作者：美国FDA局长

来源：药明康德（ID：WuXiAppTecChina）

编者按：近日，美国FDA局长Scott Gottlieb博士在华盛顿举行的2018 Health Datapalooza大会上发表了关于人工智能与数字医疗的演讲。Scott博士表示，必须始终把保护患者放在工作的首位，确保FDA的监管方法能维持检验产品安全性的科学黄金标准。他还从数字健康创新行动计划、启动将数字健康应用于药物的计划、人工智能、推出新的上市前数字安全计划和新的FDA数字健康孵化器五方面，阐述了FDA在数字化医疗时代的工作重点。下面就让我们一起来浏览Scott Gottlieb博士的完整演讲。

▲FDA局长Scott Gottlieb博士近日发表了关于人工智能与数字医疗的演讲（图片来源：FDA）

19世纪著名医生William Osler曾经说过：“好的医生治疗疾病，而伟大的医生治疗得病的患者。”但在20世纪，医学界通常将患者视为一系列症状的集合，而不是一个个体。这种情况随着医疗服务系统的发展而加剧，它让患者的护理变得零散。

数字健康有可能使William Osler医生以患者为中心的护理愿景变为现实。数字健康工具有巨大的潜力，可以提高我们准确诊断和治疗疾病的能力，并加强对个人的医疗服务，真正做到以患者为中心。

我们常常倾向于高估一项技术的短期影响，却低估它的长期效应。在我看来，数字健康的未来就是现在。FDA正在采取一些新的措施，以确保能从中受益。最值得注意的是，我们正在扩大数字医疗工具的机会，使之成为药物审查的一部分，并将这些能力与药物递送结合，形成药物递送系统。

我们正在通过新的预先认证项目，来扩展作为医疗设备的数字健康工具的上市前审查新模式。我们正在实施一种新的方法，来审查人工智能。我们还宣布了一项新的数字健康工具，来应用于我们自己的工作——药物安全性的上市前审查。最后，我们正在推出一个新的数字健康孵化器。我会依次谈谈这些举措，以及它们如何建立在我们更广阔的视野之上。

强大的数字健康技术已经得到推广，其它技术也会在未来几年得到推广。例如，移动健康应用程序（apps）已经在帮助消费者管理自己的健康，通过更好地控制饮食和生活方式。越来越多的移动健康apps在糖尿病预防、哮喘和成瘾恢复等领域显示出前景。根据业内估计，到今年年底，美国50%的智能手机和平板电脑用户会下载移动健康apps。

除了给患者赋能，数字工具也通过新的数据流为医生提供了解患者健康状况的全面视角。这些工具正在改变医生的工作流程，为患者提供更好的护理。2017年，FDA共批准了51款数字健康产品，这反映了连通和顺畅监测的趋势。这些产品中包含一款可以嵌入精神分裂症药物的传感器，这让患者可以通过医疗apps与医生共享治疗数据。

数字健康也能真正改善医疗效果，提高疗效，降低成本。以支持临床决策的医疗影像软件为例，诊断中风需要争分夺秒，因为血栓可能导致患者的功能严重丧失，并增加患者中风的几率。今年2月，FDA批准了一款临床决策支持软件，该软件使用人工智能算法以更快的速度通知神经血管专家，从而缩短诊断时间，及早采取有效的治疗挽救患者。

▲获批的软件采用人工智能算法，可以分析出癫痫发作的时间（图片来源:Empatica）

FDA在支持这种持续创新方面发挥了关键作用，这是我们保护和促进公共健康的使命的一部分。首先，我们必须确保我们的监管方法能维持我们检验产品安全性的科学黄金标准。我们必须始终把保护患者放在工作的首位。

作为使命的一部分，我们还必须采取措施，确保有利的新技术能够被有效推进，并及时为患者提供服务。因为最终，患者要从科学中受益。

这意味着我们还必须认识到，FDA的监管方法并不总是适合数字医疗等新兴技术或这一领域的快速变化。如果我们希望患者从创新中受益，FDA本身必须像我们正在监管的技术一样具有灵活性和创新性。

这要求我们在数字健康等高度创新领域要采取现代灵活的监管方法，鼓励更多的研发者将先进技术转化为临床工具，使患者受益。为了帮助推动这些机会，我在去年夏天宣布了数字健康创新行动计划（Digital Health Innovation Action Plan）。这一新计划概述了我们为重新设计FDA的方法所做的努力，以确保所有人都能及时获得高质量、安全和有效的数字健康产品。

我致力于将数字医疗作为一种工具来为消费者赋能，并打破制约提供医疗服务的障碍。为了扩大这一进展，我制定了几个关键目标。其中包括增加FDA数字医疗人员的数量和专业知识，在CDRH内推出数字健康软件预先验证试点项目，并发布新指南以实现我们的政策现代化，概述我们推动数字健康创新的努力。

数字健康创新行动计划

 

我很高兴地宣布，我们已经在这些目标上取得了实质性的进展。除了我们所取得的成就之外，我还想宣布几项首次推出的新举措。

我们发布了一项新政策，将简化包含多种功能的数字健康产品的路径，其中一些作为医疗器械受到FDA的监管。这项新指南是我们行动计划的另一部分。它解释了FDA对这些多功能数字设备的监管方法和政策，并阐明了FDA将会或不会审查这些设备中包含的某些软件，只要它们不会给FDA审查下的功能带来风险或不利影响。

 

例如，对于检测心脏节律等重要信息并将其传输到患者的电子健康记录的医院监视器，FDA只会审查心脏监护功能，除非传输功能对监护功能的安全性或有效性产生不利影响。在这种情况下，研发者只需要证明他们已经解决了两种不同功能之间存在的任何潜在不利影响。

▲ 图片来源：123RF

我们的目标是让研发者能够有效地将最新技术纳入其产品中，同时将FDA的审查重点放在用于诊断或治疗的高风险医疗设备的安全性和有效性上。我们相信这种方法将鼓励在这个重要领域取得更多创新。

我很高兴宣布该计划的另一个重要更新。我们已经取得了工作模式的初稿，它为该计划的各个方面提供了愿景，以及扩展计划的步骤。这一模式已经在我们的网站发布。我们还分享了一个新的路线图，概述我们将如何开发该计划。

这是我们对这个新计划的工作模式的几次迭代的第一个高级草案。我们需要推进的关键部分是来自研发者、患者、医生和公众的意见。因此，在整个工作模式中，你会看到我们已经制定了关于该计划各个部分需要答案和意见的“质疑问题”。我们设计的Pre-Cert项目是一个迭代的协作式体验，你的反馈是其成功的关键。

我们致力于在2018年底推出“Pre Cert 1.0”，这是该计划的首个版本。一旦我们牢牢掌握了这个框架，会在2019年对其进一步完善。

启动将数字健康应用于药物的计划

我宣布我们将扩大机会，使用数字健康工具作为药物开发的一部分，并在这个过程中实现新的创新，以改善患者的护理。

 

当我们首次开始讨论数字健康在药物开发中的作用时，我们将努力确保我们的监管方法能反映这些产品的新颖性，并鼓励和支持他们的创新。我们必须认识到，数字健康有潜力成为提高药物递送安全性和有效性的新工具。

我们知道，为了实现这些机遇，我们需要制定明确的政策，以确定如何将数字健康工具的审查和验证纳入药物开发计划。

与特定药物相关的移动设备和软件可以帮助患者持续接受治疗，因为药物依从性历来是一项挑战。例如，它可以帮助患者和医生确认患者已经服用了药物，并且可以很容易地将信息整合到电子健康记录中。再举一个例子，软件可以帮助癌症患者监测治疗的副作用，通过使用智能手机摄像头和面部识别软件来客观分类疼痛症状，并追踪认知表现。

▲可以追踪精神分裂症患者服药情况的Abilify MyCite系统（图片来源：Proteus Digital Health）

它也可以让开发者遵守上市后的监督要求。在某些情况下，这些工具可以通过嵌入智能设备中的软件或传感器收集数据，以支持安全性和有效性声明。

为了帮助扩大数字健康的潜力来实现这些可能性，我们将通过新的指南来推进这方面的政策框架。我们将向公众征询意见，如何将软件应用于处方药中，FDA如何促进数字健康功能发展的创新，以及如何将这些创新产品整合到患者的高级治疗选择中。我们还将征询意见，如何支持那些作为批准药物一部分的数字健康工具的开发，以及如何在软件经历快速更新的过程中妥善进行监管。

FDA将开放一个公共档案，以寻求开发这些前沿技术的创新者的意见，以及希望从这些进步中受益的医生和患者的意见。

我们在FDA的使命是在一个平衡、基于风险的框架下，应用我们的规定，既能保护患者，又能使处方药中的数字化蓬勃发展。

最终，我们建立此框架的目标将是开发审查和批准数字健康工具的有效途径，作为药物审查的一部分，以便这些工具充分发挥潜力帮助我们治疗疾病，并鼓励软件和治疗的协作。

人工智能

人工智能（AI）是数字健康时代最具潜力的工具之一，特别是机器学习（machine learning）。FDA正在积极开发新的监管框架，以促进这一领域的创新，并支持使用基于人工智能的技术。因此，当FDA应用Pre-Cert计划时，将考虑到机器学习的最大优点之一：继续学习和改进的能力。对AI采用Pre-Cert计划，可以允许公司对其设备进行一些小的更改，而不必每次都提交审核申请。而且，FDA将确保监管框架的其它方面（如新的软件验证工具）具有足够的灵活性，以跟上这个迅速发展领域的独特属性。FDA对AI的监管是为了给患者建立适当的保护，确保这些新技术能够达到其安全性和有效性标准，从而为患者带来益处。如果它不具有保护和改善患者情况的能力，这项技术就不会被FDA认可。

FDA预计，未来几年将会有越来越多的基于AI的工具提交审核申请，首当其冲的是医疗成像设备。FDA对AI的态度也将着重关注其处理现实世界数据的方法，包括来自病理幻灯片、电子病历、可穿戴设备和保险索赔数据的结构化和非结构化数据。FDA认为，随着更多的真实世界数据被输入到AI算法中，AI工具可以变得更具预测性。

AI也有可能显著降低由慢性疾病并发症引发的成本。比如说，糖尿病视网膜病变是导致3000多万糖尿病患者视力丧失的最常见原因，也是工作年龄段的成年人视力障碍和失明的主要原因。进行早期筛查是预防严重视力问题的关键，但大约一半的糖尿病患者都未受到眼科医生的检查。最近，FDA批准了第一款将特殊照相机和AI相结合的医疗设备，用于检测糖尿病成年患者的轻度糖尿病视网膜病变。如果AI检测到轻微视网膜病变，患者的主治医生就可以将他们转诊给眼科专家进行进一步的检查治疗，以防发展出严重后果。

▲今年4月，美国FDA批准了首款使用人工智能检测糖尿病患者视网膜病变的医疗设备IDx-DR（图片来源：IDx官方网站）

未来，AI工具可以直接集成到智能手机或可穿戴设备中用于各种早期检测应用，从而减少昂贵的专科门诊费用，同时增加尽早发现潜在严重问题的可能性。为了跟上这些快速发展领域的创新步伐，FDA本身必须在内部利用数字医疗工具来开发新的监管工具。这些数字工具可以帮助FDA使药物和设备的开发更加高效、可预测，并且更能反映患者的真实世界体验，同时简化FDA的工作流程，并将协作变得更加容易。

推出新的上市前数字安全计划

FDA这次推出了一项新的上市前数字安全计划（Premarket Digital Safety Program）。该计划针对新药研究申请（IND），要求其依据加速安全报告规定，达到IND电子报告的数据标准。这个计划首先在FDA的肿瘤卓越中心（Oncology Center of Excellence）以及该机构的药物和生物制剂中心测试。

采取这个计划是因为传统的审核方式亟待改变。在传统审核中，IND的临床试验主办者必须在15天或7天内向FDA和参与调查的人报告严重和意外的疑似不良反应，具体取决于事件类型和严重程度。FDA在IND阶段的主要任务之一是确保患者安全。但模糊不清的流程导致工作效率低下。试验主办者向FDA提交PDF或者纸质版的上市前安全报告，审查员必须分别审阅每个报告，阅读数以千计的记录，来了解受试新药的安全性。审查员花费一大块时间来管理这些安全报告，而不是检测信号和审查医疗结果。这个过程不符合数字时代的要求。

因此，FDA最近开展了一个成功的试点，开发并证明了采用数字提交的可行性，在该过程中，将上市前安全报告作为易于可视化和分析的数据集传输给FDA。FDA相信，新的数字框架可以显著提高上市前安全审核流程的效率和准确性，当该计划全面实施时，每个月可节省数百小时的审核时间。

最后，为了将数据分析整合到监管决策中，FDA正在迈出新的一步，即创建一个名为INFORMED的内部数据科学孵化器（Information Exchange and Data Transformation）。这个新科技孵化器最初的重点，是在健康技术和癌症相关的领域进行监管科学研究，目标是推进数字健康工具的框架。正在进行的INFORMED项目是与Project Data Sphere合作的，后者是一个非营利性的开放式癌症数据库，目的是利用医学成像数据开发肿瘤动力学分类算法。FDA还与美国国家癌症研究所（NCI）合作开展联合研究，设计和开发数字生物标志物作为药物开发工具，这些基础研究类型将引导行业向前发展。另外，FDA还与哈佛在AI和机器学习领域开展了一项研究项目，重点是设计、开发和实施针对监管科学应用的机器学习和AI算法。

FDA采取的这些改进将推动创新发展。我们都知道，创新的最大障碍是新产品开发的成本和风险。将有前景的想法转化为产品，以延长和改善人类生活，可能需要数十年的时间。而新的数字工具可以帮助降低成本和风险，将产品与最可能受益的患者进行匹配，或更快地识别潜在的严重副作用。新的数字工具可以打破研究和现实世界之间的人为障碍，对两者都产生裨益。因此，这些技术也促使FDA重新思考自己的使命，以及如何在这个新颖的领域实现安全有效的创新。

附FDA官网原文：

Transforming FDA's Approach to Digital Health

Remarks by Scott Gottlieb, M.D.
Commissioner of Food and Drugs
Academy Health’s 2018 Health Datapalooza
Washington, DC
April 26, 2018

(Remarks as prepared for delivery)

The eminent 19th century physician William Osler once said that “the good physician treats the disease; the great physician treats the patient who has the disease.”
 
But for most of the 20th Century, the medical community generally treated the individual as a collection of symptoms, rather than treating the patient as an individual.

This was aggravated by the evolution of our health care delivery system, which fragmented patient care, and – at times – made patients come to feel like cogs in a wheel.

Digital health holds the potential to make Dr. Osler’s vision of patient-centered care come to fruition. 

Digital health tools have vast potential improve our ability to accurately diagnose and treat disease. And to enhance the delivery of health care for the individual, making medical care truly patient centric -- empowering the individual.

I know there’s often a lot of hype around the promises of digital health.  One can ask whether digital health is victim to Amara’s law – that famous maxim coined by Stanford University computer scientist Roy Amara. The axiom says, “we tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run."  

But when it comes to digital health, from where I sit – and based on what we’re seeing at FDA – the long run is now.

And I want to talk to you today about some new steps we’re taking to ensure we reap the benefits from this change.

Most notably, we’re expanding the opportunities for digital health tools to become a part of drug review, to couple these capabilities to drug delivery to form a drug delivery system.

We’re expanding on our novel model for the pre-market review of digital health tools as medical devices, through our new pre-cert program. We’re implementing a new approach to the review of artificial intelligence; and we’re announcing today a new application of digital health tools to our own work – in this case the pre-market review of drug safety.

And finally, we’re unveiling a new digital health incubator at FDA. I want to talk about each of these things today in turn, and how they build on our broader vision.

Powerful applications of digital health technologies are already being commercialized today.

And others will be commercialized in the next few years.    

For example, mobile health apps already are empowering consumers to become co-directors of their own health by helping them take more control of things like their diet and lifestyle. And there are a growing number of mobile health applications that have shown promise in areas as diverse as diabetes prevention, asthma, and addiction recovery. ,    

According to industry estimates, by the end of this year, 50 percent of the smartphone and tablet users in the U.S. will have downloaded mobile health applications.

In addition to empowering patients, digital tools are giving providers a truly holistic view of patient health and function through new data flows.  These tools are helping redesign physician workflow to better coordinate patient care.

The trend towards connectivity and seamless monitoring is reflected in the 51 digital health products authorized by the FDA in 2017.  One of those included a sensor embedded in a drug for schizophrenia that can allow patients to share treatment data with their physician through a medical app.

Digital health also offers real opportunities to improve medical outcomes, enhance efficiency, and reduce costs. 

Take the example of clinical decision support software for medical imaging: When it comes to diagnosing stroke, every minute counts.  Every minute that blood flow to the brain is obstructed can lead to major loss of patient function, and increase the odds that a patient will succumb to the stroke.

In February, the FDA approved a type of clinical decision support software that uses artificial intelligence algorithms to notify a neurovascular specialist faster than would be possible without the software, decreasing time to treatment.

Earlier interventions with effective clot busting drugs or surgery can save brain tissue, prevent deaths, and potentially save hundreds of thousands of dollars in hospital and rehabilitation costs for a single patient.

FDA plays a critical role in supporting this continued innovation as part of our mission to protect and promote public health. First and foremost, we must make sure that our approach to regulating these technologies maintains our scientific gold standard for product safety. We must always put protecting patients at the forefront of what we do.

As part of our mission, we must also take steps to make sure that beneficial new technologies can efficiently advance, and be made availably to patients in a timely way. This is ultimately how patients are going to benefit from science.

Doing so means that we must also recognize that FDA’s usual approach to medical product regulation is not always well suited to emerging technologies like digital health, or the rapid pace of change in this area. If we want American patients to benefit from innovation, FDA itself must be as nimble and innovative as the technologies we’re regulating. 

That requires us to take modern and flexible approaches to regulation when it comes to highly novel areas like digital health --- to help encourage more developers to translate advances into clinically actionable tools to benefit patients. 

We must give companies and investors the regulatory predictability they need to make the long-term investments, and take the significant risks, that innovation requires.

Investment into the digital health space has soared over the last year. Some of this new capital will help underwrite the development of tools that go on to help patients.

To help advance these opportunities, last summer, I announced the Digital Health Innovation Action Plan.

This new plan outlined our efforts to reimagine the FDA’s approach to ensuring all Americans have timely access to high-quality, safe, and effective digital health products.

I committed to an approach that would help leverage digital health as a tool to empower consumers and to break down the fragmented silos that constrain healthcare delivery. 

To expand progress in this area, I laid out several key goals.

Among them, was increasing the number and expertise of digital health staff at the FDA, launching the digital health software precertification pilot program within CDRH, and issuing new guidance to modernize our policies and outline our efforts to promote innovation in digital health.

Digital Health Innovation Action Plan

I’m pleased to announce that we’ve made substantial progress on these goals. In addition to updating you on what we’ve accomplished, I also want to announce several new initiatives that we’re unveiling for the first time today.

These new efforts will continue to advance our framework for how we help promote innovation in this field.

Today we’re issuing a new policy that streamlines the path for digital health products that contain several functions; some of which are subject to FDA’s regulatory oversight as medical devices, while other functions, in isolation, are not.

This new guidance is another part of our action plan. It explains FDA’s regulatory approach and policy for these multiple function digital devices and clarifies where the FDA will and will not be reviewing certain software functions included in such devices, provided they don’t increase the risk or adverse effect of the function under FDA review. 

For example, take a hospital monitor that detects and transmits vital patient signal information like heart rhythms into a patient’s electronic health record. Here, FDA would only review the heart monitor function unless the transmission function adversely impacts the safety or effectiveness of the monitor capability. In that case, the developer only needs to show that they’ve addressed any potential for an adverse impact between the two different functions – the one FDA oversees, and the one it does not.

In issuing this guidance, FDA is taking additional steps to clarify what technologies won’t fall under avoidable regulation. Our goal is to allow developers to efficiently incorporate into their products the latest advances in technology, while focusing FDA’s review on the safety and effectiveness of the higher-risk medical device functions that diagnose or treat patients. We believe this approach will encourage more innovation in this important field.

Additionally, we’ve made significant progress in the last year on our device center’s precertification pilot program.

Since announcing the nine pilot participants in September, the team has been working with the participants and other stakeholders to start assembling the essential building blocks of this unique and innovative program.

I’m pleased to announce another important update on the program. We’re making available the first draft of our Working Model, which provides our vision on various aspects of the program and our steps for expanding it. The model is posted to our website. We’re also sharing a new roadmap that outlines how we plan to develop the program.

The goal of this program is to develop a tailored and pragmatic framework that trusts the excellence of organizations, but also continually verifies the safety, effectiveness, and performance of software as a medical device. It leverages the transparency of a sponsor’s culture of quality and its evidence of organizational excellence.

This is the first, high-level draft of what will be several iterations of our working model for this new program.

The critical piece we need moving forward is input from developers, patients, providers, and members of the public. 

So throughout the working model, you’ll see that we’ve drawn up “challenge questions” about each component of the program that we want answers and input on. We’ve designed the Pre-Cert program to be an iterative, collaborative experience. Your feedback is key to its success.

We’re committed to launching “Pre Cert 1.0”, a first version of the program by the end of 2018. Once we get this framework firmly in place, it’ll be further refined in 2019.

In advancing our approach to digital health, we started by focusing on efforts by our medical device center, given that Center’s long experience regulating software incorporated into medical devices. However, digital health impacts almost every facet of health care. So today I want to announce some new steps we’re taking to broaden the opportunities offered by digital health across our other medical product centers.

Launch of Program to Apply Digital Health to Drugs

Today, I’m announcing that we’ll expand the opportunities to use digital health tools as part of drug development and, in the process, enable new innovation to improve patient care.

As we begin for the first time to address the role of digital health in drug development, we’ll work to ensure our regulatory approach reflects the novel nature of these products and encourages and supports their innovation. We must recognize the potential of digital health as a new tool to improve the safety and effectiveness of drug delivery.

We know that to enable these opportunities, we need clear policies for how the review and validation of digital health tools can be baked into drug development programs.

Mobile devices and software linked to specific drugs can help patients stay on therapy for treatments where medication compliance is traditionally a challenge.  For instance, it can help patients, and their physicians, confirm that patients have taken their medication and easily incorporate information into electronic health records. 

To take another example: Software can help cancer patients monitor side effects of their treatment by using smartphone cameras and facial recognition software to objectively classify pain symptoms and track cognitive performance.

It can also allow sponsors to comply with post-market surveillance requirements. In some cases, these tools might allow for safety and efficacy claims supported by data collected through software or sensors embedded in smart devices – like increased activity, improved mood, or greater social interactions for patients treated for severe depression.

These are just a few use cases.  Given the falling costs and increasing quality of ubiquitous smart devices, developers can find new ways to use cloud based services and tools to support patient health and support more effective treatments, tailored to a patient’s needs and preferences.

To help expand the potential for digital health to address these possibilities, we’ll be advancing a policy framework in this area through new guidance. And we’ll be seeking public input on the right approach to incorporating software that’s designed to be used with prescription drugs. 

We want to assist innovation in this space. So we’re asking how FDA can promote innovation in the development of digital health functions, as well as how these novel products can be integrated into advanced therapeutic options for patients.  We’ll also be requesting input on how to support the development of digital health tools that are included as part of approved drugs, and how to properly regulate in this space when we know that software undergoes rapid cycles of innovation, and is frequently updated and improved.  

FDA will be opening a public docket to seek input from both innovators on the cutting edge of developing these technologies, as well from as from the providers and patients that hope to benefit from these advancements. 

Building off the approach of our device center, we’ll clarify that not all FDA requirements apply every time a digital health tool is employed in relation to a prescription drug. 

There’s a wide variety of ways in which apps and software can enhance health care. And our mission at the FDA is to apply our regulations under a balanced, risk-based framework—one that protects patients and allows digitization in the prescription drug setting to flourish. 

Ultimately, our goal in establishing this framework will be to develop an efficient pathway for the review and approval of digital health tools as part of drug review, so that these tools reach their full potential to help us treat illness and disease, and encourage synergies between software and therapeutics that meet FDA’s gold standard for safety and effectiveness.

Artificial Intelligence

One of the most promising digital health tools is Artificial Intelligence, particularly efforts that use machine learning. 

AI holds enormous promise for the future of medicine, and we’re actively developing a new regulatory framework to promote innovation in this space and support the use of AI-based technologies. So, as we apply our Pre-Cert program—where we focus on a firm’s underlying quality— we’ll account for one of the greatest benefits of machine learning – that it can continue to learn and improve as it is used. 

Employing the Pre-Cert approach to AI may allow a firm to make certain minor changes to its devices without having to make submissions each time. And, we’ll make sure that other aspects of our regulatory framework, such as new software validation tools, are sufficiently flexible to keep pace with the unique attributes of this rapidly advancing field.

We know that to support the widespread adoption of AI tools, we need patients and providers to understand the connection between decision-making in traditional health care settings and the use of these advanced technologies.

We know that our approach to AI regulation must establish appropriate guardrails for patients. And even as we cross-new frontiers in innovation, we must make sure that these novel technologies can deliver benefits to patients by meeting our standards for safety and effectiveness. 

The technology won’t be scaled or reimbursed without that level of confidence that it protects and promotes patients.

We expect to see an increasing number of AI-based submissions in the coming years, starting with medical imaging devices, and we’re working with experts in the field.

This includes AI experts with prior experience in sectors like finance that are already widely using AI platforms for fraud detection. As we develop our own policies, we want to understand how these technologies can be validated, and how patients and providers can be confident that they’re reliable, unbiased, and will help improve health outcomes.

Our approach to AI will also focus on the ways in which real world data flows. This includes structured and unstructured data from pathology slides, electronic medical records, wearable devices, and insurance claims data. We want to better understand, and unlock ways, that this data can be used to inform development and validation of AI devices. 

In time, AI might even be taught to explain itself to clinicians. This field has already advanced much more quickly than we first expected, as have its practical applications to patients.

We expect that AI tools can become even more predictive as additional real world data is fed into these algorithms.

This technology also has the potential to significantly reduce costs from complications of chronic disease.

For instance, diabetic retinopathy is the most common cause of vision loss among the more than 30 million Americans living with diabetes and the leading cause of vision impairment and blindness among working-age adults. 

In early April, the FDA approved the first medical device that combined a special camera and artificial intelligence to detect greater than a mild level diabetic retinopathy in adults who have diabetes in a primary care setting.

Early detection is a critical part of preventing serious vision problems. And yet eye care specialists don’t screen about half of diabetic patients on an annual basis.

If the AI detects mild retinopathy, primary care physicians can refer patients to an eye care specialist.

Eventually, AI tools could be integrated directly into smartphones or wearable devices for a variety of early detection applications, reducing the need for expensive specialists visits, while increasing the likelihood that we’re catching potentially serious problems early.

These are no longer far-fetched ideas.

We know that to keep pace with innovation in these fast-moving fields, FDA itself must do more to leverage digital health tools and analytics internally to help the agency develop new regulatory tools and advance its own work.

These include tools such as digital biomarkers for early disease diagnosis, or using data from EHRs to enable pragmatic clinical trials at the point of care.

These kinds of tools can help us make drug and device development more predictable, efficient and more reflective of patients’ real-world experience.   As we do, it’ll become easier to streamline our workflows and foster collaboration.

Launch of a New Premarket Digital Safety Program

We’re announcing one of those new steps today, with the launch of a new Premarket Digital Safety Program.

This program enables a unified data standard for meeting electronic reporting requirements under the expedited safety-reporting regulations of an Investigational New Drug Application. This new program will initially involve the FDA’s Oncology Center of Excellence, and the agency’s drugs and biologics centers. Let me give you a brief overview of how this new premarket safety program works:

Sponsors of clinical trials conducted under an investigational new drug application must report serious and unexpected suspected adverse reactions to the FDA and participating investigators in 15 or 7 days, depending on the type and severity of the event. One of the FDA’s primary roles at the IND stage of development is ensuring patient safety.

Following a detailed examination of the submission and a review process for premarket safety reports, we found inefficiencies resulting from a disjointed, analog workflow.

Sponsors are submitting premarket safety reports to the FDA on PDF files or on paper via fax or hand-delivered mail.

Medical officers and clinical reviewers must review and track each report individually, reading thousands of narratives to understand the safety profile of investigational new drugs and biologics. A recent internal survey of the FDA’s office of hematology and oncology products revealed that reviewers spend an average of 16 percent of their time managing these paper-based premarket safety reports.

Mapping our workflow made it clear that a big portion of that time is spent on administrative and file management tasks, rather than signal detection and medical review.  

Reading PDF narratives is not an ideal way to go about detecting drug safety signals. It makes the existing analog process a burden on the FDA, sponsors, investigators, and Institutional Review Boards. More important, it makes it more difficult for reviewers to focus their time on the primary mission – monitoring the safety of patients.

Sifting the signal from the noise through PDF reports is not only difficult; it precludes the use of analytical methods for improved early detection of safety signals through post-market surveillance programs. Or better understanding therapeutic class effects to inform appropriate risk mitigation strategies early in clinical development programs.

This fragmented analog workflow is also a burden on the sponsor for meeting reporting requirements. 

This process was badly in need of change.

So the FDA recently ran a successful pilot developing and demonstrating the feasibility of digital submission process where premarket safety reports are transmitted to the FDA as data sets that can be easily visualized and analyzed. 

We believe that new digital framework can significantly improve the efficiency and accuracy of the premarket safety submission and review process, saving hundreds of review hours every month when the program is fully implemented.

Launch of FDA’s New Digital Health Incubator

Finally, to support the integration of data analytics into regulatory decision making, we’re taking another new step with the creation of an internal data science incubator called the Information Exchange and Data Transformation; or INFORMED.

The initial focus of this new tech incubator will be, among other things, the conduct of regulatory science research in areas related to health technology and advanced analytics related to cancer. Our goal will be to help modernize our framework for advancing promising digital health tools.

Among the ongoing INFORMED projects are collaborations with Project Data Sphere, a nonprofit open-access cancer data repository, aimed at developing algorithms for classification of tumor dynamics using medical imaging data.

We’re also working on a joint fellowship program with the National Cancer Institute to design and develop digital biomarkers as drug development tools. These types of foundational research will guide the industry forward.

We’ve also launched a fellowship program with Harvard on AI and Machine Learning focused on designing, developing and implementing machine learning and artificial intelligence algorithms for regulatory science applications. 

It’ll look at developing new clinical endpoints and signal detection methods for evaluation of the safety and effectiveness of therapies. These efforts also will help us develop new approaches for understanding variations in individual patient experience using diverse data sets from clinical trials, EHRs, and biometric monitoring devices.

It will also look at the development of principles and definitions for validity and strength of AI/ML-derived evidence in the context of product approval and regulations. These types of foundational advances will guide the innovation forward. As many of you know, the biggest barrier to innovation is the costs and risks of new product development.  It can take decades to translate promising ideas into products to extend and improve human lives. 

New digital tools can help us reduce those costs and risks by helping us match products with the patients who are most likely to benefit, or help us identify potentially serious side effects sooner. They can break down artificial barriers between research and real-world practice, improving both. These are enabling. They’ll spark other innovations in care.

And so, these technologies are also prompting FDA to rethink our own mandate, and how we enable safe, effective innovation in this novel area. We share your urgency to advance this innovation in care. And we’re committed to doing our part to support its benefits to patients.

Thank you.

本文來源：http://mp.weixin.qq.com/s?src=11×tamp=1526562362&ver=882&signature=2ShGV7rjs4f3pUFtxlpofSCOl4u2p2n-QQYz7L1CBiVSkWygCBYJ3cxu1oQUVRAl4WRD41yGRGjrw9dEBWrJbcmIBbGhp5SxY8rkGKvQLwBP-Tc8unnoMHyflqBZDGw6&new=1