传统的医疗器械已经彻底改变了患者的诊断和治疗，并在非常广泛的应用范围内改善了患者的生活质量。应用软件技术，包括数据科学、机器学习（ML）和一般人工智能（AI），涉及许多道德和监管方面的问题。然而，在不远的将来，这些技术将推动新一轮的创新，带来积极的医疗影响。
可穿戴技术正在快速发展，随着这种进步，我们有机会捕捉到明显不同于医院记录的连续健康数据。我们如何将奇妙的新传感技术与强大的软件相结合，以充分利用智能、安全的个人健康监测的机会？
Jacob Skinner是Thrive Wearables公司的首席执行官，他利用可穿戴技术来改善医疗保健服务，并使其大众化。他完成了实验物理学的博士学位，并致力于设计以人为本的技术超过10年。他在萨塞克斯大学的传感器技术研究中心设计了商业化的电生理传感器和应用。
Will Berriss是一名软件工程师，在该领域有超过20年的经验。他是英国特许工程师（CEng）及工程与技术协会成员（MIET），并拥有医学图像处理的博士学位。
Q1：您对目前的医用可穿戴设备状况有什么看法？
Skinner：在过去，医疗设备只是在医院和医生办公室使用的东西。我们现在看到的，是我所说的医疗设备的消费化，在这种情况下，它们仍然经过医疗器械认证，并依据严格的标准制造，但它们不一定是生命攸关的；而是更倾向于监测、远程病人护理，以及支持虚拟病房。
这是行业进步的证明，因为这些医疗设备可以由人们根据自己的条件来使用，它们更容易获得。这些设备在外形上没有那么累赘，使用起来也不那么复杂。真正有趣的是这种情况发生的方式和这个领域的潜力。例如，Apple Watch是一个消费类电子设备，但在非常特殊的条件下，它也是一个医疗设备，测量特定的心电图信号。从监管的角度来看，哪些设备是医疗设备，哪些不是，仍然很清楚，但对用户来说，其界限越来越宽泛。
Berriss：感觉目前的情况还没有完全达到。我们只做了初步尝试，但如果应用程序和设备能够变得更加标准化并被采用，那就太好了。要做到这一点，技术可能生成的测量结果需要交付给临床医生，并被认为对护理可靠。
这可能会导致更多的个性化医疗和治疗，甚至更适合病人的情况。
Q2：您如何看待数据科学、机器学习和人工智能的使用对可穿戴技术行业的革新？
Skinner：医疗设备和人工智能之间有一种天然的不协调，从医疗风险的角度来看，这种不协调很棘手。从监管的角度来看，使用动态算法非常困难，因为它们可能导致结果的多样性，所以你必须非常小心地限制模型，并确保结果在指定的范围内。
Berriss：在我看来，现在有很多数据是由私人公司采集的，但未来，部分或全部数据可以集中提供，我说的集中是指通过英国的NHS或美国的Medicare或Medicaid提供。对一样东西，比如说一个肿瘤，或者甚至与多个病人有关的大量数据进行智能处理，可以更好地了解肿瘤的边界，例如，它可能增长或可能不增长的速度，这最终有助于管理它的人做出有依据的决定。
Q3：这些技术能否推动新一轮的创新？
Skinner：是的。这是它的核心。这些机会的存在是因为可穿戴技术和传感器的进步，这拓宽了医疗设备的概念。例如，肿瘤扫描一直都是数据密集型的，但现在相关数据的广度肯定在变化。
以前没有人知道一个人一天走了多少步；现在这是我们大多数人都能获知的事情，而且是显而易见的。你可以得出一些与运动有关的基本东西，然后你可以更进一步地利用这些信息。例如，你可以不断地测量心率。如果你有患心脏病或心力衰竭的风险，但有机会让你能够在NHS的这些重大负担和主要死因中更早地发现相关风险，我认为这就是更大的机遇所在。但是，知情并不总是一件好事，你不能只是给人们提供让他们感到震惊或他们无法正确解释的信息，但你也不能妨碍他们的知情权。归根结底，如果你能预测一个人将会生病，那就是另一个完整的资源要求。
Berriss：当然，例如，这些技术正用于金融界以及工程和太空。随着更多的人工智能和数据科学工作的发生，再加上日益流行的趋势，更多的见解和改进将会产生，将会有越来越多的与健康产业相关的东西。
如果有一屋子的人，一半是健康的，一些有心脏疾病，然后你去做一些测量 - 这就是重点。真的，因为模式可能在我们尚不清楚具体位置的数据中 - 人工智能通常会发现数据中的这些差异，可能在表面上看不出来。这就是它真正有趣的地方 - 模式识别和解决人类无法解决的问题。
Q4：你如何设想这项技术被用于改善人们的健康？
Skinner：这有太多的答案了! 让我们从大众化开始。如果人们懂得照顾自己的健康，那么这是一个很好的起点，因为专业卫生人员不可能在任何时候都照顾到所有人。首先，这些设备可以持续测量用户的健康状况，将信息反馈给用户，并在需要时提供给专业医疗人士。这就是可穿戴技术、持续数据、预测性和预防性医疗方面的实质性胜利。
因此，举例来说，某人是否在往不对劲的方向发展，并且在两年后就会发生心脏病？拥有这样的洞察力将是非常有益的，以便有两年的缓解策略，并使病人能够掌握自己的身体健康状态，并选择明智的生活方式。关于虚拟病房和远程病人监测，医院和家庭之间的界限正变得模糊，所以我认为真正的进步将在于半导体医疗技术或监测设备，它们不属于病危护理，却真正擅长预测性护理。
Berriss：我认为这将通过让人们更多地参与，来改善他们的健康。目前，你可以在家里测量血压，去见临床医生时，你可以提供一个附件，上面有上周的生命体征数据。但这取决于我们是否能达到这样的程度，即测量健康信号的某些方法可以非常标准化，以至于用户和专业医疗人士可以对数据的有效性感到放心。这显然需要监管，所以这些数据在医学上是有用的，而且事后不需要复制，这就是需要弥补的差距。
Q5：如何使软件足够安全，使其能够用于个人生理数据的持续监测？
Berriss：安全和保障最终要归结为对数据进行加密，并对你分享数据的途径以及分享数据的方式保持谨慎，特别是在持续监控方面。这方面已经有了很多进展，像苹果公司已经对Apple Watch进行了加密，所以它不能与任何蓝牙设备连接，以获取数据。因此，在某些方面，我们所需要的已经成为可能，但反过来，这是否能在法律上得到证明。在法庭上，我们有证据证明数据没有被篡改或落入坏人之手吗，这能成立吗？
Skinner：医疗软件必须经过高度验证和测试，因为它在连接对象、如何处理数据、谁可以访问这些数据方面要保守得多；从本质上讲，它更像是一个沙盒。因此，医疗软件在处理生理数据时一般需要额外的成本和时间投入。我认为，要考虑的一个更关键的问题是，鉴于数据包含有价值的个人信息，有被滥用的风险，那如何能安全地管理和存储数据。如果使用得当，这些数据的价值是巨大的，但也存在着巨大的风险。
通过区块链或其他相关访问协议的所有权概念很吸引人。当与加密和健康记录相结合时，它创造了一个非常有趣的空间。我相信，这种结合在不久的将来会变得越来越重要，并产生重大影响。此外，如果个人拥有更多的数据，并希望以自己的条件访问这些数据以获得自己的医疗保健，他们可能需要在某些条件下获得对其他人的汇总数据的控制性访问，以比较他们的数据并做出结论。这意味着需要对目前的数据使用方式进行巨大的改变，目前的数据使用方式是极其自上而下的，并且存储在大的数据库中，访问权限有限。转移数据或对数据有所洞悉并不容易，而且目前数据的使用方式可能非常有用。
此外，还有可能以尊重隐私和匿名的方式购买和出售数据，这对研究目的很有用。然而，这还没有公开或大规模地进行。
Berriss：使数据匿名的一种方法是生成分配给用户的随机数字代码，而不是真实姓名。在COVID-19追踪应用程序中，他们使用数字代码来识别用户，而不是他们的真实姓名，这可能是一个潜在的途径。所以，这并非不可能，只是需要更多的努力。
Q6：其中有哪些道德和监管方面的问题？
Berriss：我想，从道德的角度来看，使用区块链来保持安全性，并在个人和公司或医疗机构之间订立合同是很重要的。许多年来，人们一直担心自己的个人数据被用来对付自己。例如，如果你测量了某些健康指标，并发现你有可能在10年内导致你死亡的疾病，你可能不希望与你的健康保险公司分享这些信息，因为他们可能会拒绝为你提供已知或预测的疾病保险。
因此，个人在使用追踪其健康数据的设备之前，了解其潜在的影响是至关重要的。重要的是，公司要预先披露他们可能发现的信息种类，以及他们可能与谁分享这些信息。这对那些更容易受到伤害的个人来说尤其如此，他们的健康数据可能会导致负面后果。
Skinner：是的，这是一个关键问题。如果我们不采取适当的措施，利用区块链和其他技术保护个人数据，这些信息极有可能受到黑客的攻击。这种漏洞的后果可能很严重，尤其是对NHS这样的医疗机构。在监管方面，我认为目前存在两个不同的问题。首先，正在被访问以及存储在各种数据库中的数据量呈指数级增长。这种数据流的增加正在造成一个问题，因为有一百万倍的数据和数百万的不同节点。令人担忧的是这些数据是如何被访问和存储的，以及与之相关的潜在风险。
第二个问题与医疗标准的放宽有关。虽然这种放宽可能有很好的理由，但也有相关的风险。美国食品药品监督管理局（FDA）通过了许多通常不会批准的技术。这方面的例子包括由于移动电话的兴起而出现的许多数字解决方案。这些数字系统通常被归入更严格的医疗设备法规之外（根据所谓的510(k)提交），以及无数的I类和II类设备，它们基本上只是以合格的方式感应和传递信息。如果把这一点做到极致，就会出现通常被冠以“健康”之名的设备，它们（在市场上）被定位为非医疗性质。这为医疗效果以及医疗技术的可信度定下了一个移动的目标。很难知道界线在哪里，而且很有可能技术进步和监管在大多数时候都不能很好地保持一致。这是一个细微的问题，但重要的是要意识到快速发展的技术在医疗领域的潜在影响。
Q7：将这些技术整合到可穿戴设备及其应用程序时，您认为有哪些挑战？
Berriss：我认为与我们在技术方面讨论过的任何东西进行整合的关键困难在于所涉及的数据量太大。首先，如果你想把数据传输到其他地方，就需要维持网络带宽问题，如果你不这样做，那么你就需要在本地处理，这就带来了一系列的挑战。传输如此大量的数据往往是不可行的。
如果集成到一个移动应用中，虽然手机能够处理复杂的任务，但仍有挑战，比如说，你要进行什么处理，什么数据会被传输。还可能存在一个问题，即用户对如此强大的设备有什么兴趣，这也可能影响到如何创建这样的设备，并确定它的形式。
Skinner：我相信主要的挑战在于证明可穿戴技术的价值。将使用技术的价值与它所产生的冲突联系起来的等式是公认的。从本质上讲，如果有人从使用一项技术中获得很多价值，他们就更有可能采用它。然而，需要近距离接触的以人为本的技术可能对个人空间有相当大的侵犯性，因此很难克服这一障碍。因此，为了鼓励技术的采用，该技术的价值必须被证明是非常高的。
即使在它可能意味着生与死的区别的情况下，比如可以检测潜在心脏病发作的可穿戴技术，如果感觉太过笨重或碍眼，人们仍然可能会抵制使用它。此外，即使该技术提醒用户有潜在的健康问题，他们也可能不会采取行动来解决这个问题。简而言之，最大的挑战是通过提高可穿戴技术的价值，使其超越它所产生的冲突，从而改善可穿戴技术的采用。
Q8：这个领域下一步会发生什么？请向我们分享您的预测。
Skinner：本质上，差不多是我们已经讨论过的内容。我们谈到的主题都有时间表。这些包括医疗保健的大众化，利用预防和预测措施让人们更好地了解和处理自己的健康状态，并通过这些措施减少医院就诊的次数。此外，通过利用虚拟病房和远程监控帮助人们早点回家护理，是我的主要预测。
Berriss：根据我的经验，我发现当你在媒体或这个期间听到它作为一个反复讨论的话题时，往往会知道什么会成为下一件大事。例如，我已经看到许多关于心率和心率变异性的讨论。如果你问我紧接着是什么，我会认为是这个领域的东西。
英文原文：
A conversation with Jacob Skinner and Will Berriss, Thrive Wearables
Conventional medical devices have revolutionized patient diagnostics and treatments and improved quality of life across a staggering breadth of applications. Applying software techniques, including data science, machine learning (ML), and general artificial intelligence (AI), has many ethical and regulatory dimensions. However, the future is heading rapidly toward a point where these techniques are driving a new wave of innovation and positive health impacts.
Wearable technology is advancing at a rapid pace and with this advancement comes the opportunity to capture a very different kind of continuous health data than that recorded in a hospital setting. How do we combine incredible new sensing technologies with robust software to take full advantage of the opportunity for intelligent, safe personal health monitoring?
Jacob Skinner is the CEO of Thrive Wearables, where he uses wearable technology to improve healthcare and democratize access to it. He completed a D.Phil. in experimental physics and has designed human-centered technology for over 10 years. He has designed commercially available electrophysiology sensors and applications at the University of Sussex's Sensor Technology Research Centre.
Will Berriss is a software engineer with over 20 years of experience in the field. He is a Chartered Engineer (CEng) and member of the Institution of Engineering and Technology (MIET) and has a Ph.D. in medical image processing.
What is your view on the current medical wearable device landscape?
Skinner: In the past, medical devices were things that we just used in hospitals and in doctors’ offices. What we see now is what I call the consumerization of medical devices, in which they are still medical device certified and built within strict standards, but they are not necessarily life critical; they are more geared to monitoring, remote patient care, and in supporting virtual wards.
It’s a testament to advances in the industry, because these medical devices can be used by people on their own terms and they are much more accessible. They're not as cumbersome physically or as complex to use. What is really interesting as well is the way in which this is happening and the potential in this space. For example, the Apple Watch is a consumer electronics device, but under very particular conditions it's also a medical device, measuring specific ECG signals. It's still clear from a regulatory point of view which devices are medical and which are not, but for the user the boundaries are increasingly broad.
Berriss: It feels like the current landscape is not quite there yet. We're sort of dipping a toe in the water, but it would be great if apps and devices could become more heavily standardized and adopted. For this to happen, the measurements that tech could generate would need to be delivered to clinicians and be considered reliable for care.
This could result in more personalized healthcare and treatment that is even more tailored to a patient.
How do you see the use of data science, machine learning, and AI revolutionizing the wearable tech industry?
Skinner: There's a natural dissonance between medical devices and artificial intelligence that is tricky to navigate from a medical risk point of view. From a regulatory perspective, it's very hard to use dynamic algorithms because of the diversity of outcomes they might lead to, so you have to constrain the models very carefully and ensure outcomes are within specified boundaries.
Berriss: The way I see things, there is a lot of data being captured by private companies but, going forward, some or all of that data could be made available centrally, and by centrally I mean the NHS in the U.K or, perhaps Medicare or Medicaid in the USA. A huge amount of data processed intelligently about one thing, say a tumor, or even in relation to multiple patients, could give a much better understanding of the boundary of the tumor, and, for example, how quickly it may or may not grow, which ultimately helps those managing it to make informed decisions.
Could these techniques drive a new wave of innovation?
Skinner: Yes. That's the core of it. These opportunities exist because of wearable technology and advances in sensors, which broaden the concept of medical devices. For example, tumor scanning has always been data heavy, but the breadth of what data is relevant now is definitely changing.
Nobody used to know how many steps they took in a day; now it’s something most of us are aware of and can easily find out, and you can derive some basic stuff relating to exercise, but then you can take that so much further. For example, you could be measuring heart rate constantly. If you're at risk of heart disease or heart failure and there was a chance that you might be able to detect it just that little bit earlier, I think that's where the bigger opportunities are, in these big burdens on the NHS and big causes of death. But knowledge isn’t always a good thing, and you can’t just give people information that alarms them or that they can’t interpret properly, but you also can’t shield them from their rights to be informed. Ultimately, if you can predict that someone is going to get ill, that’s a whole other resource requirement.
Berriss: Definitely, these techniques are being used in the financial world and in engineering and space, for example. As more AI and data science work happens, and it becomes more popular, more of those insights and improvements will get generated and there will be more and more that is relevant to the health industry.
If you had a room full of people and half were healthy and some had a heart condition and then you go and take some measurements – and that’s the whole point, really, because the pattern could be in the data somewhere we don’t know about already – AI would typically spot those differences in the data that might not be visible on the face of it. And that’s where it becomes really interesting – pattern recognition and solving things that humans can’t.
How do you envisage this technology being used to improve people’s health?
Skinner: There are so many answers here! Let's start with democratization. If people are looking after their own health, then that's a great starting point, because health professionals cannot look after everyone at all times. First, these devices could start constantly measuring users’ health and feeding information back to users and potentially escalating to medical professionals as needed. That's kind of the bread and butter win in terms of wearable tech, constant data, and predictive and preventive healthcare.
So, for example, is somebody moving in the wrong direction and is two years away from a heart attack? Having that kind of insight would be incredibly beneficial in order to have two years’ worth of a mitigation strategy and to empower patients to take ownership of their physical health and make informed lifestyle choices. In regard to virtual wards and remote patient monitoring, it’s blurring the boundary between hospitals and homes, so I think the real advances will be in semi-medical technologies or monitoring devices that are not critical care but are really good at predictive care.
Berriss: I think it would improve people's health by involving them more. Currently, you can take your blood pressure at home, and when you go to see a clinician you could provide an attachment with vitals data from the last week. But it depends on whether we can get to a point where certain approaches to measuring health signals can be so standardized that users and medical professionals can feel confident in the validity of the data. This obviously needs regulating, so this data is medically useful and doesn’t need to be replicated afterward, and that’s the gap that needs closing.
How can software be made safe and secure enough for it to be used in the continuous monitoring of personal physiological data?
Berriss: Safety and security ultimately comes down to encrypting data and being careful with the way in which you share it and also how you share it, especially with continuous monitoring. There is a lot of progress in this already happening, with companies like Apple, which has encrypted the Apple Watch so it cannot connect with just any Bluetooth device to retrieve data from it. So, in some respects, what we need is already possible, but on the flip side is whether this can be legally proven. Would it hold up in a court of law that we have proof the data hasn't been tampered with or been placed into the wrong hands?
Skinner: Medical software has to be highly validated and tested, as it's much more conservative in terms of what it's connecting to, how it's processing data, who's got access to it; essentially, it's all much more of a sandbox. So, there's a general additional cost and time investment required in medical software processing physiological data. I believe that an even more crucial question to consider is how data can be managed and stored securely, given that it contains valuable personal information that is at risk of being misused. The value of this data is enormous if it is used appropriately, but there is also a significant risk.
The concept of ownership through a blockchain or other associated access protocols is fascinating. When combined with encryption and health records, it creates a very interesting space. I believe that this combination will become increasingly important and have a significant impact in the near future. In addition, if individuals have more data and want to access it on their own terms for their own healthcare reasons, they may need to be given controlled access to other people's aggregated data on certain terms to compare their data and make conclusions. This represents a need for a sea change in how data is currently used, which is extremely top-down and stored in big databases with limited access. It is not easy to shift data around or gain insights from it, and it is not currently being used in a way that could be incredibly useful.
Furthermore, there is the potential for buying and selling of data in a way that respects privacy and anonymity, which could be useful for research purposes. However, this is not being done openly or on a large scale yet.
Berriss: One way to anonymize the data could be by generating random number codes assigned to users instead of real names. In the COVID-19 tracking apps, they used a number code to identify people instead of their real names, and this could be potentially one avenue to follow. So, it’s not impossible, it just needs more work.
What are the ethical and regulatory dimensions at play here?
Berriss: I suppose using a blockchain to keep things secure and establish contracts between individuals and companies or medical bodies is important from an ethical perspective. For many years, people have been concerned about their personal data being used against them. For example, if you measure certain health metrics and discover that you are predisposed to a condition that could lead to your death in 10 years, you may not want this information to be shared with your health insurance company because they could deny you coverage for known or predicted conditions.
Therefore, it is crucial for individuals to understand the potential implications before using devices that track their health data. It is important for companies to disclose up front what kind of information they may discover and with whom they may share this information. This is especially true for individuals who are more vulnerable and may have health data that could lead to negative consequences.
Skinner: Yes, this is a critical issue. If we don't implement proper measures to secure personal data using a blockchain and other technologies, it is highly likely that this information will be vulnerable to hacking. The consequences of such breaches could be severe, especially for healthcare agencies like the NHS. In terms of regulation, I think there are two different issues taking place. The first is the exponential increase in the amount of data that is being accessed and stored in various databases. This increase in data flow is causing a problem, as there is a million times more data and millions of different nodes. The concern is how this data is being accessed and stored and the potential risks associated with it.
The second issue is related to the relaxation of medical standards. While there may be good reasons for this relaxation, there is a risk associated with it. The FDA is allowing many technologies to pass that traditionally wouldn't have. Examples could include many digital solutions that have come into being due to the emergence of mobile phones. These digital systems are usually classified outside of the more stringent medical device regulations (under what is called a 510k submission), as well as a myriad of Class I and II devices, which are essentially just sensing and passing on the information in a qualified way. Taking this to the extreme leads to what are often termed “wellness” devices, which are very much positioned (in the market) as non-medical in nature. This creates a moving target in terms of what medical efficacy is and what medical technology credibility is. It's hard to know quite where the line is, and there's a strong chance that technology advances and regulation will not be well aligned most of the time. This is a nuanced discussion, but it's important to be aware of the potential implications of fast-tracked technologies in the medical field.
What challenges do you see in the integration of these technologies into wearable devices and their apps?
Berriss: I think the key difficulties with integrating with anything we've discussed on the technical side lies in the sheer amount of data involved. First, there are network bandwidth issues that need to be maintained if you want to transmit the data elsewhere, and if you don't, then you'll need to process it locally, which presents its own set of challenges. It's often not feasible to transmit such large amounts of data.
If you integrate into a mobile application, while mobile phones are capable of handling complex tasks, there are still challenges with what works, for example, what processing you do and what data will be transmitted. There also may be a question about what appetite users have for a device that's so powerful, which could also impact how such a device is created and define what form it takes.
Skinner: I believe the main challenge lies in proving the value of wearable technology. The equation that relates the value of using the technology to the friction it creates is well recognized. Essentially, if someone perceives a lot of value from using a piece of technology, they are more likely to adopt it. However, human-centered technologies that require close proximity can be quite invasive to personal space, making it difficult to overcome this barrier. Therefore, in order to encourage adoption, the value of the technology must be proven to be very high.
Even in cases where it could mean the difference between life and death, such as with wearable technology that could detect a potential heart attack, people may still resist using it if it feels too bulky or obtrusive. Additionally, even if the technology alerts them to a potential health issue, they may not take action to address it. In short, the biggest challenge is improving the adoption of wearable technology by increasing its value beyond the friction it creates.
What’s next in this area? Please give us your predictions.
Skinner: Essentially, I would say more of what we've already discussed. The themes we talked about all have timelines. These include the democratization of healthcare, using preventive and predictive measures for people to better understand and engage with their health, and reducing the number of hospital visits through these measures. Additionally, helping people go home sooner by utilizing virtual wards and remote monitoring are the key predictions.
Berriss: In my experience, I find you tend to know what's going to be the next big thing when you hear it as a recurring topic in the media or in this space. For example, I’ve seen many discussions about heart rate and heart rate variability. If you ask me what's immediately next, I would say something in that space.
