AI in healthcare – reflections from regulatory authorities

It’s a busy period, so many activities, so little time, so much to learn.

In June I will take part in a panel on Inclusion, equity and women in pharma and lead one on data analytics, digital islands, and AI.

I’m also working on a paper on social dreaming, running a coaches’ retreat, where we will discuss self-authorisation and taking another training course on constellations, which is a powerful method to help teams and individuals grasp underlying causes that are impacting performance or development.

Many different topics, but all are close to my heart. It’s a busy year.

Today’s blog topics:

– AI in healthcare – thoughts based on texts from EMA and FDA
– Blueprint for success
– Accumulating knowledge when time is finite, and knowledge is infinite
– How to be a celebrity at work

AI in healthcare – thoughts based on texts from FDA and EMA

Not a day goes by without an article on the use of ChatGPT and AI in the workplace. My opinion remains what it was. ChatGPT can be useful to accelerate certain work areas, as a title or abstract generator, but it’s only as good as the user, and I find that ChatGPT generated text has a recognisable “muzak-like” quality to it. Like a pocket calculator or a bicycle, it is a tool that can deliver great results when used appropriately.

The truth is that creativity and thinking cannot be delegated, and that however much we desire that a neutral unbiased “entity” will come along and make our world a better place, that is not set to happen either.

One thing we need to keep in mind: AI is trained on data sets, there are humans behind the algorithms, there are humans behind the selection of data, there are humans, who identify what is right and wrong, what answers are inappropriate and where users need nudging. Where the topic is technical, this may not be a problem, however where decisions about humans are involved, bias becomes a real risk.

I recently asked ChatGPT a question on religion, the response I received ended with “it is important to respect the diversity of beliefs and practices around the world and to approach the topic of religion and spirituality with sensitivity and open-mindedness”. This made me stop and think, I asked for some facts, I received a nudge, which will vary, in line with different societal beliefs from AI to AI around the world. I find this prospect frightening. The opinion wasn’t declared as such, it was just fed into the response I received. An article on bbc.com quotes experts as saying that “ChatGPT and other machine learning models are constantly updating and advancing as their datasets become broader, the technology means being able to catch biases, mistakes, and errors at the cutting edge”. The technology can help us do this, but at the start are the humans who design the AI. AI is not a separate entity, it is not divorced from human error, it is not neutral, unbiased, or infallible. We would do well to remember this.

Curious to get a sense of what regulators say, I reviewed several documents including a recent discussion paper published in May 2023 by the FDA (1), a study written in June 2022 by the European parliamentary research service (EPRS) Panel for the future of science and technology (2), a publication comparing the non-healthcare specific US algorithmic accountability act of 2022 versus the EU artificial intelligence act and others (3).

The papers themselves are worth reading, they are referenced below.

What I took away from my reading, sources are referenced in brackets and text is in italics where directly quoted:

Your AI is only as good as your programmers “Systemic human biases often make their way into AI models, including widespread and rooted bias based on sex and gender, race and ethnicity, age, socioeconomic status, geographic location, and urban or rural contexts” (2) this poses obvious risks to patients where decisions are automated based on AI.
Your AI output is only as good as your dataset: “The most common causes of AI biases in the healthcare sphere are due to biased and imbalanced datasets which may be based on structural bias and discrimination” (2)
AI has great potential over humans where we are faced with large datasets: that need analysing or where modelling is required including in drug discovery, target identification, molecule modelling, or to manage quality for example in process design and quality control (1)
Standards are not harmonised, but there is more guidance than you might think: Regulatory authorities are working on setting standards; however, the field is evolving fast (3)

In conclusion “AI tools, even when accurate and robust, are dependent on how human beings use them in practice and how the results they produce are used” (2), don’t believe everything you read.

Key takeaways: 1) AI is not a silver bullet 2) the field is evolving fast 3) there is more guidance available than you might be aware of

Blueprint for success

I used to believe that “more is more” and “faster is faster”. When faced with additional workload I just worked faster. This can be successful, but it is not sustainable and I don’t recommend it. While a planned approach is obviously the best approach, often a demanding workload and timelines seduce teams into building the plane while flying. Sometimes, it is unavoidable, but generally it is not. I meet people almost every day who despite knowing better get seduced into trying to deliver work although they know intuitively that doing so will push them beyond their limits. So as a reminder: “Success depends on stopping before you start and expending energy wisely”

So, when you are tasked to deliver a project, especially if the deadlines have already passed, here are my recommendations:

Before you go into solution mode. Stop and ask yourself:
- Is the project within my remit, meaning do I have the authority to deliver it?
- Do I have bandwidth with current staffing? If not, what do I need? What can my team do? What would I need to delegate? Do I need additional resources/external support?
- Are the timelines realistic, if not, what needs to happen to make them realistic?
- Are the timelines fixed – for example, tied into an external deliverable such as a regulatory authority submission, or are they negotiable?
Understand who the key supporters of the project are and the political relevance of the project
- Whose vision and goals does the project support, what are they and are they congruent with the organisation?
  - If this is not clear it helps to clarify this first
- Who are the stakeholders?
Revisit timelines, budget, resourcing etc. with your leadership team and propose a plan that is realistic. It is harder to negotiate additional resources once you have started.
Engage external experts judiciously – try to avoid bringing in external teams to work in isolation on aspects of your project. If you do need external support, make sure it is coordinated.

In summary, approach any problem comprehensively, avoid a scattershot approach and expend energy wisely

Key take-ways: 1) For painless success in any project always start with your Vision/North star and go from there. 2) Strive for simplicity 3) Healthy teams deliver better products

Accumulating knowledge when time is finite, and knowledge is infinite

Time is finite, knowledge is infinite. It’s a challenge and it is frustrating. Every day I identify topics I desire to know more about, and I have realised, that here, as everywhere else, careful planning is everything. The beauty of knowledge is of course that the more you know, the more the shape of the world changes, and the more you can change the shape of your world. This will influence your life, your career trajectory, the possibilities you see, the opportunities you are offered and the beliefs you hold that limit and empower you.

Knowledge comes from many sources including experience, other people, coworkers, mentors and coaches, and continued education and training courses. Different types of knowledge stem from different sources. A balance approach is the way to go.

I take the accelerated route to knowledge wherever possible. If there is something I want to learn I hire an expert to teach me, if there are areas I want to improve in, I work with coaches and mentors, if the learning is experiential, I take part in courses and group work and if it’s academic, I study. This last is where I started, but I realised over the years that the fastest route to knowledge is tapping into respected and vetted resources, who can guide me on the journey.

To free up time for things, I want to learn, I delegate tasks outside my areas of interest to professionals.

Key-takeaways: 1) Follow your interests, never stop learning, find the fastest route, use vetted resources and teachers where possible 2) Respect that non-academic knowing and knowledge take time to settle and cannot be rushed.

How to be a celebrity at work

Virtual collaboration is here to stay, indeed, for many of us it was a reality before COVID. Virtual engagement works particularly well for individuals and teams who have existing relationships, and who have met at least once, and for tasks that are “maintenance” activities, or tie into the type of day-to-day activities, that are performed in markets across the world.

However, for anything new, such as transformation projects involving the formation of international cross-functional project teams, nothing beats bringing individuals together. The pandemic reminded us that we are physical beings. Face to face meetings accelerate trust building, make creative collaboration easier, help to build a team spirit. In my experience this increases the speed of travel for complex projects by close to 50%. So you also save money.

When deciding whether a face to face meeting, an offsite meeting or a zoom interaction will provide most value, an article discussed on Nature.com provides some food for thought: it indicates that scientists are more likely to cite research they experienced by attending a presentation, than talks they most likely missed (link). N.B. The article has not yet been peer reviewed, but I find the data worth sharing regardless.

The authors reviewed a conference attendee scheduling app noting where conference attendees had meeting conflicts, versus where they did not. Subsequently, the research team assessed if there was a correlation between talks that were presumably attended and citations in later work by the conference attendee scientists.

The Nature review states: After taking other effects into account, the authors found that, after correcting for various factors, meeting attendees cited liked papers 52% more often when they could see them in person than when they couldn’t. Author Teplitskiy says “That’s pretty sizable”.

Working off the assumption that this observation is transferable to other professionals, and citations in the scientific world, translate to endorsement in the professional world, this information is worth taking into consideration when you are next making a case for a project or want to seed interest in a group of your peers for a change you’d like to accelerate.

In our new virtual world, it will be harder to motivate individuals to attend a face-to-face meeting, however, in some circumstances it’s worth trying, hard.

Key reminder: Technology has its uses, but it’s always worth considering what the meeting is for, what the desired outcomes are and then planning the format accordingly. It is easier to excite people for your projects and to find project champions when you have engaged them face to face.

I hope my blog provides you with some useful insights and, as ever, I look forward to hearing your thoughts. And if you have a challenging project, are working through a team or personal challenge and would like to discuss coaching to help you achieve that next level, please reach out for an informal chat.

Best wishes

Isabelle C. Widmer

Text from Barbican exhibition AI more than human
Exhibit “the Coded Gaze”