Things & Thinks XX
In this issue of my fortnightly newsletter, I review an interesting paper on AIML product approvals by USFDA and EU. I follow this up with a superb, thought-provoking reflections on user-centered design. As usual, I also curate some of the important news from the digital health world, and we end with a Chart of the Month and a Tweet of the Month.
AIML device approvals-comparison between USFDA and EU pathways
Let’s begin with this interesting paper published in The Lancet.
The authors reviewed governmental and non-governmental databases to identify 222 devices approved in the USA and 240 devices in Europe. Here are some key findings-
- The number of approved AI/ML-based devices has increased substantially since 2015, with many being approved for use in radiology.
- However, few were qualified as high-risk devices. Of the 124 AI/ML-based devices commonly approved in the USA and Europe, 80 were first approved in Europe. Among these, most were approved for radiology (80 devices [65%]), followed by 18 (15%) cardiovascular devices, and 9 (7%)
neurological devices.
The authors point out interesting observations and limitations of the ecosystems-
- One possible reason for CE mark in Europe before approval by the FDA might be the relatively less rigorous and decentralised evaluation of medical devices in Europe. This hypothesis is strengthened by an FDA report highlighting 12 devices that were only approved in Europe and later found to be unsafe or ineffective.
- Evidence on the safety and performance of the European approach to the regulation of medical devices, including AI/ML-based medical devices, is scarce. The current European Commission’s database on medical devices (Eudamed2) is a restricted access (ie, not publicly available) repository for information on market surveillance exchanged between national competent authorities and the European Commission.
- This doesn’t mean US pathway is better! The information available, via USFDA, in the summaries and statements of approved medical devices is vague and scarce, which makes it challenging to identify AI/ML-based devices and might confine the device’s capabilities.
- Finally, these two statements sum this paper up nicely!
There is no unified definition for AI/ML. A broader definition of AI/ML could lead to more AI/ML-based medical devices being approved by the FDA in the USA and CE marked in Europe. Second, it was not always possible to determine whether a medical device was AI/ML based (even after contacting the manufacturers).
Can there be too much of ‘user-centered design’?
In this post by Alexis Lloyd, Devin Mancuso, Diana Sonis and Lis Hubert, there are some thought provoking thoughts about user-centered design and its limitations. More importantly, they also suggest how to go beyond these pitfalls to build a truly well-designed product. Here are some points that resonated especially with me.
- User-centered design (UCD) was developed in reaction to the blind spots coming from a highly focused lens of business needs. UCD advocated for a design practice that instead focused on the person using the technology, and was intended to create experiences based on an understanding of their needs and goals.
- However, UCD brought with it its own limitations, which are getting clearer only now; primarily UCD leads to Obscuring participants, Obscuring friction and Obscuring possibility.
- There is a whole set of second-order experiences that we don’t actively design, but happen as a consequence of what we design. Which means that there’s the potential for a great deal of positive change that can be created simply by shifting how we look and what we look at.
- Obscuring participants: The authors cite Kevin Slavin to explain this focus on users, thereby obscuring the rest of the participants in the ecosystem-“When designers center around the user, where do the needs and desires of the other actors in the system go? The lens of the user obscures the view of the ecosystems it affects.”
- Obscuring friction: By privileging ease-of-use above all else, we have at times obscured friction to the detriment of users. We’ve over-optimized, creating experiences that are addictive, irresponsible, and at times, too easy to use. In addition, friction often doesn’t get removed from an experience, but instead is shifted on to other parts of the system.
- Obscuring possibility: Within the confines of our website or app we optimize for our product’s success, guiding the user through their ‘happy path’ to the intended engagement or transaction. But when we look at the user experience solely through the lens of optimizing for success, we often fail to design appropriate failure states for when things don’t go as expected.
So what are the remedies? The authors suggest the following strategies-
- Uncover the exploits: Go beyond the happy path. Take a “white hat” approach to design by actively exploring unintended consequences.
- If this, then what?: Consider second and third-order effects as well as alternative outcomes in order to understand the potential consequences of your system.
- System mapping: Make the invisible actors visible. Mapping the various relationships and transactions between the actors in our system allows us to better understand how each group is incentivized to behave.
- Design for excluded users: Practice conscious design by looking beyond expected user groups to include a diverse range of people with a need for your product/service.
- Ethics-oriented competitive research: Examine the impact of similar products or systems to gather insight on which outcomes you want your system to replicate and which you want to avoid.
I find this quite interesting in the context of healthcare and the hype of user-centricity that we see in this sector. Apart from the terribly high levels of lip-service many healthcare innovators give to being user-centric, there are of course tons of examples of half-baked applications, narrowly defined user personas and ill-considerations of the complex interlinkages within each healthcare micro-system. All of this sounds like a ready recipe for future problems!
Digital Healthcare news-
Several new updates both from tech and non-tech perspective-
- Big Tech in Healthcare: Both Google and Microsoft continued their health data innovations. Google announced expansion of its Care Studio , an ongoing pilot with Ascension, to an expanded userbase of clinicians and nurses. Google describes Care Studio as a software solution that provides a comprehensive view of a patient’s records and allows clinicians to quickly search through complex patient information. Microsoft announced that its Microsoft Cloud for Healthcare will soon get an update that will include new features for virtual health, continuous patient monitoring, and care coordination, and support for eight new languages. In other news, Google’s Verily and its chronic care subsidiary Onduo extended collaboration with Highmark Health. Google Health also announced that its Google Fit mobile app will use machine learning to give users heart rate and respiratory rate readings without the need for any hardware other than their smartphone’s camera.
- On the COVID-19 front, everyone has marvelled at the speed with which Israel’s vaccination has proceeded however it is now revealed that it had entered into a data-for-vaccine deal with Pfizer, promising access to vaccination data at the most granular level to Pfizer in return of a swift vaccine rollout.
- Pharma also continued its digital health partnerships-Astra Zeneca will work with AliveCor for kidney focused collaboration to use AliveCor’s Kardia-K AI, which is designed to measure a patient’s potassium levels.
- India announced its latest budget and there were several expectations from the healthcare industry. The budgetary allocation for sanitation and access to clean water was counted as healthcare expenditure for the first time. There was welcome investments in disease control as well as the COVID-19 vaccination drives however experts were disappointed especially with the low allocation towards the government’s flagship insurance scheme Ayushman Bharat amongst several other things. National Digital Health Mission, an ambitious program to digitization of India’s healthcare ecosystem, received less than 0.01% allocation for the pilots that will be run in the next financial year.
- Finally, cybersecurity continues to be a red flag for digital health; according to a recent investigation, thirty mobile health apps were susceptible to a broken object level authorization (BOLA) attack.
Tidbits
- Healthcare Chart of the Month: There is a lot going on in this chart, that show change in annual revenue for largest healthcare companies/systems in the US. As noted by GIST, the largest companies listed here, including Walmart, Amazon, CVS, and UnitedHealth Group, continue to double down on vertical integration strategies, configuring an array of healthcare assets into platform businesses focused on delivering value to consumers.
- Healthcare Tweet of the Month: This twitter thread describes a recent paper, pointing out listening to patients can go beyond softer feel-goody factors to actual impact on clinical practice; it is just fantastic.