This is a portion of my talk on technology and palliative care. I presented along with two great physicians, Drs. John Boll and Alex Nesbitt.
Let’s first consider the
development of both human expertise and technology. If we posit the notion that
the ability of medical personnel has improved over time to get closer to some
notion of perfection we can also then consider that technology, in a general
sense, has improved also. In many cases the technology has helped the doctors,
in some cases advances in understanding have helped doctors. However, in the
end doctors do not achieve a level of perfection, and technology will never be
able to completely replace doctors, which is something we will consider later.
When we consider medical
care for resource poor areas we have three approaches: more healthcare
providers, technology assisted care, and autonomous care. I want to explore the
notions of technology assisted and autonomous care. We can get defensive about
autonomous medical care but this inquiry relates to places that don’t enjoy the
medical infrastructure that we do.
For example, with a Mars
mission, we may have to rely on a lot of technology and minimally trained
personal. In Boston, you trip over doctors. We naturally resist technology’s
incursion into medical care because people recognize the empathy and
sacrificial attributes brought by real people. However, technology can, at the
least, assist doctors in diagnoses and the development of a patient care plan.
One element of
technology I would like to talk about is artificial intelligence, which mimics
human knowledge and cognition. It can use rules of thumb and learning
capabilities to make recommendations. It can also include sensing technologies
that mimic human abilities from seeing to speaking.
It uses rules of thumb,
called heuristics, and confidence factors to come up with a conclusion. That is
what these IF/THEN and CF statements represent. But machines can’t handle
problems that haven’t been thought of, they can’t think of “what if”.
Some appealing prospects
for resource poor areas are to use an artificial intelligence system to suggest
a diagnosis and treatment using a dialogue with the health care provider or family
member, which I will call the “small black box”. A grander approach is to use a
system of AI that learns about the patient over time, often called deep
learning, as well as other inputs. I will call this the “big black box”.
The small black box uses
patient signs and symptoms and rule-based/heuristic analysis to generate a
suggested patient treatment plan.
However, let’s consider what
could occur beyond using rules of thumb for diagnosis. Neural networks allow
software to learn by identifying patterns from a mass of data. They have been
used in the financial world for a long time but how far can they extend into
medicine? A recent article in Nature illustrates the potential for these deep
learning machines to go beyond heuristics.
The researchers considered
melanoma diagnoses. This diagnosis is typically guided by rules described by
the mnemonic: ABCD. Where they consider asymmetry, borders, color and
diameter of the lesions.
Researchers went beyond rule-based
diagnoses by using 14,000 images previously diagnosed by dermatologists. Could
an AI system categorize the images as benign lesions, cancerous lesions and
non-cancerous growths? The AI system was accurate 72% of the time.
This work was followed up by
a test set of 2,000 biopsy proven images. These were fed into the neural
network. They compared the computer findings with dermatologists’ conclusions.
This graph shows the
performance of the neural network system versus dermatologists. You can see the
wide range of assessments by the dermatologists but generally the AI algorithm
was superior.
However, I recognize this
might be considered a simple, visual evaluation far different then the
variegated nature of pain and palliative care.
I would like to
introduce an autonomous diagnosis and treatment system that is a futuristic concept
but offers potential in resource poor areas. This system would use the small
black box AI system with its heuristics and patient inquiries along with non-invasive
measurements. In this big black box, the system receives patient signs and
symptoms every year and learns about the individual patient. By this process,
the black box learns about the peculiarities of the patient and can recommend
additional required procedures or treatment plans.
Noninvasive evaluations would obtain data from head to toe, perhaps including brain imaging
along the way. What might this look like? Perhaps it is like a space suit or a
gelatinous bath. Maybe you would walk around like Darth Vader for an hour, I
don’t know.
However, there is
something missing. Namely the patient’s distinctive community and beliefs. Hard
data can only go so far. Cultural norms and spiritual needs must be part of the
mix. Designers use ethnography to gain this kind of nuanced insight. Ethnography
usually relies on observations and interviews to develop a textured
understanding of people. This information can also be part of the black box. In
this way, the black box can understand the presentation of pain, the importance
of dignity, and the elements of faith traditions that might not be otherwise
considered in the ‘hard science’ aspect of artificial intelligence.
Ethnographic data is
usually obtained by observation and surveys. We are all ethnographers of sorts,
we quickly learn to read people and develop insights. For example, this can be
summarized by Dr. Nesbitt’s ongoing question for his patients, “what do I need
to know about you to so that I might treat you better?”
We all carry our
cultural values and past experiences with us. This can make it difficult to see
things through someone else’s eyes. For example, consider the arrow hidden in
the FedEx logo.
Using AI for patient
care does present problems. It can effectively program in prejudice so that the
ethnographic data that makes sweeping conclusions such as “a certain population
collectively care for the elderly” doesn’t always work. You could well have a
case where an elderly person is ostracized for a variety of reasons that
ethnographic approaches might not capture.
In addition, data from
individuals becomes part of some sort of database that can present a myriad of
privacy issues.
Finally, there is the
impact of a machine replacing a human in any way. It is an affront to our
pride, identity perhaps even dignity.
Therefore we believe
that the output of AI, even a comprehensive “big black box” system needs to be
moderated by a loving caregiver. Someone who has a relationship with the
patient and can deliver automated treatment plans through the affectionate mind
of a caregiver.
This is a
democratization of medical care where a minimally trained but loving caregivers
have the tools to execute a palliative care plan.
Fundamentally we believe
that in resource poor areas, the best person for supervising medical care is a
person who has a caring relationship with the patient. Technology can act as an
adjunct to allow them to do a better job.
People have the
distinctive ability to empathize with a patient. We also recognize and
appreciate the personal sacrifice given by a caregiver. Additionally, people
have the wonderful ability to develop creative approaches that machines will
never obtain.
