March 25, 2025 • 43 mins
In this two part episode, we talk to Dr. Jonathan Gregory, a former NHS cancer surgeon turned healthcare innovator, to explore the intersection of data, digital tools, AI, and patient-centred design in transforming cancer pathways.
With over 20 years in frontline surgery and leadership roles, Jonathan now works. as clinical advisor for Macmillan Cancer Support, and in roles at Imperial College, and NHS innovation programmes to rethink how healthcare is delivered—from AI-powered end-of-treatment communication to understanding the lived experiences of cancer survivors. He also runs his own consultancy Pivotal Health working with the NHS, academia, startups, and third-sector organisations to develop and implement digital, AI, and data-driven tools.
In part one we explore:
- What it's like working as a surgeon
- Redesigning cancer pathways and why the NHS struggles, despite simple solutions being within reach
- Health inequalities in cancer care and how systems can be re-engineered to work for everyone
In part two we explore:
- Where the real power of AI lies—not in replacing doctors but to challenge bias and support better decisions
- AI's role in better patient communication
- A groundbreaking national research trial, which is rapidly becoming the largest of its kind
Jonathan’s insights will challenge how you think about healthcare transformation, showing how human-centred design, behavioural science, and digital innovation can unlock real improvements—if we let them.
If you’re interested in the future of healthcare, cancer treatment, health inequalities, or AI’s role in medicine, this is an episode you won’t want to miss.
Problems Worth Solving is brought to you by Healthia, the collaborative service design consultancy for health, care and public services.
Find out more about our work at healthia.services.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
SPEAKER_01 (00:00):
Hello, this is Problems Worth Solving, the
podcast where we meet peopletransforming health and care
through human-centred design anddigital innovation.
I'm Sam Mentor, ManagingDirector at Healthier, the
(00:21):
collaborative service designconsultancy.
If you enjoy listening, you cansubscribe to this podcast and
the accompanying newsletter athealthier.services.
So today I'm joined by DrJonathan Gregory.
Jonathan is a cancer surgeon bybackground who worked in the NHS
for over 20 years and heldseveral management positions.
(00:44):
He now works to improve patientcare and outcomes using data,
digital and AI via hisconsultancy Pivotal Health.
He's National Clinical Advisorat Macmillan.
He's digital theme lead at anational institute for health
and care research and he's anhonorary research fellow at
Imperial College London.
We've split this episode intotwo parts because we had lots to
(01:06):
discuss.
You're listening to part twowhere we explore Jonathan's work
with Macmillan where he's usingAI to improve patient
communication.
We will also dig into hisgroundbreaking work to
understand the experiences ofcancer survivors in the
community and discuss his visionfor the future.
In part one, previouslypublished, we explored
Jonathan's background, what it'slike working as a surgeon and
(01:28):
practical experiences of healthinequalities.
We also dug into Jonathan's workto improve cancer pathways and
the critical importance ofgetting the right balance
between digital and humansupport.
Part of being a doctor and asurgeon is, of course, your
medical expertise, but it's alsoyour communication and the way
(01:52):
that you can communicate withthat patient around their
condition and reassure them orsupport them and that side of
things.
You've been doing some work withMacmillan looking at
communication and the end oftreatment summary side of
things.
Can you tell me a bit more aboutthat work?
SPEAKER_00 (02:08):
Yes, so it's based in cancer care and actually
across really all care butthere's what are called pillars
of personalised care and theyrelate really to things like
having support and informationthat you can access They're
about having your holisticneeds, your needs greater than
the medical needs beingaddressed.
(02:30):
And they're about, at the end oftreatment, having what's called
a treatment summary that sort oftries to wrap everything
together.
And these sort of pillars ofcare are very big in cancer,
certainly a big priority forMiller.
And so we tried a variety ofthings, but around treatment
summaries, Things had reallyfailed, which probably sounds
(02:51):
terrible to the listeners, youknow, why it's a letter at the
end.
This letter is meant to beslightly different.
It's meant to be a letter thatreally tells the patient where
they were, where they are todayand where they're going.
So it's subtly different from astandard clinic letter, which
makes it slightly hard fordoctors and nurses perhaps to
(03:11):
generate those letters.
Most letters we write are toother medics.
So even if you're meant to bewriting to the patient, it's not
always second nature.
But also time.
If you say to me, John, there'sfive patients here who need
their treatment summary left todoing.
Okay, do you want me to do thator go see the 10 patients who
just need to be diagnosed withcancer now?
(03:33):
The front of the pathway willalways win because that's where
they need.
So this has been stubbornlylearned.
The national average completionrate is about 8%.
Now, the data, I think, can bequestioned, but I think we can
say it's very low.
So I happened to, you know, thatplace I sit in, I'm always
looking for ideas to beg, stealand borrow and bring across.
(03:56):
I was at a conference and I sawsomeone talking about RPA, robot
process automation.
I thought, I think we could usethat to deliver treatment
summaries.
So that started the seed of anidea.
And over time, I've managed tobuild a collaboration that now
consists of two canceralliances, which cover 11
integrated care boards,Macmillan Cancer Support, the
(04:19):
NHS Automation Accelerator andsix acute NHS trusts.
We've started work at the momentto try and build an automated
solution to deliver thesetreatment summaries for
patients.
SPEAKER_01 (04:30):
So this end of treatment letter is a really
important thing.
And essentially what you'resaying is it's not happening
because there are otherpressures on the surgeons to
make more people better.
When it does happen, what doesthat kind of end of treatment
conversation and letter looklike from a patient's
perspective?
SPEAKER_00 (04:48):
I think where it's done well, it is brilliant and
actually is very hard to improveupon.
I've had some meetings withclinical teams and I'm amazed at
what they're delivering.
You know, let's be clear,there's some absolute real
excellence out there.
But at a lot of places, it'seither not happening or Really,
the letter that goes out is verymedicalised.
(05:08):
It still begins with the list ofyour diagnosis in very medical
terms, all about the receptorsthat were wrong and the
chemotherapy drugs you've had.
And in the workshops we've donewith patients, they all say, I
would just like a letter I canlook at that perhaps makes me
feel a bit less like the done-topiece of me, Tucker.
(05:30):
I could look at it withoutperhaps shuddering.
I mean, we know that, you know,again, back to our previous
conversation about people'scognitive bandwidth.
A lot of people, as they'regoing through their initial
cancer diagnosis and treatment,all the letters, all the
booklets, they really get put ina drawer somewhere out of the
way, like the dining room.
And they try very hard not to goback in there because it really
(05:52):
takes you back to that moment.
And we've heard in our workshopsa real desire to have a letter
that almost allows them to closethe chapter.
But they need somewhere to goback to, you know, for two
months' time.
Something's hurting.
What am I meant to do?
I've got a new cough.
Or my daughter's asked me about,could this have a problem for
(06:16):
our grandson?
Where do I go?
So eventually I've got aslightly friendlier letter.
I won't mind going back to that.
And those conversations I don'tthink happen widely at the
moment.
Not that they don't happen atall.
But I think in that, perhapsusing accessible language,
flagging to local resources,we're looking at doing something
(06:40):
that delivers far more in termsof signposting outside of the
hospital to areas of support andresources that can be used to
signpost to things that probablyseem obvious to the clinical
team.
If I've told you, Sam, duringyour surgery and chemotherapy,
stop smoking, lose weight,that's it.
(07:02):
I don't really imagine I need totell you again.
But the reality is you were tooworried or too ill or hadn't got
the cognitive bandwidth to makethose life decisions at that
moment.
But now that treatment'sfinished, it's a teachable
moment again.
And patients have said to us,I'm actually ready now, but no
one tells me to do anything now.
(07:22):
No one's saying lose weight.
I can't remember where I put thestop smoking leaflet.
So some of it is just about, inan automated way, recycling
stuff.
in the hope that it just landson slightly more fertile ground
this time round.
So I think the conversationsthat happen now, and where they
happen are brilliant, but we'retrying to build something that
(07:44):
compensates for where they don'thappen or where the excellence
is predicated on a group ofhighly motivated, highly skilled
individuals.
And of course, sickness or staffturnover can shake that.
And what we want is somethingthat at least delivers a good
enough for everyone every day.
(08:04):
And to put it in context, it's330,000 people diagnosed with
cancer a year.
So that should be approximately330,000 of these treatment
summary letters going out ayear.
So the impact is potentiallyvery large.
Even if only 1% of people thatletter has an impact for, I'll
still take that number as beingmeaningful.
SPEAKER_01 (08:27):
And from a practical perspective, how does this tool
or this system you're working onwork?
SPEAKER_00 (08:32):
So we're currently building it.
So I started off with highideals and all projects.
I'm compromising those as I goalong.
So the aim was to buildsomething where we'd ideally
need next to no humaninvolvement because that's how
we most release clinical staffand remove barriers to adoption
and scaling.
But because we're trying todrive this off data that
(08:55):
clinical teams receive, ofcatching anyway.
That was the plan.
If you're doing all this stuffanyway, can we use those bits of
data to drive letter content?
It transpires we can and can't.
So we're needing to probablybring in a couple of
human-driven prompts to justhelp the system.
(09:15):
But at this stage, we're using1970s-style AI, Sam.
AI that people don't think isAI.
So it's logic.
It's if it says this, then writethat.
So, for example, if the medicalrecords say breast carcinoma,
let's say breast cancer in theletter.
If the document, if the medicalrecords say stage T10, T1, N0,
(09:40):
M0, we'll just put you had avery small cancer in your
breast.
So it's very like that.
Now, that's been donedeliberately because it will
build confidence.
It's human written information.
We have no risks.
So we've got a low barrier totrust.
It does mean our initial productis going to probably be a little
bit clunky.
(10:01):
But once we've got it workingsound and trust built in it,
then we could look at usingperhaps a very constrained large
language model to round it off,to soften the edges.
But so that's a phase two,whereas having seen already when
some trusts have thought thatwhat we're doing up front is,
AI, the antibody response hasbeen extremely large, Sam, and
(10:26):
caused a lot of problems.
So, which reaffirms that I thinkwe're right to start with this
technologically not very sexysolution, but being really
clear, that's to get to EverestBase Camp.
And how are you keeping thehuman in the loop on this?
So, the trigger will be, at thisstage, will be a human will say,
will trigger the bot, thevirtual worker, to generate the
(10:49):
letter.
And it will come back to thathuman to be sort of checked or
signed off.
Over time, as reassurance, wevalidate it.
Then hopefully we can get lessand less human needed.
Because even checking 25 lettersa week is still checking 25
letters a week, isn't it?
It's another job to do.
But I think people, the teams wespeak to see the value in that
(11:12):
they recognize that they can'tpossibly remember every support
center and every piece of usefulinformation resource they flag
to.
So for example, by knowing yourpostcode and your diagnosis, I
can flag the nearest lung cancersupport group to you is here.
And someone else's postcode isthere.
(11:34):
A doctor or nurse filling one ofthese in, they've not got that
on the tip of their tongue.
So we can lean in that wayaround health inequalities is
one of the drivers here, becauseif you're a very activated
patient, being very engagedthrough your clinical course, I
don't think this letter willmake much difference to you,
being honest.
But if you find, you know,health literacy is not so good,
(11:56):
English is a second language,then actually these sorts of
letters, that's a single placeto go that summarises clearly
what I need to do, I would hopeis helpful.
But also it opens the door to ustaking steps towards
translation.
You know, there's lots of peoplewho have good spoken and
listening skills English skills,but their reading of English is
(12:19):
challenging.
So even if we can, using theautomation, identify their
preferred language.
So at the top of the letter, wecan put in their preferred
language, just an explanation ofwhat the letter contains and
lets us know if you need us totranslate this or it's to be
called to access.
That's a start.
(12:40):
It means they're not waiting fora friend or a neighbor or
someone to read this letter forthem.
We can then, of course, moveover time to translate them
fully with a large languagemodel.
But again, the risk of that isslightly higher at this stage.
But we can at least make a step,and that's a step we're not
generally making routinelyacross the NHS today.
SPEAKER_01 (12:59):
And this is work you're doing with Macmillan.
So have you been able to makeuse of their huge network to
come and get this out there?
SPEAKER_00 (13:06):
Yeah, and certainly I think so.
And I think the other thing I'dsay is that Macmillan's is
helping me to try and holdpeople to account on the
project.
It's like a poacher turnedgamekeeper, isn't it?
I think I'm well positioned toboth sympathise with clinical
teams and say, that's nevergoing to work, that's too big a
ask, but equally to channel intoclinical teams to say, you and I
(13:30):
both know that's not true, orsorry, that's not good enough,
you need to come back again.
And I think that Macmillan is ofa trusted person.
brand by a lot of people in thatspace I think help is helping us
keep everyone on board beinghonest I sense everyone is
supportive and everyone wants todo it but you can imagine the
visions people are holding ofcourse subtly different but of
(13:54):
course Macmillan as well withits work previously around you
know information for patientsand carers We can draw on that
resource of how should weexplain this to people?
Well, they've got materialthat's been tested widely with
patients and the public.
So we can draw on that tosupport the project.
(14:14):
And we've got, let's say, we'vegot current patient partners
supporting the project.
And where that's invaluable, andI'm sure a lot of people
recognise that listening tothis, is they often have lower
wants than everyone wouldassume.
So we asked a group ofspecialist nurses what they
thought should be in the letter.
(14:34):
And then we asked a patientworkshop.
The patient workshop wantedabout half the content the
nurses thought they'd want.
And that's really a principle ofvalues-based healthcare, isn't
it?
Actually, if you measure valueat the level of the patient,
make sure you deliver that.
And often that value may well beeasier or smaller than we in the
system imagine it to be.
SPEAKER_01 (14:56):
So you've been taking a user-centred approach
by involving patients and nursesin this process.
Can you tell me a bit more aboutthe process you've gone through
in designing this?
SPEAKER_00 (15:05):
Yes, and shall we say it's still very much
actively in construction.
So the way we've done it to dateis identify the stakeholders and
I've just ground them intosubmissions, Sam.
Lots of one-to-one meetings atthe start, trying to build a
relationship, build a coalition.
and gradually bringing more andmore people in together.
So I suppose going from that,and sorry, I'm making myself
(15:27):
sound like Machiavelli here, buttrying to make individuals feel
bought into the project at aone-to-one level.
And then you bring into a groupand trying to use that both
ways.
Firstly, so people realisethey're part of a bigger whole
and this is hopefully going toscale through the whole NHS.
So positive spin.
It's also...
you can use that to reduce a bitof pressure.
(15:48):
Because if you're the onehospital out of six who's not
done something, no one likes tobe in that place.
So he's trying to use thattension, I think, and hold that.
And then with patients, we'verun three patient workshops to
date.
So we started off surfacingtheir experiences around this,
trying to surface their wants.
(16:09):
What would they like from this?
And then Some of these letters,some parts are generic.
Do we want the letter to beginwith dear Mr.
or Mrs.
So-and-so?
Or do we want first name?
Let's pair it right back.
What's wanted there?
So we've got a generic componentwhere we're exploring with them
and got a meeting next weekexactly doing this.
(16:32):
So how do you want the letter tosign off?
And then we've got more specificcontent.
So if we've got content that'sreally about the lived
experience after prostatecancer, Then we've got some more
smaller focused groups to say,how would you feel about it
being explained this way?
Does that feel right?
I had the pleasure to take oneof the patient information forum
(16:55):
classes, PISS classes a coupleof years ago about plain English
writing for medicalcommunication.
I'm no expert, but it's reallytrying to bring it in there.
National reading age 9 to 11.
If everyone picks the easy read,shall we just make it an easy
read then?
We don't need the hard read.
And I think a lot of this ishopefully bringing other experts
(17:16):
in who know more than me, but meperhaps just holding enough of
these little bits to know whenwe need help from people at
which times to bring theirexpertise in.
SPEAKER_01 (17:31):
So this work has been around the letters that
people receive at the end oftreatment.
And obviously lots of thosepeople have recovered and they
are then out of the system.
You're doing some work tounderstand the experiences of
people who have recovered.
What are you doing and why isthis important?
SPEAKER_00 (17:48):
Yeah, myself and my colleagues at Imperial College
London, we've launched anational research trial to try
and understand the experiencesand outcomes of people who have
recovered from cancer, but it'sstill affected by it, obviously.
And so trying to do that atscale is difficult.
(18:09):
So there's lots we know aboutpatients after their treatment
for cancer.
But most of that relates tosurveys done in hospitals,
studies that were maybe attachedto a drug trial or something
like that.
When you actually look at it, wedon't know very much about
people's lives when they're justgetting on.
And it might shock people toknow, but we know very little
(18:31):
about the services they use totry and minimize any impacts of
cancer or its treatment.
If you had radiotherapy to yourneck, for example, You may have
difficulty swallowing for therest of your life.
You may see a dieticianintermittently.
If that all happens outside ofhospital, that's unseen really
(18:52):
to the NHS, that cost, thatneed.
So what we've done is design asurvey to try and surface this.
But that in itself, Iappreciate, doesn't sound very
exciting.
But what we've tried to do isdesign this in such a way that
it scales quickly and easily.
So at its core, We designed itto be very easy for GP practices
(19:15):
and people to support.
We tried to think, you know,what's the value proposition to
GPs?
So how do we deliver that?
We worked with some patientpartners to design the initial
survey.
So we surfaced what they feltwas important.
And we worked with a couple ofpeople who got deep experience
really of questionnaire surveysafter cancer.
(19:38):
But then we tested the surveywith 50 people from a variety of
places to get feedback, tryingto understand what questions
worked.
And then we ran the study as apilot.
So we piloted it in NorthwestLondon and had 2000 people
participate.
And we then reviewed theiranswers, looked at the questions
that didn't work so well, wherethe answers went right, had more
(19:59):
formal usability testing,brought along eight more fresh
faced patient partners.
to refresh and review everyproposal we made for the new
one.
And then we launched itnationally last week.
And we've got 3,000 participantsin only eight days nationally.
So potentially this is going tobe the largest study of its type
(20:21):
in the world.
And we've got people in thestudy who were diagnosed and
treated in 1970 all the waythrough to diagnosed and treated
in 2023.
Our oldest participant today is94.
Our youngest is 18.
We've got a real mix in thereand every cancer type.
That's one of the things is alot of work today perhaps is
(20:44):
focused on one group.
But by making this available toevery cancer, anyone diagnosed
over the age of 16, you're sortof hoping to try and identify
groups of people who have moreproblems with a particular thing
than another group because thenmaybe we can identify
coalitions.
Maybe actually people who aretreated for kidney cancer And
(21:06):
for a blood cancer, some oftheir lived experience after
treatment is really similar,even though they're very
different diseases.
But perhaps the support servicescould help or marry up there.
Or is there anything we can spotthat seems to particularly
predicate a good outcome or agood feedback?
So it's been really interesting.
And within the survey, we askpeople to provide feedback on
(21:29):
what they think is good aboutthe survey, bad about the
survey.
we encourage free text responsesbecause we're curious to see
whether do free text responsesencourage people to take part?
I think we might imagine they dobecause it's like saying, tell
us, we really want to know.
But maybe the blank space isintimidating to people with
(21:51):
lower literacy skills.
Maybe people do fill the freetext in and ends the survey.
It's like a mic drop.
Told you all I want to know, I'moff.
And I'm not sure We know that.
SPEAKER_01 (22:03):
I think one of the really exciting things about AI,
I mean, it's a small detail, butthe chat GPT and the large
language model is its ability tointerpret thousands of free text
responses in a survey.
It's something that has neverbeen done well before.
And that suddenly opens up awhole new branch of kind of,
it's qual research at scale,which has never existed before.
SPEAKER_00 (22:23):
Absolutely, Sam.
And we've actually, that is oneof our, we've got lots of hidden
aims in this study.
predominantly it's a methodologystudy.
How do you reach people livingtheir lives to find out about
their long-term effects oftreatment for cancer?
But hidden in it are, yeah, howaccurate are large language
models at producingtranslations?
(22:45):
Can large language modelsinterrogate the free text into
categories and buckets?
We're also asking about, doesthe sequencing of questions, how
much does that influencepeople's responses?
So we asked a very I think acomplex question about whether
people are happy or sorry,whether people consent for us to
(23:07):
access their NHS records to lookat their cancer treatment to
hospital.
And you're participating in thisstudy because you received an
SMS message or you saw it onsocial media.
So you can think there might besome scepticism or concern.
So we wanted to test whether ifwe ask that question near the
start, are people a bitapprehensive put off and say no
(23:30):
and is it different if you askat the end when they've built up
maybe some trust through theproject through the server and
so we've got lots of theseinteresting I'd say almost
design pieces within it andcertainly what we're trying to
advocate here is reallyiterative research trial design
it's not we've got a protocol wecan't waver from this we're
(23:52):
going to deliver it we'rechanging and dropping questions
as we go with this study, as wesee what works, what was better.
And there's a common theme thatwill be the same, but we're
quite happy to change it inlight.
And just this morning, I've hada message about a particular
difficulty science habitssurvey, and you look at it and
go, oh yes, I hadn't spottedthat.
(24:15):
So virus small amendment, we canchange that.
So I think it's trying to, as Isay, deliver at scale, cheaply,
representative data.
that can then be used to designservices, to audit services and
to improve care.
When you think, you know, sowe've now got 5,000 responses
(24:36):
and we have had a very smallresearch grant for this.
And yet most studies at thatsort of scale, you'd be at half
a million pounds, probably amillion pounds.
We're a fraction of that becausewe took a commercially available
customer satisfaction platform
SPEAKER_01 (24:55):
And what will you do with the results once you've got
those in?
Is this a long-term study or isthis you're going to do a piece
of research and then share theinsight?
What's next for it?
SPEAKER_00 (25:03):
Yeah, so the study is open until December.
And if anyone's interested, ifthey go to
computationaloncology.netforward slash indigo, it's the
indigo study.
And the aim here is, Sam, is Ithink multiple in that, firstly,
if this proves to be a way toreach people, We want to be able
to give advice that says, ifyou're looking to find people
(25:26):
aged 20 to 40 who have beenaffected by these cancers, we
found Facebook liberated themost returns and participation.
If you're looking for 50 to 70with this, then actually it was
SMS from the GP practice.
That sort of piece to supportothers.
(25:48):
But then looking at the serviceuse, I mean, that data is
actionable almost immediately.
We know what lifestylebehaviours people have wanted to
change.
We've got these other things.
So we'll be looking at trying topresent that at a simple level
to integrated care boards assoon as we're able.
And then they say we'll workthrough the meat of it.
But these numbers are going tobe large.
(26:09):
We've already got a thousandparticipants who have breast
cancer, for example.
So once we can get their NHSEngland data and to say, about
80% of people are agreeing tothat.
So it goes to show if you buildtrust and understanding, people
will share their healthcaredata.
It's just you've got to buildtrust and understanding.
(26:30):
Then suddenly we can go fiveyears down the line, the people
whose life looked like this, isthere any commonality in their
treatment that could be causingthat?
What drug did they receive orwas it oh my gosh, there's a
difference between people inthis part of the country and
this part.
So we'll use it.
Realistically, this is thestart, not the end.
(26:51):
We can use it to drive aninterventional study or some
other thing.
So for example, this couldbecome a treatment where you're
sent a survey and you completeit and the results are
interpreted in real time.
And if you score a certainthing, it could signpost you to
particular resources.
It could say, do you want aphone call from a specialist
(27:12):
nurse?
It could Some hospitals aredeveloping late effects clinics,
but maybe this could be a frontdoor to that and try and get
people to the right place at theright time.
So I think this is just thestarting point, but quite a
large starting point.
SPEAKER_01 (27:27):
How are you distributing the survey?
SPEAKER_00 (27:30):
So the survey at the moment is scaling through
primary care.
So GP practices who want to takepart in the research look at
their patient lists and send outan SMS to people who qualify.
That finishes around Eastertime.
And at that point, it's socialmedia scaling.
So we've got charity partnersand anyone who remotely wants to
(27:52):
share a link to participate.
And we're snowballing thatbecause you can imagine
particularly disease charities,if you push this into your
community, you're getting yourcancer's voice louder in this
data set.
So we're quite hopeful there.
And we're trying a variety ofplatforms.
(28:12):
So Instagram, Facebook, Google,in a variety of places and
monitoring which people comefrom which platforms, when, what
the completion rates, et cetera,and looking to iterate the
content we have on social media,which we've developed with the
public.
But actually, is it working theway we want?
(28:33):
Does it speak to all groupsequally?
You know, there'll be a process,I think, subsequently of sort of
enrichment where We look at thedemographics of our study and
realise who we've not heard loudenough, and then directly going
out and trying to work to engagemore people from that community.
I always say, if we're in aroom, who's not in the room?
(28:54):
Who's not here?
Who's not in this data set?
We know who should be in thedata set.
We know there's, what, 2.2million people living beyond the
diaspora cancer in the UK, andwe've got crude figures.
We need ethnicity, gender, sex,cancer type to look like the
national picture for the dataset to help everyone.
And that's essentially what ourdrive will be.
SPEAKER_01 (29:16):
And what about low digital skills?
Because obviously a lot ofpeople will be older who've had
cancer.
Have you got specific techniquesfor reaching those people?
SPEAKER_00 (29:23):
You undoubtedly, that is a weakness of the study
in that we don't offer a hardcopy version.
It is, in terms ofparticipation, people are
allowed help to take part, etc.
So it's been designed and theplatform it's on automatically
works with iPad readers and someof these other things.
(29:44):
So there's sort of, shall wesay, I think it has good
accessibility, but it is farfrom perfect.
For example, we're currentlyworking on translations, but
you're undoubtedly right.
My counter to that isaccessibility.
I know that there are going tobe people who can't take part
because of digital literacy,maybe digital exclusion.
(30:09):
Equally, though, we knowcurrently most people don't take
part in these trials becausethey don't go to hospital very
often or they're not asked whenthey're in hospital.
So I think whilst there'sundoubtedly going to be a bias
in our data set, my sense is itshould be smaller than in other
studies and we actually askquestions about people's
(30:31):
educational attainment andthings because we know again
digital confidence often relatesto academic attainment levels
for example so I think we'reunder no illusions that there
are groups of people who willfind it harder to take part in
this but I suppose we at leastare trying.
SPEAKER_01 (30:48):
And worth pointing out that you're doing this on a
small research grant.
You're not NHS England going outand surveying the population.
And you've got huge ambitionsfor this piece of work.
And it sounds like you're makinggreat progress with it.
SPEAKER_00 (30:59):
Yes.
Thank you.
Yes.
To date, I think we've spentabout£50,000 to date.
And that's probably slightly thedisruptive element of me and my
colleagues, Matt Williams andCurlander Calvez.
I think we quite like if peoplesay you can't do it.
I think we'd quite like to tryand show that we can.
SPEAKER_01 (31:16):
Yeah, get stuff done.
It's a good attitude.
I'd like to talk a little bitabout the future.
So this is the closing sectionof the podcast where we go on to
what you're thinking about andwhere you're going.
What are you excited about,Jonathan?
SPEAKER_00 (31:29):
I'm excited about lots, Sam.
It doesn't take much to get meexcited, I have to be honest.
I am excited about AI cominginto healthcare, but I think for
reasons different than mostpeople.
I often give a talk where We'retalking about bias in AI.
And I say, well, I'm the mostbiased thing in the clinic room.
(31:49):
Because as a doctor, mostdoctors are good because they've
become highly biased algorithms.
I'm looking as you walk in.
What's your gait?
Who's with you?
Got a walking stick?
What's the heels of your shoes?
Have you got nicotine stainingon your fingers?
They're all feeding into myalgorithm of what might be wrong
with you.
And that, to large amount of thetime, works really well.
(32:11):
It's why the medical studentgoes, Gosh, how did the
consultant know it was thatwithin a minute?
Because it's all the unspokenstuff.
But I think we're using AI tochallenge our bias.
Because whilst my bias workswell in certain patient groups,
it doesn't work all the time.
Does it work as well at 4 a.m.
(32:34):
for a man of black ethnicityaged 75 as it does for a
21-year-old white female?
Maybe it does, but it probablydoesn't.
And I'm not suggesting that's...
We're talking about bias at thelevel here of deliberative.
It's really, we're talking aboutour heuristics here, our recall
(32:54):
bias, what's our most recentepisodes and recollections.
And so I think having AI canmaybe point out, you think this
person's having a stroke, butgiven their age and this, that,
and the other, the most likelything is this.
It's not that we need AI to tellus what it is.
It's just AI to go, justchecking.
(33:14):
Do you think it's like the nicecolleague who's just going, oh,
it's interesting.
I've seen one of those, but itturned out to be this.
And I think that narrative isfirstly less challenging to
clinicians, but I think it putsAI is biased because the data
that goes into it and guesswhere that came from, humans.
And so it's more about admittingthat.
(33:36):
that it's impossible really tohave anything that's not biased.
But you just need that bias tonot overlap over the same
domains.
And then we can challenge eachother's bias.
So that really interests me indealing with decisions under
uncertainty.
In medicine, you go through aphase where when you're a young
(33:57):
doctor, you think, I'll be greatwhen I know everything.
Can't wait.
Then you're this sort ofregistrar doctor and you think
you do know everything becauseyou're currently sitting in
loads of exams, professionalexams, so you're quite bright.
Then you become exultant and youfeel like you know nothing
again.
But it's because you're facedoften with the edge cases, with
the nuance where there's trueequipoise.
(34:17):
And using AI to look at futurepossibilities and map those, we
can sit down and say, should westart your chemotherapy today or
hold off?
And at the minute, that's builton maybe a research paper, my
understanding of that paper,what I'm remembering today and
talking to you.
(34:37):
But what would be nice is to go,actually, Sam, we can look at
the AI goes, if we start today,this and this and that, this
isn't saying definitely willhappen, but it just structures
our conversation.
You say, well, actually, John,for me right now, most important
thing is I stay at work.
And option B looks like for ahundred Sam's, more would stay
at work doing that.
(34:58):
And that is true informeddecision-making, I think.
So those sorts of changes, Ithink, Sam, to me feel like real
paradigm shifts.
I don't think they're comingimmediately, but I think they'll
come.
We know from studies presently,at least, that patients all want
AI to be, shall we say,subservient to the doctor.
I think it'll be one of mycolleagues, Danny Rutter, said,
(35:20):
gave the analogy of lifts and Ithink you go back to whatever
the 1920s, there was always abellboy or a Lyft operator
because we didn't trust theLyft.
And then now you'd think it wasmadness if you saw an operator
in the Lyft buttons.
And I think he's probably rightthat maybe for the next 20, 30
years, there's a doctor in frontof this.
And then gradually everyonegoes, I don't know why I'm
(35:43):
waiting for it.
He just tells me what's on thecomputer anyway.
And then using it more in theback office to get rid of low
value, high volume, repetitivetasks like doctors checking
blood test results that arenormal and having to do a letter
to the GP to say it was normal.
That could all be vanishing out,you know, via an AI, couldn't
it?
So are you optimistic?
(36:06):
Good question.
I'm a pessimistic optimist.
I think everything will happenin the end.
Do I think it'll be easy?
No.
To put it in context, thistreatment summaries project, we
might be talking aboutindividual NHS trusts needing to
pay two to five thousand poundsa year for the platform beyond
the pilot.
(36:26):
And a lot of them have said, no,we can't afford that.
So that's the sort of barrier touse and implementation.
That's obviously concerning.
I think being devil's advocate,I think we've got a big forcing
function coming.
because the NHS, as it is, isnot going to be able to cope
with the scale of demand.
And so in a funny way, I thinkautomation will, a bit like
(36:49):
through COVID, will be pushedrapidly on us.
And so I think it'll feel likenothing's happening and then
suddenly, like dominoes, it willgo.
I think I am hopeful.
I'm hopeful as well that I dosit in far more conversations
now where social determinants ofhealth inequality are actually
considered a thought about Notenough and not adequately, but
(37:13):
when I compare that to, say,five years ago, I do feel that
it is different.
So I think I'm hopeful.
SPEAKER_01 (37:20):
If you could wave a magic wand and change one thing
about the way we treat cancertoday, what would you change?
SPEAKER_00 (37:29):
Oh, you've stumped me, Sam.
SPEAKER_01 (37:31):
Or it could be change about the health system.
SPEAKER_00 (37:33):
Yeah, yeah.
You've silenced the surgeon,Sam.
Amazing.
You need to keep this in thepodcast.
The world will be like, so muchfor him being an innovator.
Is that because
SPEAKER_01 (37:47):
there's so much that you would change?
Or is that because it's soperfect?
SPEAKER_00 (37:52):
I think it's because I'm searching for, if I've got
one wish, it's got to be themost impactful.
And that's probably, I'm thenthinking, oh, what do you know
what I mean?
That like, oh, what does thatlook like?
I think, and this perhapssounds, well, it's not
innovative.
I think if every person goingthrough cancer treatment didn't
have to worry about theirfinances, I think there's
(38:17):
something that would just make abig difference for a large
number of people and actuallydoesn't rely on technologists or
other things.
But to be worried about puttingthe eating on when you're on
chemotherapy, to be thinking howare we going to get stuff for
the kids, just that to me on topof everything else I think is
(38:37):
That is something that wecouldn't, society could choose
to alleviate.
SPEAKER_01 (38:41):
So it's a policy thing.
It's not a medical thing.
It's a policy thing.
SPEAKER_00 (38:44):
Yeah,
SPEAKER_01 (38:44):
I think so.
What's next for you, Jonathan?
Where are you going this year?
SPEAKER_00 (38:48):
So hopefully finish scaling the trial, delivering
the end of treatment summariesproject.
And then really, I think I'dlike to be, I'm thinking about
doing a piece of work aroundracism in healthcare.
And I've had the, recently readthe book, Divided.
by Annabelle Suemio and that'sopened my eyes to how baked some
(39:11):
of that is into the wayhealthcare is delivered and I'd
be interested in perhaps tryingto bring some of that into my
conversations and my work toperhaps challenge looking around
and challenging and findingwhere we're I think nowadays I
would say accidentallyperpetuating what has happened
certainly when I readAnnabelle's book I was
completely unaware of most ofthis so I think it's about
(39:33):
surfacing it and then trying towork to address that.
So I think that will be afeature for the next year.
But then say my work with theHealth Tech Research Center with
Macmillan and my own consultancyis really about working with
companies to develop productsthat meet an unmet need, that
(39:57):
are safe, that work, that canprove they work and have a value
proposition that means they canbe procured be adopted and
scaled in the NHS.
And I think if that's what I'mtrying to do, there's probably
going to be enough for me to dothis year, Sam.
Not a bit worried I might haverun out of things to do next
year, but we'll cross thatbridge when we come to it.
SPEAKER_01 (40:17):
A year's quite a long time really, isn't it?
Trying to think what you'regoing to be doing in 12 months'
time, let alone five years.
SPEAKER_00 (40:22):
I suppose the only thing that I had noted down
about the future, but I don'tthink it's probably worth the
bit, but it ties in, I think,with what you would be doing
It's trying to bring behavioralscience and design and human
psychology much more into theforeground of healthcare.
(40:43):
It feels to me a bit likeoutside of public health, where
we think about public healthmessaging.
I'm not sure we really apply itproperly to everything.
I don't think we use it whencompanies or people are
designing things for cliniciansor nurses to use.
I don't think we use it when wecommute to work.
(41:03):
public or with patients andsystem design.
So I'm always looking foropportunities where try and
surface that into the foregroundof the way services are designed
and delivered and rather lessthan perhaps being either an
afterthought or not thought ofat all.
SPEAKER_01 (41:23):
It still feels quite novel.
We work with lots of differentdepartments and often the work
also involves a communicatingthe why you're doing what you're
doing as well as doing whatyou're doing which is that's a
big overhead but it's aneducation piece as well
SPEAKER_00 (41:38):
yeah I agree as I've said I don't I honestly don't
believe I'm an expert inanything I just read a bit of
stuff and listen to people andtry and yet some of this I think
I think I'm the only personwho's in this room who's
mentioning behavioural scienceor whatever you think we should
be beyond that now Kahneman'swork was what the 1970s And I
(42:00):
think because it got listedunder behavioral economics,
everyone thinks it's onlyeconomics.
It's about decision-making.
And as we actually didn'tmention in the talk, the
scarcity book from SundellMillarison, that's the same.
It's decision-making underuncertainty and cognitive load,
isn't it?
And I think that should all bemandated into healthcare's
(42:21):
awareness if we just followedthose principles and some basic
human psychology.
I think we'd get an uplift onour services that would be far
greater than the amount ofeffort that went in.
SPEAKER_01 (42:34):
Jonathan, it's been a pleasure talking to you.
Thanks so much for taking thetime to join me today.
SPEAKER_00 (42:38):
Thank you very much for having me, Sam.
SPEAKER_01 (42:51):
Problems Worth Solving is brought to you by
Healthier, the collaborativeservice design consultancy for
transformation in health careand public services.
Find out more about how we canhelp you deliver user-centered
change at healthier.services.
Hello, this is Problems Worth Solving, the
podcast where we meet peopletransforming health and care
through human-centred design anddigital innovation.
I'm Sam Mentor, ManagingDirector at Healthier, the
(00:21):
collaborative service designconsultancy.
If you enjoy listening, you cansubscribe to this podcast and
the accompanying newsletter athealthier.services.
So today I'm joined by DrJonathan Gregory.
Jonathan is a cancer surgeon bybackground who worked in the NHS
for over 20 years and heldseveral management positions.
(00:44):
He now works to improve patientcare and outcomes using data,
digital and AI via hisconsultancy Pivotal Health.
He's National Clinical Advisorat Macmillan.
He's digital theme lead at anational institute for health
and care research and he's anhonorary research fellow at
Imperial College London.
We've split this episode intotwo parts because we had lots to
(01:06):
discuss.
You're listening to part twowhere we explore Jonathan's work
with Macmillan where he's usingAI to improve patient
communication.
We will also dig into hisgroundbreaking work to
understand the experiences ofcancer survivors in the
community and discuss his visionfor the future.
In part one, previouslypublished, we explored
Jonathan's background, what it'slike working as a surgeon and
(01:28):
practical experiences of healthinequalities.
We also dug into Jonathan's workto improve cancer pathways and
the critical importance ofgetting the right balance
between digital and humansupport.
Part of being a doctor and asurgeon is, of course, your
medical expertise, but it's alsoyour communication and the way
(01:52):
that you can communicate withthat patient around their
condition and reassure them orsupport them and that side of
things.
You've been doing some work withMacmillan looking at
communication and the end oftreatment summary side of
things.
Can you tell me a bit more aboutthat work?
SPEAKER_00 (02:08):
Yes, so it's based in cancer care and actually
across really all care butthere's what are called pillars
of personalised care and theyrelate really to things like
having support and informationthat you can access They're
about having your holisticneeds, your needs greater than
the medical needs beingaddressed.
(02:30):
And they're about, at the end oftreatment, having what's called
a treatment summary that sort oftries to wrap everything
together.
And these sort of pillars ofcare are very big in cancer,
certainly a big priority forMiller.
And so we tried a variety ofthings, but around treatment
summaries, Things had reallyfailed, which probably sounds
(02:51):
terrible to the listeners, youknow, why it's a letter at the
end.
This letter is meant to beslightly different.
It's meant to be a letter thatreally tells the patient where
they were, where they are todayand where they're going.
So it's subtly different from astandard clinic letter, which
makes it slightly hard fordoctors and nurses perhaps to
(03:11):
generate those letters.
Most letters we write are toother medics.
So even if you're meant to bewriting to the patient, it's not
always second nature.
But also time.
If you say to me, John, there'sfive patients here who need
their treatment summary left todoing.
Okay, do you want me to do thator go see the 10 patients who
just need to be diagnosed withcancer now?
(03:33):
The front of the pathway willalways win because that's where
they need.
So this has been stubbornlylearned.
The national average completionrate is about 8%.
Now, the data, I think, can bequestioned, but I think we can
say it's very low.
So I happened to, you know, thatplace I sit in, I'm always
looking for ideas to beg, stealand borrow and bring across.
(03:56):
I was at a conference and I sawsomeone talking about RPA, robot
process automation.
I thought, I think we could usethat to deliver treatment
summaries.
So that started the seed of anidea.
And over time, I've managed tobuild a collaboration that now
consists of two canceralliances, which cover 11
integrated care boards,Macmillan Cancer Support, the
(04:19):
NHS Automation Accelerator andsix acute NHS trusts.
We've started work at the momentto try and build an automated
solution to deliver thesetreatment summaries for
patients.
SPEAKER_01 (04:30):
So this end of treatment letter is a really
important thing.
And essentially what you'resaying is it's not happening
because there are otherpressures on the surgeons to
make more people better.
When it does happen, what doesthat kind of end of treatment
conversation and letter looklike from a patient's
perspective?
SPEAKER_00 (04:48):
I think where it's done well, it is brilliant and
actually is very hard to improveupon.
I've had some meetings withclinical teams and I'm amazed at
what they're delivering.
You know, let's be clear,there's some absolute real
excellence out there.
But at a lot of places, it'seither not happening or Really,
the letter that goes out is verymedicalised.
(05:08):
It still begins with the list ofyour diagnosis in very medical
terms, all about the receptorsthat were wrong and the
chemotherapy drugs you've had.
And in the workshops we've donewith patients, they all say, I
would just like a letter I canlook at that perhaps makes me
feel a bit less like the done-topiece of me, Tucker.
(05:30):
I could look at it withoutperhaps shuddering.
I mean, we know that, you know,again, back to our previous
conversation about people'scognitive bandwidth.
A lot of people, as they'regoing through their initial
cancer diagnosis and treatment,all the letters, all the
booklets, they really get put ina drawer somewhere out of the
way, like the dining room.
And they try very hard not to goback in there because it really
(05:52):
takes you back to that moment.
And we've heard in our workshopsa real desire to have a letter
that almost allows them to closethe chapter.
But they need somewhere to goback to, you know, for two
months' time.
Something's hurting.
What am I meant to do?
I've got a new cough.
Or my daughter's asked me about,could this have a problem for
(06:16):
our grandson?
Where do I go?
So eventually I've got aslightly friendlier letter.
I won't mind going back to that.
And those conversations I don'tthink happen widely at the
moment.
Not that they don't happen atall.
But I think in that, perhapsusing accessible language,
flagging to local resources,we're looking at doing something
(06:40):
that delivers far more in termsof signposting outside of the
hospital to areas of support andresources that can be used to
signpost to things that probablyseem obvious to the clinical
team.
If I've told you, Sam, duringyour surgery and chemotherapy,
stop smoking, lose weight,that's it.
(07:02):
I don't really imagine I need totell you again.
But the reality is you were tooworried or too ill or hadn't got
the cognitive bandwidth to makethose life decisions at that
moment.
But now that treatment'sfinished, it's a teachable
moment again.
And patients have said to us,I'm actually ready now, but no
one tells me to do anything now.
(07:22):
No one's saying lose weight.
I can't remember where I put thestop smoking leaflet.
So some of it is just about, inan automated way, recycling
stuff.
in the hope that it just landson slightly more fertile ground
this time round.
So I think the conversationsthat happen now, and where they
happen are brilliant, but we'retrying to build something that
(07:44):
compensates for where they don'thappen or where the excellence
is predicated on a group ofhighly motivated, highly skilled
individuals.
And of course, sickness or staffturnover can shake that.
And what we want is somethingthat at least delivers a good
enough for everyone every day.
(08:04):
And to put it in context, it's330,000 people diagnosed with
cancer a year.
So that should be approximately330,000 of these treatment
summary letters going out ayear.
So the impact is potentiallyvery large.
Even if only 1% of people thatletter has an impact for, I'll
still take that number as beingmeaningful.
SPEAKER_01 (08:27):
And from a practical perspective, how does this tool
or this system you're working onwork?
SPEAKER_00 (08:32):
So we're currently building it.
So I started off with highideals and all projects.
I'm compromising those as I goalong.
So the aim was to buildsomething where we'd ideally
need next to no humaninvolvement because that's how
we most release clinical staffand remove barriers to adoption
and scaling.
But because we're trying todrive this off data that
(08:55):
clinical teams receive, ofcatching anyway.
That was the plan.
If you're doing all this stuffanyway, can we use those bits of
data to drive letter content?
It transpires we can and can't.
So we're needing to probablybring in a couple of
human-driven prompts to justhelp the system.
(09:15):
But at this stage, we're using1970s-style AI, Sam.
AI that people don't think isAI.
So it's logic.
It's if it says this, then writethat.
So, for example, if the medicalrecords say breast carcinoma,
let's say breast cancer in theletter.
If the document, if the medicalrecords say stage T10, T1, N0,
(09:40):
M0, we'll just put you had avery small cancer in your
breast.
So it's very like that.
Now, that's been donedeliberately because it will
build confidence.
It's human written information.
We have no risks.
So we've got a low barrier totrust.
It does mean our initial productis going to probably be a little
bit clunky.
(10:01):
But once we've got it workingsound and trust built in it,
then we could look at usingperhaps a very constrained large
language model to round it off,to soften the edges.
But so that's a phase two,whereas having seen already when
some trusts have thought thatwhat we're doing up front is,
AI, the antibody response hasbeen extremely large, Sam, and
(10:26):
caused a lot of problems.
So, which reaffirms that I thinkwe're right to start with this
technologically not very sexysolution, but being really
clear, that's to get to EverestBase Camp.
And how are you keeping thehuman in the loop on this?
So, the trigger will be, at thisstage, will be a human will say,
will trigger the bot, thevirtual worker, to generate the
(10:49):
letter.
And it will come back to thathuman to be sort of checked or
signed off.
Over time, as reassurance, wevalidate it.
Then hopefully we can get lessand less human needed.
Because even checking 25 lettersa week is still checking 25
letters a week, isn't it?
It's another job to do.
But I think people, the teams wespeak to see the value in that
(11:12):
they recognize that they can'tpossibly remember every support
center and every piece of usefulinformation resource they flag
to.
So for example, by knowing yourpostcode and your diagnosis, I
can flag the nearest lung cancersupport group to you is here.
And someone else's postcode isthere.
(11:34):
A doctor or nurse filling one ofthese in, they've not got that
on the tip of their tongue.
So we can lean in that wayaround health inequalities is
one of the drivers here, becauseif you're a very activated
patient, being very engagedthrough your clinical course, I
don't think this letter willmake much difference to you,
being honest.
But if you find, you know,health literacy is not so good,
(11:56):
English is a second language,then actually these sorts of
letters, that's a single placeto go that summarises clearly
what I need to do, I would hopeis helpful.
But also it opens the door to ustaking steps towards
translation.
You know, there's lots of peoplewho have good spoken and
listening skills English skills,but their reading of English is
(12:19):
challenging.
So even if we can, using theautomation, identify their
preferred language.
So at the top of the letter, wecan put in their preferred
language, just an explanation ofwhat the letter contains and
lets us know if you need us totranslate this or it's to be
called to access.
That's a start.
(12:40):
It means they're not waiting fora friend or a neighbor or
someone to read this letter forthem.
We can then, of course, moveover time to translate them
fully with a large languagemodel.
But again, the risk of that isslightly higher at this stage.
But we can at least make a step,and that's a step we're not
generally making routinelyacross the NHS today.
SPEAKER_01 (12:59):
And this is work you're doing with Macmillan.
So have you been able to makeuse of their huge network to
come and get this out there?
SPEAKER_00 (13:06):
Yeah, and certainly I think so.
And I think the other thing I'dsay is that Macmillan's is
helping me to try and holdpeople to account on the
project.
It's like a poacher turnedgamekeeper, isn't it?
I think I'm well positioned toboth sympathise with clinical
teams and say, that's nevergoing to work, that's too big a
ask, but equally to channel intoclinical teams to say, you and I
(13:30):
both know that's not true, orsorry, that's not good enough,
you need to come back again.
And I think that Macmillan is ofa trusted person.
brand by a lot of people in thatspace I think help is helping us
keep everyone on board beinghonest I sense everyone is
supportive and everyone wants todo it but you can imagine the
visions people are holding ofcourse subtly different but of
(13:54):
course Macmillan as well withits work previously around you
know information for patientsand carers We can draw on that
resource of how should weexplain this to people?
Well, they've got materialthat's been tested widely with
patients and the public.
So we can draw on that tosupport the project.
(14:14):
And we've got, let's say, we'vegot current patient partners
supporting the project.
And where that's invaluable, andI'm sure a lot of people
recognise that listening tothis, is they often have lower
wants than everyone wouldassume.
So we asked a group ofspecialist nurses what they
thought should be in the letter.
(14:34):
And then we asked a patientworkshop.
The patient workshop wantedabout half the content the
nurses thought they'd want.
And that's really a principle ofvalues-based healthcare, isn't
it?
Actually, if you measure valueat the level of the patient,
make sure you deliver that.
And often that value may well beeasier or smaller than we in the
system imagine it to be.
SPEAKER_01 (14:56):
So you've been taking a user-centred approach
by involving patients and nursesin this process.
Can you tell me a bit more aboutthe process you've gone through
in designing this?
SPEAKER_00 (15:05):
Yes, and shall we say it's still very much
actively in construction.
So the way we've done it to dateis identify the stakeholders and
I've just ground them intosubmissions, Sam.
Lots of one-to-one meetings atthe start, trying to build a
relationship, build a coalition.
and gradually bringing more andmore people in together.
So I suppose going from that,and sorry, I'm making myself
(15:27):
sound like Machiavelli here, buttrying to make individuals feel
bought into the project at aone-to-one level.
And then you bring into a groupand trying to use that both
ways.
Firstly, so people realisethey're part of a bigger whole
and this is hopefully going toscale through the whole NHS.
So positive spin.
It's also...
you can use that to reduce a bitof pressure.
(15:48):
Because if you're the onehospital out of six who's not
done something, no one likes tobe in that place.
So he's trying to use thattension, I think, and hold that.
And then with patients, we'verun three patient workshops to
date.
So we started off surfacingtheir experiences around this,
trying to surface their wants.
(16:09):
What would they like from this?
And then Some of these letters,some parts are generic.
Do we want the letter to beginwith dear Mr.
or Mrs.
So-and-so?
Or do we want first name?
Let's pair it right back.
What's wanted there?
So we've got a generic componentwhere we're exploring with them
and got a meeting next weekexactly doing this.
(16:32):
So how do you want the letter tosign off?
And then we've got more specificcontent.
So if we've got content that'sreally about the lived
experience after prostatecancer, Then we've got some more
smaller focused groups to say,how would you feel about it
being explained this way?
Does that feel right?
I had the pleasure to take oneof the patient information forum
(16:55):
classes, PISS classes a coupleof years ago about plain English
writing for medicalcommunication.
I'm no expert, but it's reallytrying to bring it in there.
National reading age 9 to 11.
If everyone picks the easy read,shall we just make it an easy
read then?
We don't need the hard read.
And I think a lot of this ishopefully bringing other experts
(17:16):
in who know more than me, but meperhaps just holding enough of
these little bits to know whenwe need help from people at
which times to bring theirexpertise in.
SPEAKER_01 (17:31):
So this work has been around the letters that
people receive at the end oftreatment.
And obviously lots of thosepeople have recovered and they
are then out of the system.
You're doing some work tounderstand the experiences of
people who have recovered.
What are you doing and why isthis important?
SPEAKER_00 (17:48):
Yeah, myself and my colleagues at Imperial College
London, we've launched anational research trial to try
and understand the experiencesand outcomes of people who have
recovered from cancer, but it'sstill affected by it, obviously.
And so trying to do that atscale is difficult.
(18:09):
So there's lots we know aboutpatients after their treatment
for cancer.
But most of that relates tosurveys done in hospitals,
studies that were maybe attachedto a drug trial or something
like that.
When you actually look at it, wedon't know very much about
people's lives when they're justgetting on.
And it might shock people toknow, but we know very little
(18:31):
about the services they use totry and minimize any impacts of
cancer or its treatment.
If you had radiotherapy to yourneck, for example, You may have
difficulty swallowing for therest of your life.
You may see a dieticianintermittently.
If that all happens outside ofhospital, that's unseen really
(18:52):
to the NHS, that cost, thatneed.
So what we've done is design asurvey to try and surface this.
But that in itself, Iappreciate, doesn't sound very
exciting.
But what we've tried to do isdesign this in such a way that
it scales quickly and easily.
So at its core, We designed itto be very easy for GP practices
(19:15):
and people to support.
We tried to think, you know,what's the value proposition to
GPs?
So how do we deliver that?
We worked with some patientpartners to design the initial
survey.
So we surfaced what they feltwas important.
And we worked with a couple ofpeople who got deep experience
really of questionnaire surveysafter cancer.
(19:38):
But then we tested the surveywith 50 people from a variety of
places to get feedback, tryingto understand what questions
worked.
And then we ran the study as apilot.
So we piloted it in NorthwestLondon and had 2000 people
participate.
And we then reviewed theiranswers, looked at the questions
that didn't work so well, wherethe answers went right, had more
(19:59):
formal usability testing,brought along eight more fresh
faced patient partners.
to refresh and review everyproposal we made for the new
one.
And then we launched itnationally last week.
And we've got 3,000 participantsin only eight days nationally.
So potentially this is going tobe the largest study of its type
(20:21):
in the world.
And we've got people in thestudy who were diagnosed and
treated in 1970 all the waythrough to diagnosed and treated
in 2023.
Our oldest participant today is94.
Our youngest is 18.
We've got a real mix in thereand every cancer type.
That's one of the things is alot of work today perhaps is
(20:44):
focused on one group.
But by making this available toevery cancer, anyone diagnosed
over the age of 16, you're sortof hoping to try and identify
groups of people who have moreproblems with a particular thing
than another group because thenmaybe we can identify
coalitions.
Maybe actually people who aretreated for kidney cancer And
(21:06):
for a blood cancer, some oftheir lived experience after
treatment is really similar,even though they're very
different diseases.
But perhaps the support servicescould help or marry up there.
Or is there anything we can spotthat seems to particularly
predicate a good outcome or agood feedback?
So it's been really interesting.
And within the survey, we askpeople to provide feedback on
(21:29):
what they think is good aboutthe survey, bad about the
survey.
we encourage free text responsesbecause we're curious to see
whether do free text responsesencourage people to take part?
I think we might imagine they dobecause it's like saying, tell
us, we really want to know.
But maybe the blank space isintimidating to people with
(21:51):
lower literacy skills.
Maybe people do fill the freetext in and ends the survey.
It's like a mic drop.
Told you all I want to know, I'moff.
And I'm not sure We know that.
SPEAKER_01 (22:03):
I think one of the really exciting things about AI,
I mean, it's a small detail, butthe chat GPT and the large
language model is its ability tointerpret thousands of free text
responses in a survey.
It's something that has neverbeen done well before.
And that suddenly opens up awhole new branch of kind of,
it's qual research at scale,which has never existed before.
SPEAKER_00 (22:23):
Absolutely, Sam.
And we've actually, that is oneof our, we've got lots of hidden
aims in this study.
predominantly it's a methodologystudy.
How do you reach people livingtheir lives to find out about
their long-term effects oftreatment for cancer?
But hidden in it are, yeah, howaccurate are large language
models at producingtranslations?
(22:45):
Can large language modelsinterrogate the free text into
categories and buckets?
We're also asking about, doesthe sequencing of questions, how
much does that influencepeople's responses?
So we asked a very I think acomplex question about whether
people are happy or sorry,whether people consent for us to
(23:07):
access their NHS records to lookat their cancer treatment to
hospital.
And you're participating in thisstudy because you received an
SMS message or you saw it onsocial media.
So you can think there might besome scepticism or concern.
So we wanted to test whether ifwe ask that question near the
start, are people a bitapprehensive put off and say no
(23:30):
and is it different if you askat the end when they've built up
maybe some trust through theproject through the server and
so we've got lots of theseinteresting I'd say almost
design pieces within it andcertainly what we're trying to
advocate here is reallyiterative research trial design
it's not we've got a protocol wecan't waver from this we're
(23:52):
going to deliver it we'rechanging and dropping questions
as we go with this study, as wesee what works, what was better.
And there's a common theme thatwill be the same, but we're
quite happy to change it inlight.
And just this morning, I've hada message about a particular
difficulty science habitssurvey, and you look at it and
go, oh yes, I hadn't spottedthat.
(24:15):
So virus small amendment, we canchange that.
So I think it's trying to, as Isay, deliver at scale, cheaply,
representative data.
that can then be used to designservices, to audit services and
to improve care.
When you think, you know, sowe've now got 5,000 responses
(24:36):
and we have had a very smallresearch grant for this.
And yet most studies at thatsort of scale, you'd be at half
a million pounds, probably amillion pounds.
We're a fraction of that becausewe took a commercially available
customer satisfaction platform
SPEAKER_01 (24:55):
And what will you do with the results once you've got
those in?
Is this a long-term study or isthis you're going to do a piece
of research and then share theinsight?
What's next for it?
SPEAKER_00 (25:03):
Yeah, so the study is open until December.
And if anyone's interested, ifthey go to
computationaloncology.netforward slash indigo, it's the
indigo study.
And the aim here is, Sam, is Ithink multiple in that, firstly,
if this proves to be a way toreach people, We want to be able
to give advice that says, ifyou're looking to find people
(25:26):
aged 20 to 40 who have beenaffected by these cancers, we
found Facebook liberated themost returns and participation.
If you're looking for 50 to 70with this, then actually it was
SMS from the GP practice.
That sort of piece to supportothers.
(25:48):
But then looking at the serviceuse, I mean, that data is
actionable almost immediately.
We know what lifestylebehaviours people have wanted to
change.
We've got these other things.
So we'll be looking at trying topresent that at a simple level
to integrated care boards assoon as we're able.
And then they say we'll workthrough the meat of it.
But these numbers are going tobe large.
(26:09):
We've already got a thousandparticipants who have breast
cancer, for example.
So once we can get their NHSEngland data and to say, about
80% of people are agreeing tothat.
So it goes to show if you buildtrust and understanding, people
will share their healthcaredata.
It's just you've got to buildtrust and understanding.
(26:30):
Then suddenly we can go fiveyears down the line, the people
whose life looked like this, isthere any commonality in their
treatment that could be causingthat?
What drug did they receive orwas it oh my gosh, there's a
difference between people inthis part of the country and
this part.
So we'll use it.
Realistically, this is thestart, not the end.
(26:51):
We can use it to drive aninterventional study or some
other thing.
So for example, this couldbecome a treatment where you're
sent a survey and you completeit and the results are
interpreted in real time.
And if you score a certainthing, it could signpost you to
particular resources.
It could say, do you want aphone call from a specialist
(27:12):
nurse?
It could Some hospitals aredeveloping late effects clinics,
but maybe this could be a frontdoor to that and try and get
people to the right place at theright time.
So I think this is just thestarting point, but quite a
large starting point.
SPEAKER_01 (27:27):
How are you distributing the survey?
SPEAKER_00 (27:30):
So the survey at the moment is scaling through
primary care.
So GP practices who want to takepart in the research look at
their patient lists and send outan SMS to people who qualify.
That finishes around Eastertime.
And at that point, it's socialmedia scaling.
So we've got charity partnersand anyone who remotely wants to
(27:52):
share a link to participate.
And we're snowballing thatbecause you can imagine
particularly disease charities,if you push this into your
community, you're getting yourcancer's voice louder in this
data set.
So we're quite hopeful there.
And we're trying a variety ofplatforms.
(28:12):
So Instagram, Facebook, Google,in a variety of places and
monitoring which people comefrom which platforms, when, what
the completion rates, et cetera,and looking to iterate the
content we have on social media,which we've developed with the
public.
But actually, is it working theway we want?
(28:33):
Does it speak to all groupsequally?
You know, there'll be a process,I think, subsequently of sort of
enrichment where We look at thedemographics of our study and
realise who we've not heard loudenough, and then directly going
out and trying to work to engagemore people from that community.
I always say, if we're in aroom, who's not in the room?
(28:54):
Who's not here?
Who's not in this data set?
We know who should be in thedata set.
We know there's, what, 2.2million people living beyond the
diaspora cancer in the UK, andwe've got crude figures.
We need ethnicity, gender, sex,cancer type to look like the
national picture for the dataset to help everyone.
And that's essentially what ourdrive will be.
SPEAKER_01 (29:16):
And what about low digital skills?
Because obviously a lot ofpeople will be older who've had
cancer.
Have you got specific techniquesfor reaching those people?
SPEAKER_00 (29:23):
You undoubtedly, that is a weakness of the study
in that we don't offer a hardcopy version.
It is, in terms ofparticipation, people are
allowed help to take part, etc.
So it's been designed and theplatform it's on automatically
works with iPad readers and someof these other things.
(29:44):
So there's sort of, shall wesay, I think it has good
accessibility, but it is farfrom perfect.
For example, we're currentlyworking on translations, but
you're undoubtedly right.
My counter to that isaccessibility.
I know that there are going tobe people who can't take part
because of digital literacy,maybe digital exclusion.
(30:09):
Equally, though, we knowcurrently most people don't take
part in these trials becausethey don't go to hospital very
often or they're not asked whenthey're in hospital.
So I think whilst there'sundoubtedly going to be a bias
in our data set, my sense is itshould be smaller than in other
studies and we actually askquestions about people's
(30:31):
educational attainment andthings because we know again
digital confidence often relatesto academic attainment levels
for example so I think we'reunder no illusions that there
are groups of people who willfind it harder to take part in
this but I suppose we at leastare trying.
SPEAKER_01 (30:48):
And worth pointing out that you're doing this on a
small research grant.
You're not NHS England going outand surveying the population.
And you've got huge ambitionsfor this piece of work.
And it sounds like you're makinggreat progress with it.
SPEAKER_00 (30:59):
Yes.
Thank you.
Yes.
To date, I think we've spentabout£50,000 to date.
And that's probably slightly thedisruptive element of me and my
colleagues, Matt Williams andCurlander Calvez.
I think we quite like if peoplesay you can't do it.
I think we'd quite like to tryand show that we can.
SPEAKER_01 (31:16):
Yeah, get stuff done.
It's a good attitude.
I'd like to talk a little bitabout the future.
So this is the closing sectionof the podcast where we go on to
what you're thinking about andwhere you're going.
What are you excited about,Jonathan?
SPEAKER_00 (31:29):
I'm excited about lots, Sam.
It doesn't take much to get meexcited, I have to be honest.
I am excited about AI cominginto healthcare, but I think for
reasons different than mostpeople.
I often give a talk where We'retalking about bias in AI.
And I say, well, I'm the mostbiased thing in the clinic room.
(31:49):
Because as a doctor, mostdoctors are good because they've
become highly biased algorithms.
I'm looking as you walk in.
What's your gait?
Who's with you?
Got a walking stick?
What's the heels of your shoes?
Have you got nicotine stainingon your fingers?
They're all feeding into myalgorithm of what might be wrong
with you.
And that, to large amount of thetime, works really well.
(32:11):
It's why the medical studentgoes, Gosh, how did the
consultant know it was thatwithin a minute?
Because it's all the unspokenstuff.
But I think we're using AI tochallenge our bias.
Because whilst my bias workswell in certain patient groups,
it doesn't work all the time.
Does it work as well at 4 a.m.
(32:34):
for a man of black ethnicityaged 75 as it does for a
21-year-old white female?
Maybe it does, but it probablydoesn't.
And I'm not suggesting that's...
We're talking about bias at thelevel here of deliberative.
It's really, we're talking aboutour heuristics here, our recall
(32:54):
bias, what's our most recentepisodes and recollections.
And so I think having AI canmaybe point out, you think this
person's having a stroke, butgiven their age and this, that,
and the other, the most likelything is this.
It's not that we need AI to tellus what it is.
It's just AI to go, justchecking.
(33:14):
Do you think it's like the nicecolleague who's just going, oh,
it's interesting.
I've seen one of those, but itturned out to be this.
And I think that narrative isfirstly less challenging to
clinicians, but I think it putsAI is biased because the data
that goes into it and guesswhere that came from, humans.
And so it's more about admittingthat.
(33:36):
that it's impossible really tohave anything that's not biased.
But you just need that bias tonot overlap over the same
domains.
And then we can challenge eachother's bias.
So that really interests me indealing with decisions under
uncertainty.
In medicine, you go through aphase where when you're a young
(33:57):
doctor, you think, I'll be greatwhen I know everything.
Can't wait.
Then you're this sort ofregistrar doctor and you think
you do know everything becauseyou're currently sitting in
loads of exams, professionalexams, so you're quite bright.
Then you become exultant and youfeel like you know nothing
again.
But it's because you're facedoften with the edge cases, with
the nuance where there's trueequipoise.
(34:17):
And using AI to look at futurepossibilities and map those, we
can sit down and say, should westart your chemotherapy today or
hold off?
And at the minute, that's builton maybe a research paper, my
understanding of that paper,what I'm remembering today and
talking to you.
(34:37):
But what would be nice is to go,actually, Sam, we can look at
the AI goes, if we start today,this and this and that, this
isn't saying definitely willhappen, but it just structures
our conversation.
You say, well, actually, John,for me right now, most important
thing is I stay at work.
And option B looks like for ahundred Sam's, more would stay
at work doing that.
(34:58):
And that is true informeddecision-making, I think.
So those sorts of changes, Ithink, Sam, to me feel like real
paradigm shifts.
I don't think they're comingimmediately, but I think they'll
come.
We know from studies presently,at least, that patients all want
AI to be, shall we say,subservient to the doctor.
I think it'll be one of mycolleagues, Danny Rutter, said,
(35:20):
gave the analogy of lifts and Ithink you go back to whatever
the 1920s, there was always abellboy or a Lyft operator
because we didn't trust theLyft.
And then now you'd think it wasmadness if you saw an operator
in the Lyft buttons.
And I think he's probably rightthat maybe for the next 20, 30
years, there's a doctor in frontof this.
And then gradually everyonegoes, I don't know why I'm
(35:43):
waiting for it.
He just tells me what's on thecomputer anyway.
And then using it more in theback office to get rid of low
value, high volume, repetitivetasks like doctors checking
blood test results that arenormal and having to do a letter
to the GP to say it was normal.
That could all be vanishing out,you know, via an AI, couldn't
it?
So are you optimistic?
(36:06):
Good question.
I'm a pessimistic optimist.
I think everything will happenin the end.
Do I think it'll be easy?
No.
To put it in context, thistreatment summaries project, we
might be talking aboutindividual NHS trusts needing to
pay two to five thousand poundsa year for the platform beyond
the pilot.
(36:26):
And a lot of them have said, no,we can't afford that.
So that's the sort of barrier touse and implementation.
That's obviously concerning.
I think being devil's advocate,I think we've got a big forcing
function coming.
because the NHS, as it is, isnot going to be able to cope
with the scale of demand.
And so in a funny way, I thinkautomation will, a bit like
(36:49):
through COVID, will be pushedrapidly on us.
And so I think it'll feel likenothing's happening and then
suddenly, like dominoes, it willgo.
I think I am hopeful.
I'm hopeful as well that I dosit in far more conversations
now where social determinants ofhealth inequality are actually
considered a thought about Notenough and not adequately, but
(37:13):
when I compare that to, say,five years ago, I do feel that
it is different.
So I think I'm hopeful.
SPEAKER_01 (37:20):
If you could wave a magic wand and change one thing
about the way we treat cancertoday, what would you change?
SPEAKER_00 (37:29):
Oh, you've stumped me, Sam.
SPEAKER_01 (37:31):
Or it could be change about the health system.
SPEAKER_00 (37:33):
Yeah, yeah.
You've silenced the surgeon,Sam.
Amazing.
You need to keep this in thepodcast.
The world will be like, so muchfor him being an innovator.
Is that because
SPEAKER_01 (37:47):
there's so much that you would change?
Or is that because it's soperfect?
SPEAKER_00 (37:52):
I think it's because I'm searching for, if I've got
one wish, it's got to be themost impactful.
And that's probably, I'm thenthinking, oh, what do you know
what I mean?
That like, oh, what does thatlook like?
I think, and this perhapssounds, well, it's not
innovative.
I think if every person goingthrough cancer treatment didn't
have to worry about theirfinances, I think there's
(38:17):
something that would just make abig difference for a large
number of people and actuallydoesn't rely on technologists or
other things.
But to be worried about puttingthe eating on when you're on
chemotherapy, to be thinking howare we going to get stuff for
the kids, just that to me on topof everything else I think is
(38:37):
That is something that wecouldn't, society could choose
to alleviate.
SPEAKER_01 (38:41):
So it's a policy thing.
It's not a medical thing.
It's a policy thing.
SPEAKER_00 (38:44):
Yeah,
SPEAKER_01 (38:44):
I think so.
What's next for you, Jonathan?
Where are you going this year?
SPEAKER_00 (38:48):
So hopefully finish scaling the trial, delivering
the end of treatment summariesproject.
And then really, I think I'dlike to be, I'm thinking about
doing a piece of work aroundracism in healthcare.
And I've had the, recently readthe book, Divided.
by Annabelle Suemio and that'sopened my eyes to how baked some
(39:11):
of that is into the wayhealthcare is delivered and I'd
be interested in perhaps tryingto bring some of that into my
conversations and my work toperhaps challenge looking around
and challenging and findingwhere we're I think nowadays I
would say accidentallyperpetuating what has happened
certainly when I readAnnabelle's book I was
completely unaware of most ofthis so I think it's about
(39:33):
surfacing it and then trying towork to address that.
So I think that will be afeature for the next year.
But then say my work with theHealth Tech Research Center with
Macmillan and my own consultancyis really about working with
companies to develop productsthat meet an unmet need, that
(39:57):
are safe, that work, that canprove they work and have a value
proposition that means they canbe procured be adopted and
scaled in the NHS.
And I think if that's what I'mtrying to do, there's probably
going to be enough for me to dothis year, Sam.
Not a bit worried I might haverun out of things to do next
year, but we'll cross thatbridge when we come to it.
SPEAKER_01 (40:17):
A year's quite a long time really, isn't it?
Trying to think what you'regoing to be doing in 12 months'
time, let alone five years.
SPEAKER_00 (40:22):
I suppose the only thing that I had noted down
about the future, but I don'tthink it's probably worth the
bit, but it ties in, I think,with what you would be doing
It's trying to bring behavioralscience and design and human
psychology much more into theforeground of healthcare.
(40:43):
It feels to me a bit likeoutside of public health, where
we think about public healthmessaging.
I'm not sure we really apply itproperly to everything.
I don't think we use it whencompanies or people are
designing things for cliniciansor nurses to use.
I don't think we use it when wecommute to work.
(41:03):
public or with patients andsystem design.
So I'm always looking foropportunities where try and
surface that into the foregroundof the way services are designed
and delivered and rather lessthan perhaps being either an
afterthought or not thought ofat all.
SPEAKER_01 (41:23):
It still feels quite novel.
We work with lots of differentdepartments and often the work
also involves a communicatingthe why you're doing what you're
doing as well as doing whatyou're doing which is that's a
big overhead but it's aneducation piece as well
SPEAKER_00 (41:38):
yeah I agree as I've said I don't I honestly don't
believe I'm an expert inanything I just read a bit of
stuff and listen to people andtry and yet some of this I think
I think I'm the only personwho's in this room who's
mentioning behavioural scienceor whatever you think we should
be beyond that now Kahneman'swork was what the 1970s And I
(42:00):
think because it got listedunder behavioral economics,
everyone thinks it's onlyeconomics.
It's about decision-making.
And as we actually didn'tmention in the talk, the
scarcity book from SundellMillarison, that's the same.
It's decision-making underuncertainty and cognitive load,
isn't it?
And I think that should all bemandated into healthcare's
(42:21):
awareness if we just followedthose principles and some basic
human psychology.
I think we'd get an uplift onour services that would be far
greater than the amount ofeffort that went in.
SPEAKER_01 (42:34):
Jonathan, it's been a pleasure talking to you.
Thanks so much for taking thetime to join me today.
SPEAKER_00 (42:38):
Thank you very much for having me, Sam.
SPEAKER_01 (42:51):
Problems Worth Solving is brought to you by
Healthier, the collaborativeservice design consultancy for
transformation in health careand public services.
Find out more about how we canhelp you deliver user-centered
change at healthier.services.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Special Summer Offer: Exclusively on Apple Podcasts, try our Dateline Premium subscription completely free for one month! With Dateline Premium, you get every episode ad-free plus exclusive bonus content.