 我们必须把病人放在第一位。

In some ways that’s easier said than understood. Putting patients’ interests first is really not that simple. Doing what’s best for the patient involves the balancing of many factors, some of which are often in direct opposition to each other. Below I’ll list some of the larger factors, but I’m not going to go through each of these in detail. The point is simply that figuring out the best approach here is not based on one simple factor, even though the patient is at the center of the equation.

在某些方面说起来容易理解难。 将患者利益放在首位真的不是那么简单。 为患者做最好的事情涉及许多因素的平衡，其中一些因素常常相互直接对立。 在下面，我将列出一些较大的因素，但我将不对其进行详细介绍。 关键是，即使患者处于等式的中心，在这里找出最佳方法也不是基于一个简单的因素。

· Patient safety and effectiveness. Figuring out how to achieve safety and effectiveness for the patient involves the examination of scientific evidence, but the actual level of acceptable risk-benefit is ultimately still subjective. Under the Hippocratic Oath, doctors are first called upon to do no harm. But it’s important to understand that safety — which is usually thought of as harm prevention — often incorporates measures of effectiveness, in the sense that using a product that doesn’t work might mean that a patient is not using something else that would work. As a result, in designing AI, as with all medical devices, it’s important that the product not only have the appropriate level of safety, but also have evidence of effectiveness. For a product that adapts over time, this also means using postmarket vigilance to make sure that the product continues to be effective and we are continuing to use the product with appropriate patients, given the training of the software.

· 患者的安全性和有效性 。 弄清楚如何实现对患者的安全性和有效性需要对科学证据进行检查，但是可接受的风险收益的实际水平最终仍然是主观的。 在希波克拉底誓言下，首先要求医生不要伤害。 但是，重要的是要理解，安全性-通常被认为是预防伤害-通常包含有效性的度量标准，即使用不起作用的产品可能意味着患者没有使用其他可能起作用的东西。 因此，与所有医疗设备一样，在设计AI时，重要的是该产品不仅要具有适当的安全级别，而且还必须具有有效性的证据。 对于随时间变化的产品，这还意味着要保持售后警惕，以确保该产品继续有效，并且经过软件培训，我们将继续在适当的患者中使用该产品。

· Empowered patient decision-making. This is an important philosophical need. In medicine, we have the idea of informed consent with an emphasis on the word “informed.” Patients must know enough about the risks and benefits to make an informed decision on whether they want a certain kind of care. Frankly this issue also has economic overtones as we want to make sure that patients and their caregivers make appropriate purchasing decisions so that good products prevail in the marketplace and bad ones fall by the wayside. There’s also an element here of consumer confidence and trust that if purchasers believe that products are high quality and effective, the size of the market will increase and that has the effect of expanding access as well as product research and development.

· 赋予患者决策权 。 这是重要的哲学需要。 在医学上，我们的想法是知情同意，重点是“知情”一词。 患者必须充分了解其风险和收益，才能做出是否需要某种护理的明智决定。 坦率地说，此问题也具有经济意义，因为我们要确保患者及其护理人员做出适当的购买决定，以使优质产品在市场上占主导地位，而劣质产品则落在后面。 消费者信心和信任中还有一个要素，即如果购买者认为产品质量高，效果好，则市场规模将增加，并具有扩大进入和产品研发的作用。

· Patient privacy. There is, in the case of medical AI, a clamoring for transparency — information shared by software developers. As the developers consider sharing that information, they must make sure that doing so doesn’t unintentionally compromise patient privacy for those already treated.

· 患者隐私。 就医疗AI而言，要求透明性-由软件开发人员共享的信息。 当开发人员考虑共享该信息时，他们必须确保这样做不会无意间损害已经治疗者的患者隐私。

· Availability/access and outcomes. To state the obvious, technology like AI can only save lives if healthcare professionals and patients have access to it. In order to have access, obviously the products have to first be invented. Innovation is dependent on an adequate return on investment that requires protection of intellectual property. As obvious as all this sounds, it can often be undervalued in policymaking. Open source technology and transparency to the point of not protecting trade secrets will not always lead to adequate incentives for software development.

· 可用性/获取和结果 。 显而易见，像AI这样的技术只有在医疗保健专业人员和患者可以使用时才能挽救生命。 为了获得使用权，显然必须首先发明产品。 创新取决于需要保护知识产权的适当投资回报。 听起来似乎很明显，但在决策过程中它常常被低估了。 开源技术和透明性到不保护商业秘密的地步，并不总是会导致对软件开发的充分激励。

· Cost-effectiveness. In a similar vein, technology does no good if it isn’t affordable. Technology has to be considered in the larger scope of healthcare economics. Further, products have to be sustainable, and that may depend on whether the costs are paid by third-party payers or by the patient. Again, this seems obvious, but the cost consequences of regulatory policy are often undervalued.

· 成本效益。 同样，如果技术负担不起，技术也不会有好处。 必须在更大范围的医疗保健经济学中考虑技术。 此外，产品必须具有可持续性，这可能取决于费用是由第三方付款人还是由患者支付。 同样，这似乎很明显，但是监管政策的成本后果常常被低估了。

· Legal accountability. Federal regulation of medical technology exists to hold the developers and others in the chain of distribution accountable for the technology they distribute to the market. Consequently, the system needs to include appropriate government oversight and corporate responsibility.

· 法律责任。 存在联邦医疗技术法规，要求开发商和分销链中的其他人对他们向市场发布的技术负责。 因此，该系统需要包括适当的政府监督和公司责任。

The challenge is finding the right balance of all of those factors that optimizes the development of AI for patients. Those factors are not all aligned in the same direction. They require trade-offs among them.

面临的挑战是在所有这些因素之间找到适当的平衡，从而优化患者的AI发育。 这些因素并非都朝着同一方向发展。 它们需要在它们之间进行权衡。

结论 (Conclusion)

In this series of articles, I’ve tried to take those factors into account, balancing them against each other where necessary, in developing the proposed best practices found in Part II, as well as in the limited role FDA should play as explained in Part III.

在本系列文章中，我试图考虑这些因素，在制定第二部分中建议的最佳实践以及FDA应在第二部分中说明的有限作用中，在必要时将它们相互平衡。 三 。

[1] https://www.fda.gov/media/119722/download

[1] https://www.fda.gov/media/119722/download

[2] https://www.fda.gov/files/medical%20devices/published/US-FDA-Artificial-Intelligence-and-Machine-Learning-Discussion-Paper.pdf

[2] https://www.fda.gov/files/medical%20devices/published/US-FDA-Artificial-Intelligence-and-Machine-Learning-Discussion-Paper.pdf

[3] Steven McShane and Mary Von Glinow, Organizational Behavior, 8th Edition, 2018.

[3] Steven McShane和Mary Von Glinow，组织行为，第8版，2018年。

[4] Weber, Max, The Theory of Social and Economic Organization. Translated by A.M. Henderson and Talcott Parsons. London: Collier Macmillan Publishers, 1947.

[4] Weber，Max，《社会经济组织理论》。 由AM Henderson和Talcott Parsons翻译。 伦敦：Collier Macmillan出版社，1947年。

[5] https://www.medpagetoday.com/washington-watch/fdageneral/30954

[5] https://www.medpagetoday.com/washington-watch/fdageneral/30954

[6] https://www.reuters.com/investigates/special-report/usa-fda-cases/

[6] https://www.reuters.com/investigates/special-report/usa-fda-cases/

[7] https://www.upi.com/Archives/1991/04/24/FDA-seizes-orange-juice-in-labeling-battle/9056672465600/

[7] https://www.upi.com/Archives/1991/04/24/FDA-seizes-orange-juice-in-labeling-battle/9056672465600/

翻译自: https://towardsdatascience.com/postmarket-responsibilities-for-medical-ai-part-i-fdas-plan-to-increase-its-oversight-dae6d840ffc6

matlab .fda文件