Player is loading...

Embed

Embed code

Transcriptions

Note: this content has been automatically generated.
00:00:01
it's wonderful to be back and then with my old friend j. f. we go back a long long way
00:00:08
my or my only young friday
00:00:13
um i got two reasons for being here one is to talk to you about what i'm interested in which is h. c. c.
00:00:19
the other is that i am very possible by nash and apple so i come to learn
00:00:25
as much as i do to to lecture so what i'm going to talk to about
00:00:29
today is one of the problems how we get uh the diagnosis of h. c. c.
00:00:39
if maybe
00:00:42
huh
00:00:48
action and so this is the outline of the talk a little bit of philosophy to start with good place to start
00:00:57
talk about one of the big problems i set the
00:00:59
main problem we face in h. c. c. focusing on match
00:01:05
some ideas about a solution to the problem uh look into the future uh look
00:01:10
into the past and then some conclusions so let's start with a little bit of philosophy
00:01:17
this is quite just how gracious flak us known asked horace
00:01:25
one of the founders of western literature and this is one of his famous
00:01:32
oh it's if you read on you would come to copy idea grass today
00:01:38
and this is what he says in his code is talking to his
00:01:42
girlfriend or is news you can know a which literally means the ad so
00:01:50
he says to have do not ask about the outcome of my day ease or yours look only
00:01:57
it's a secret beyond us and don't attend abstruse calculations well
00:02:03
that was just b. c. three coming into the third millennium
00:02:10
we can do this now what he's saying is don't try and foresee the future
00:02:16
don't try and get mathematical calculations to work out what's
00:02:19
gonna happen in the future doesn't work well now it does
00:02:24
and this is progress so we gonna move on from horace
00:02:30
to the days when we can do the things that the old gots used to claim to do
00:02:37
so here's the big problem
00:02:40
this is recent trends in us cancer mortality
00:02:45
on college rests up doing quite well so
00:02:48
this is the annual percentage change in mortality
00:02:53
or various kansas and you can see the progress
00:02:57
is being made the mortality from all kansas is
00:03:01
folding by about one and a half to two percent per annum but there are a few kansas
00:03:09
on the right where it's going up and the most traumatic one where it's going up
00:03:15
is liver cancer primary liver cancer which on vertical h. c. c. from now on
00:03:21
this is quite old now that's to two thousand and three if you look at the latest version
00:03:27
it was over here somewhere so there is a massive increase in crime
00:03:32
liver cancer and it's going to reach epidemic proportions as we will see
00:03:41
this is the problem of effective treatment as you all know is predicated on early diagnosis
00:03:48
all the guidelines say you should do your
00:03:53
screening for early diagnosis using ultrasound
00:03:59
implementation and compliance in the west is very cool
00:04:04
most people most setups the try and do ultrasound screening not do very well
00:04:13
we know who just screen because we look for people
00:04:16
with on the line chronically disease hepatitis b. c. alcohol
00:04:22
i'm particularly fatty liver 'cause this is the new frontier
00:04:29
so there is you problem how do you may early
00:04:32
diagnoses went ultrasound isn't particularly efficient particularly in fatty liver
00:04:39
yeah this is a
00:04:41
paper that comes from not far from here by somebody call philly cody and j. f. to fall
00:04:48
and they talk about the epidemic up out the whole it
00:04:51
actually live with a non alcoholic activity disease and h. c. c.
00:04:56
and in less summary won't get to read it at the back they say
00:05:03
they need to be enrolled in the screening program one of the challenges with
00:05:07
nash induced h. c. c. is the off the cases are announcer ought to patients
00:05:12
there's a need to identify these patients in order to screen them
00:05:16
for h. c. c. the obesity obese patients is a major challenge
00:05:22
so ultrasound don't work well if you've got an obese patients
00:05:29
so they limitations of ultrasound are twofold first of all
00:05:34
land lodge in most centres it needs a separate infrastructure
00:05:39
physicians call do in japan the physicians do their own ultrasound but
00:05:44
in most places in the west it needs to go to separate department
00:05:49
but the main thing is that ultrasound performs pulley
00:05:53
in the obese and this is a a diagram from
00:06:04
'kay
00:06:08
which
00:06:09
uh_huh
00:06:11
mm
00:06:14
going on strike
00:06:26
started off again to have a a well those with you on that one
00:06:35
yeah okay
00:06:38
okay so this is from toronto
00:06:42
which is the pinnacle of ultrasound screening in the west it is
00:06:46
that business they do big studies on screening for h. c. c.
00:06:51
and this was a figure they showed recently and it said about the quality of the
00:06:55
ultrasound picture by which they may get they get a good look at the whole liver
00:07:02
and the performance was good it only six percent of patients poll
00:07:07
in twenty nine point three and fat in sixty four percent so there are big
00:07:12
problems with using ultrasound for screening in
00:07:15
the general population the obese it's even worse
00:07:24
we need a simple blood test
00:07:30
we have to acknowledge that it's impossible to start screening the entire obese population
00:07:36
it's impossible statistically because the numbers that you would be trying to pick up
00:07:41
ah excruciatingly small you have your
00:07:44
radiologist all day looking normal fatty levels
00:07:48
and once in a while picking up an x. t. c. i. logistically it would be very difficult to setup so what we
00:07:55
need to do is to risk stratified population so that you
00:07:59
focus your sources of people who are highly likely to get cancer
00:08:04
and then implemented test early diagnosis in that high
00:08:08
risk population so we're separating out two things here
00:08:12
want is identify high risk population and then in that high risk population
00:08:20
attempting early diagnosis okay so we separating out those two concepts
00:08:26
so i think we can get someone into doing that i'm going to show you that now
00:08:31
so the aim of our work this is what i have done in my retirement
00:08:37
to develop a plot based test for h. c. c.
00:08:41
that could be used routinely in a diagnostic setting so dated a political protest
00:08:48
and also for risk stratification and early diagnosis
00:08:52
and you could use it potentially a either alone or in combination with ultrasound so that's
00:08:58
the aim of the project i'm pleased to shout out any questions that make it very informal
00:09:09
so we use green blood tests one of which will be very familiar to your
00:09:15
alpha feature protein true very ice a full of alpha feature protein called l. three
00:09:23
and another one which you may have heard of research was pave
00:09:27
ca same thing guys gamma cock supra from but all t. c. p.
00:09:33
so these are three potential by marcus know why choose these
00:09:38
now let me emphasise that there is no scientific rationale here whatsoever
00:09:45
it is and highly
00:09:50
entirely
00:09:56
area
00:09:57
why choose these three they can all be measured on the same system
00:10:03
secondly they already have f. d. a. approval for h. c. c.
00:10:08
risk they are very stable and well car tries but most importantly
00:10:14
they have extensive well annotated data have been recorded and all available to us
00:10:21
so we have zoological amounts of data and that's what i like
00:10:27
so that's why we chose it because we had dated to play with
00:10:34
important to note there is nothing new here all this data has been available for thirty years
00:10:43
but if it was knew it would be of no use to me whatsoever 'cause
00:10:48
the value of this is that we can go back in time because it is all
00:10:54
so there is huge normalcy pious inmates and they're all data has a lot of value
00:11:05
so we take those three by marcus
00:11:09
here they are l. three off of each protein and d. which is guys common desktop box across from them
00:11:17
put them together with gender and age and we get to school by the letters that and it's called gal at
00:11:31
no it looks horrendous but we'll come to that in a minute so the guy that school
00:11:36
zed is put into this algorithm age gender a f. p. l. three and d. c. p.
00:11:45
and that gives you a scroll
00:11:57
yeah we're
00:11:59
this i extract information from your data base is a decision to so you get your information
00:12:06
here and you get a decision tool on the side and and effect what you are doing
00:12:17
is changing analogue data into digital data
00:12:22
the principles are very rigorous statistical methodology to get this model
00:12:29
all the variables are treated in that continues for no putting
00:12:34
in high and low f. p. o. f. p. positive and negative
00:12:39
you put in the exact value sort all continuous
00:12:43
and then extensively validated in your own data set you
00:12:47
own hospital and then do it in external data sets
00:12:52
so that's the guy that model
00:12:57
now here is the way the individual work so here's the three by marcus
00:13:04
in pink is a controlled population who have been followed up we know who
00:13:09
do not half h. c. c. they've all got chronic liver disease also roses
00:13:14
and the blue all the h. c. c.s and you
00:13:18
can see each case elevated but there's plenty of overlap yeah
00:13:27
not you assess how cold the quality of your by market is by doing a rock
00:13:34
curve a receiver operating characteristic curve part sensitivity
00:13:39
against well minus specificity and you get to the
00:13:45
the area under the curve tells you how good your by our market is
00:13:50
if it was fifty percent and the line went like that would be a totally useless file market
00:13:58
if it went straight up like that and across and the area
00:14:01
was a hundred percent then it would be a perfect by market
00:14:06
so what you can see immediately is that each of these
00:14:10
is quite good it has area on the rock curve of
00:14:14
point take four point nine so that's what the three of them on
00:14:18
their own can do but when you combine them in they got scroll
00:14:29
so that's the original before you put them into the model this is data from japan you know get at the top
00:14:44
there's there's there's there's no you can see the blue line
00:14:47
and now it goes up to almost perfect point nine seven
00:14:54
no i have to tell you when we first show these results um
00:15:00
several of the key opinion leaders shall we say
00:15:05
dismissed it it was p. g. t. v. t. two good for
00:15:10
and um i was advisory boards and they told the companies that we're developing their own diagnostic test don't
00:15:16
bother about this it couldn't possibly be it couldn't possibly be true but you just don't see point nine seven
00:15:24
but it was true
00:15:26
so this is the international validation we did big numbers of patients more than six thousand
00:15:33
u. k. japan hong kong germany and you can see in all cases the
00:15:40
guy that model point nine seven point nine three point nine six point nine four
00:15:47
seem to work in all of them
00:15:54
yeah
00:15:58
our actual
00:16:00
so the improvement of the individuals is modest but very consistent
00:16:06
but there is a problem of course it was and da not done in
00:16:10
america so if it's not done in america it's still i still a bit wobbly
00:16:20
but then it was done in america and publish at the
00:16:23
end of last year and i think much to my colleagues
00:16:27
surprise they got the identical results so this was
00:16:32
published in a. t. and you see it comes from
00:16:36
about a thousand patients in all the big live a sentence across the united states
00:16:43
and they got exactly the same results so this is point nine five the
00:16:49
overall group and interesting way only went down a little bit to point nine two
00:16:56
but the the disease this means patients with cancer is
00:16:59
less than three centimetres those you're looking for the screen program
00:17:06
and it was almost embarrassing the the yeah the paper
00:17:11
and this is what it said in in in the um discussion yeah excellent
00:17:16
performance of got that was confirmed for the first time in the u. s.
00:17:20
showing to be superior to ultrasound so ultrasound had an
00:17:25
area under the curve about point eight compared to point nine
00:17:31
five or something like that it was maintained for the uh the
00:17:34
detection of h. c. c. including chew most with negative i have a
00:17:40
it was not affected by age gender etiology or the
00:17:45
severity of the liver dysfunction and the outstanding performance was confirmed
00:17:50
for the first time in a multi sent to us cohort for
00:17:53
the the stage actually see so um it looks like job done
00:18:02
i'm particularly important that it was early stage h. c. c.
00:18:07
i'm particularly important that it was not affected
00:18:10
by etiology but they didn't lock at nash
00:18:15
or they didn't have the details to breaking down according to match but fortunately
00:18:23
we've been working with a group in germany this is from young best
00:18:28
and this is um presented these all and it's on the revision with the journal now
00:18:34
and this is in a german sent us a hundred and twenty one patients with nash
00:18:40
and two hundred and twenty four nash control patients and you
00:18:45
can see again in red the guy that gives just the same
00:18:50
and in the early patients that's less than three centimetres there weren't
00:18:54
a lot only thirty i think but it's still point nine three
00:19:03
i was even more interesting is because of my wonderful colleagues in japan you can look at what happened
00:19:10
for that kind of scroll before they actually get
00:19:14
the cancer because they got samples going back many years
00:19:18
and if you plot them out you can see those that get cancer of separated out
00:19:24
six years before they actually get the transit
00:19:29
detected within the screening program will come and that's
00:19:33
so let me emphasise that what you see now i think that's all true i mean it's all published anyway whether or not
00:19:38
it's true is a different matter it's all published what i'm going
00:19:41
to show you the last half is um unpublished work and um
00:19:50
probably a bit speculative
00:19:54
now if we go back here i don't if anybody's noticed it but there is something strange going on
00:20:05
if you've worked him by omar 'cause you know
00:20:09
that they don't generally work in the the stage disease it
00:20:13
seems crazy the tape by our market can be works well
00:20:18
in this tiny too much as it does in the picture something very strange going on we're going to look at that
00:20:28
so
00:20:30
we had a look at the effect of
00:20:33
i'm too much size on how well it worked and here's the data
00:20:38
so again this is japan 'cause they have the biggest number of patients
00:20:42
but look less than one centimetre point nine for less than three point nine less than five point nine
00:20:51
even if local
00:20:53
less than ten doesn't make any difference what the
00:20:56
sizes it works just as well and that is very
00:21:02
very strange and i think i can show you why that is i don't know if anyone would like to suggest the reason
00:21:10
of course this get chicks said
00:21:18
oh no come back to that yeah so if this happens the question is is it possible the guy that's cool
00:21:25
could be effective oh elevated before h. c. c. is seen clinically
00:21:31
no most kansas you could never asked a question imagine breast cancers that you wanted to know what
00:21:37
was happening to a walker ten years before the lady developed breast cancer the data just isn't available
00:21:45
but in japan
00:21:47
in a little district hospital like hacked the data on all these things
00:21:53
going back twenty five years and very generously gave me all that data
00:22:00
in fact to kind of given me just about the whole national data set which is just amazing
00:22:07
and this is the data set we got from that this is a summary of what we've got now wasn't the original
00:22:14
six thousand patients with chronic liver disease about accord with h. p. v. fifty percent with
00:22:20
h. c. v. twenty one percent with non a non b. many of those will be nash
00:22:26
they were in an intensive screening program they were followed up for more than fifteen
00:22:32
years with a hundred and fifty thousand observation time points and each of those time points
00:22:40
h. e. geology the buyer marcus marcus uh fibrosis can function and i
00:22:46
lefties were measured and sent to us in a nice big excel spreadsheet
00:22:51
over that period six hundred developed h. c. c.
00:22:57
the only medium size of cat detection one point eight centimetres
00:23:02
and five thousand four hundred didn't know that's the sort of data you can work on
00:23:09
and i have to give acknowledge to my
00:23:12
good friend uh had nor a toyota and catchy commodity to just gave us the data
00:23:21
the problem is all these patients all this data so some patients had
00:23:29
a hundred and five observations over thirty years
00:23:33
how on earth do you aggregate data when it is collected a regularly
00:23:39
very nicely done every six months and it's done six
00:23:43
months then nine months than twelve months very difficult to tell
00:23:48
so
00:23:50
that said technically very challenge to aggregate all this
00:23:53
data because it's collected thirty regular time points and um
00:24:00
my statistical colleagues were not particularly forthcoming on how to do this
00:24:06
and i complained to my daughter oh my watch here so it's our daughter
00:24:10
that uh we didn't know how on a us to aggregate all this data and she came up with
00:24:17
the immortal line at that let us get the kids to bed and then we'll do it for you
00:24:23
and she and her husband settle down and did it and uh um
00:24:30
i tilted at the mass and a husband that the programming and they set the computer going over night on a
00:24:36
packet of frozen peas because it not so hot it was so much work but in the morning to have an answer
00:24:45
and this was it
00:24:49
and i told them with typical professorial pomposity that this couldn't
00:24:54
possibly be right so let me take you through the slide
00:24:58
this is the guy that scroll going up here and this ha i'm going backwards in this access
00:25:05
so this is six years before they would be like nodes
00:25:10
on screen e. here to medium size of one point eight centimetres
00:25:15
yeah
00:25:17
see looking back in time so that by marcus all separated here six years before
00:25:30
very much the key observation at these levels all low
00:25:35
these are with a in your daily prime is what you would
00:25:39
call the normal range this is sort of eight and this is true
00:25:43
not a great but yeah he but statistically that completely separate
00:25:47
that if you put the confidence intervals all they don't overlap
00:25:53
i thought this was wrong but it looks like it's not and they were absolutely right
00:25:58
and i think what this tells us
00:26:01
is that this is defining the population who are at risk
00:26:07
now when i last saw j. f. i was telling people that this was probably very early cancer i don't think it is
00:26:14
i think this is just telling you that population at risk of getting comes up
00:26:22
you'll notice we see this in all the data sets we look at
00:26:27
there's a terminal right c. c.'s inflection point here
00:26:30
and i think this is when the clinical cancer develops
00:26:36
someplace terminal rise is reason is driven by all three bile marcus that's got that
00:26:44
this here is just driven by a f. p. so patients
00:26:48
who are going to get liver cancer have elevated a f. p.
00:26:53
many many years if you are a biologist you can think of this as a
00:26:58
not sent to hitchhike offices kept the first haiti's here on the second it is here
00:27:06
but two separate things separation driven by i have eighty years
00:27:10
before terminal rice driven by all three or by marcus uh
00:27:19
again greats get to system
00:27:22
two things people said well perhaps the bile marcus i'll coming from that you must live or
00:27:28
all the tomorrow
00:27:30
they that perhaps the by marcus are coming from the non shoe must live up not that you
00:27:36
second thing that people said with with classical western arrogance
00:27:41
well japanese h. c. c.s are just different yeah forgetting
00:27:48
but both were wrong both were wrong
00:27:52
so you can't do the clinical test to find out if they're coming
00:27:55
from the tomorrow not all you do is take serial samples after re section
00:28:01
and here it is
00:28:04
here's the got school is time after reception
00:28:09
section because here so what's happening is the true but we got score starts here
00:28:14
and then they get reset and it falls so this graph will fall on that side
00:28:19
and these are the people who rica and these are the people who get no real comes after section and you can see
00:28:26
whether or not you get recurrence you get to dramatic fall got so it tells you pretty certain they that
00:28:34
that by omar 'cause i actually coming from the true but it also tells you that
00:28:38
you can tell who's going to get to recover buttons within three months of doing the operation
00:28:45
so the second thing is well is it really just a japanese phenomena well again we were very lucky
00:28:54
we have friends in scotland who are still part of the united kingdom for a while um okay so
00:29:01
this is now forty this is what you've seen from japan this is six years before they get the cancer
00:29:09
here's the f. p. conceit separate and you can see the terminal rise
00:29:13
yes these are the people who don't get the cancer flat as a pancake
00:29:19
now look at scotland
00:29:22
just the same separated out at the beginning
00:29:26
and then a shop terminal rice completely independent
00:29:31
data nothing to do with me japan does that look else what else it shows you
00:29:38
this is a scottish times go this is fifty years before
00:29:45
and this data is
00:29:48
available on the web scottish group put it out into the community so anybody here could just go to the journal
00:29:54
download the data on plot it and you'll find this it was a complete surprise to them
00:30:02
so if we just summarise how far we've come
00:30:07
f. p. levels i'll re used in those destined to develop h.
00:30:10
c. c.'s at least ten years before clinical detection of h. c. c.
00:30:17
and they are two related
00:30:19
there's an inflection point two years before the clinical detection here
00:30:25
and we've now had data from two distinct japanese centres in osaka
00:30:31
they show exactly the same thing so we've got three independent data sets all showing the same thing
00:30:39
i think it's real
00:30:43
notice
00:30:46
for scientists and there is nothing new here at all
00:30:52
there is nothing experiment no experiment we just looking deeply into data
00:31:00
just observation
00:31:02
totally and fun double can you imagine going to a funding agency
00:31:06
and say we'd like to measure real by marcus for twenty five years
00:31:12
nobody would find it
00:31:14
and of course it's data that you could never obtained now
00:31:18
if you what bill gates you couldn't buy thirty year old serial data absolutely unique
00:31:26
now there's two issues with such mathematical models that's the guy that's cool
00:31:32
it locks on friendly
00:31:35
so you can imagine saying to one of the hard work doctors on the wall oh just ah
00:31:41
calculate the got scroll where you put all these variable in it and
00:31:45
tell me there's so it looks very unfriendly and secondly they buried black boxy
00:31:53
i don't know if that translates to switzerland it means you can't see what's going on yeah
00:32:01
you can't see what's going on inside the mall
00:32:07
so the first way of doing it is you have a graphical user interface and the mayo clinic with
00:32:15
the ones who let the american study that i just showed you put this on their website very quickly
00:32:22
and all you do is you put a gender age
00:32:26
a f. p. l. three basically hit submit and it
00:32:31
tells you what the percentage likely what is your patient
00:32:34
with chronic liver disease has is operating at h. c. c.
00:32:41
my suspicion is that not many good things that come out of england it will be
00:32:46
born in the u. k. and raised in america but we will we will we will see
00:32:52
certainly american farmer how much more interested than english for
00:32:58
so how can you visualise what's going on in this black box that's much more challenging
00:33:06
so this is how we've done it some people can see the some can you see this in three dimensions
00:33:13
so what we're going to do is take those three by omar 'cause the c. p. a. f. p. and l. three right
00:33:21
are we going to lock them in each individual patient in three d. space
00:33:28
so that's the control group you can see this two populations but they're blue okay those are people with chronic liver disease
00:33:35
now you put on p. h. p. c.'s and then right
00:33:41
and can you see that they are sorta separated
00:33:46
page three dimensions so what you can't put is a straight line through
00:33:51
them what you have to put is a sort of blanket in three dimensions
00:33:56
that separates them out optimal and that mathematically is very challenging the p. h. d. student
00:34:04
who did it it took him three weeks i'm very politely said it is not trivial
00:34:11
so there's a lot of programming in matlab but he get it and our credit in at the end it's great
00:34:20
so now we have to plot what's called a hyper play in three d.
00:34:24
space but ultimately separates out those two populations and this is what it looks like
00:34:31
so can you see the reds out here
00:34:34
and the plumes
00:34:36
by and large underneath the hype play yeah not perfect
00:34:42
and then
00:34:44
some people say that very pretty pretty curve
00:34:50
this is a claim in mathematics is known as a decision surface
00:34:55
or decision boundary so actually optional curve the separating out these two populations
00:35:04
and
00:35:06
we can spend this round in three dimensions become
00:35:10
now what you think j. f. can we come down and
00:35:15
there it is if you just hit that left button well done
00:35:21
there it is and then you can interrogated
00:35:24
and you can say you can stop it wherever you want
00:35:30
yeah
00:35:32
for very pretty
00:35:41
using that you can actually get the statistics come to that
00:35:46
we haven't done this properly yet so this is just a preliminary lock see
00:35:53
but the important thing to notice is that
00:35:57
normal humans like us can only work in three dimensions mathematicians have
00:36:02
no problem in working in any number of dimensions that you want
00:36:08
and you can figure that out if you like it's difficult but they do that we have
00:36:14
got them to work in five dimensions so they can put all they got that criteria and and
00:36:22
it's at least as good as the by statistical model it's probably better so if you look at the u. k.
00:36:27
correctly classify it was ninety percent with the old model in ninety four
00:36:33
percent with the um with the fifth generation got that safe uh early
00:36:41
eighty seven and ninety three so it looks as though it may have utility but it's very early days
00:36:50
interesting way etiology affects the boundary
00:36:55
so the h. c. v. patients have a very different shape
00:37:00
than the h. p. v. patients
00:37:07
i'll just show you this because i think it looks pretty it's not particularly informative
00:37:15
but you can stop it and look down the channel
00:37:18
anyway you could you say that they have they have different boundaries
00:37:22
so let me just see if i can go about or
00:37:27
now what you immediately think i'm sorry this is confusing because
00:37:32
the blues of the kansas here the rights of the controls and vice versa
00:37:37
what your
00:37:40
right
00:37:42
what you are i know you all burst into x. a. is presumably
00:37:48
these red it's the kansas once upon a time live under the boundary
00:37:55
they must have travelled across over time yes
00:37:59
same for these they must've gone from there over to there and again because we've got the data
00:38:06
we can see that so now we've plotted the
00:38:10
three d. boundary and here's a patient undergoing surveillance
00:38:16
this is the time from here
00:38:20
until he gets his cancer okay
00:38:24
and you're going to see him move and cross that boundary and go
00:38:28
on to the bad side the cans aside from the chronic disease yeah
00:38:36
yeah
00:38:40
harry goes so you can see the time at the top counting down
00:38:44
so he's twelve months nine months and the areas cross the barrier
00:38:55
yep
00:38:57
so you can see him crossing and he crosses a long time
00:39:02
that's ten spit right around the cross as a long time before
00:39:07
a um
00:39:09
actually gets the cancer diagnosed so you'll see a lifetime speed it up so this is three seconds
00:39:16
to one year so this is life in pure
00:39:19
holy abstract mathematical terms speeded up about a million times
00:39:27
now of course the next thing y'all longing to ask is what happens if you chop the chew me out there
00:39:35
does it go back to where it's come from
00:39:42
this is challenging me
00:39:45
yeah
00:39:48
so here's three d. changes in response to treatment and so here's the patient
00:39:58
so you can say you know the patient on the surveillance has a cross the barrier
00:40:04
and ended up here now we're going to chop the tube out and see what happens
00:40:16
what time is it yeah
00:40:18
okay five minutes
00:40:25
now he has the operation and you'll see very quickly
00:40:30
well there goes back on the nice
00:40:36
time from treatment at the top time to recover ins so four years later he rica's
00:40:45
and actually comes and you see what happens
00:40:53
twelve months to go
00:40:56
like a coffee
00:40:59
there and how to comes
00:41:02
now he's rica interesting saying is
00:41:07
the t. because
00:41:09
along almost the identical plane
00:41:13
probably reflecting clone now it's a it's probably true requirements
00:41:18
but it doesn't always happen
00:41:21
this is the the last one
00:41:26
his uh not it's not the same one is it
00:41:29
no i don't think so so here's another one yeah time to diagnose this you
00:41:34
can see is unstable coming up and hitting the barrier thirty two months twenty nine
00:41:47
harry goes out and he's rises can so
00:41:51
now swivels round
00:41:55
and he's
00:42:01
just waiting to cooperation time
00:42:05
now he has his operation
00:42:09
now back on the nice
00:42:13
stays underneath for
00:42:15
two years and now he rica's
00:42:19
but this time it doesn't come out the same way
00:42:23
so my guess is that this is
00:42:26
different clothes
00:42:28
and i wonder if we are detecting effectively approach your mic or functional
00:42:35
analogy
00:42:37
i don't know yeah lots of work to do
00:42:42
so just to finish off the show you where this goes
00:42:46
so
00:42:50
this office as a practical approach risk ratification so this and is tell you which risk
00:42:57
group you're in this is early diagnosis and i'm going to show you how we use this
00:43:06
so what we have to do is we have a calculator again you can have it on your i. phone
00:43:16
i hope this is going to work
00:43:21
click start
00:43:24
patient information in order to use this calculator you must've been diagnosed
00:43:28
before today uh please insight today so data for screening we've got
00:43:35
january nineteen
00:43:37
male or female what was your first ages screening was u. f. p. level then
00:43:45
thank you for filling this and
00:43:48
now you update your serial f. please
00:43:53
so here we are we putting them in six eight eleven twenty sixty three and then you hit submit
00:44:03
and it was it out for you and it tells you your current risk of developing h. c. c. is ninety eight percent
00:44:10
competence intervals ninety five to ninety nine yeah this isn't
00:44:14
you measure this is not the chance that you've got it
00:44:19
this is a lifetime risk going forward so it tells you what your risk is not
00:44:26
if you've got that chance they usually want to be going screen for early diagnosis yeah
00:44:33
now if you see somebody has a patient to jeff p. stays low for two three three four to
00:44:40
e. flat as a pancake your current risk of developing actually sees one percent
00:44:46
and we are ninety five percent confident your skis between one zero so you could effectively say these people
00:44:54
if you have a flat a f. p. you do not need to
00:44:56
be screened every six months go to your three years i don't know
00:45:03
so it seems like a f. p. the oldest serology to market is
00:45:09
the teacher risk stratification h. t. c. some of you will know that
00:45:14
the been articles written a f. p. has had its own a butchery
00:45:18
written in the literature so i think it's still quite a big deal
00:45:25
so i'm i'm gonna skip this one no not just got two minutes i care
00:45:31
um
00:45:34
when you use when you use 'em when you've identified your high risk group of patients you want to go into early
00:45:40
diagnosis and thought that you use a technique called statistical process
00:45:45
control and very briefly what this does is it uses the
00:45:51
not the absolute levels of cow that but the changes so this is the raw data this
00:45:58
is what happens this is what you see the separation early on and then the rice and then
00:46:05
those that don't get it if you use this
00:46:07
statistical process methodology called cumulative sum it exaggerates that
00:46:13
and it makes it much more sensitive and you can see that patients and get h. c. c. f. to rise
00:46:20
but you can detect long before you would see it in the in
00:46:24
the native and yes they non h. c. c.s no change at all
00:46:30
the next version of the program is going to apply artificial intelligence to the model
00:46:36
so the model the this drawing routine clinical application
00:46:39
and that by continually improves its performance so this is
00:46:43
what i think it will look like this is just as quick scratch off through a couple of hours ago
00:46:49
you have a central computer let's say the is the
00:46:52
centre of switzerland everything revolves around but but then keeps they
00:46:59
computer here with the model that syria f. pays off that in any time
00:47:04
identify is people who are high risk group they then have doubt that measured regularly
00:47:11
that makes uh the diagnosis and then
00:47:16
once they have um early diagnosis on on gal ad and confirmed by scanning all biopsy
00:47:26
the end it will tell you what the optimal treatments are
00:47:30
because we got all the data on the treatments of these are the patients and it will tell you what the outcome is
00:47:37
and this is how it feeds out the data it will tell you if you have a
00:47:41
transplant reception regulation face systemic therapy what your six months one year three a five year survival hours
00:47:47
and what your chance of q. or is we can talk about that like if you're interested
00:47:54
what's nice is you probably don't have to do trials because the model the system
00:47:59
i will tell you what it's performances each yes it sure it will say
00:48:05
you are picking up x. percent of your actually sees the early
00:48:10
lesson three sentences and it will tell you what the survival is and it didn't do comparisons
00:48:15
with other health care services see if you're keeping up with the people of the road is your
00:48:23
so um last slide just to show you how long the sort
00:48:28
of work right this is a paper that i wrote when i was
00:48:33
younger uh before most of you in that many of
00:48:36
you in this room aboard this is ninety eighty ninety eight
00:48:43
in those days
00:48:45
to measure a f. p. i had to make my own i say i have
00:48:49
to i said i'd isolate i have p. from patients serum injected into bunny rabbit
00:48:55
make an antibody build the up radioed in you know as say
00:49:00
and then click the samples it was very tough in those days
00:49:05
and it says here and this is ninety yeah j. n.
00:49:08
c. i. could journal nineteen eighties of very nearly forty years ago
00:49:13
based on a f. p. concentrations at the time the presentation and i have p. gobbling times
00:49:18
suggests that increasing concentrations of a f. p. detectable by radio meaner say they were quite novel and
00:49:25
maybe maybe present eighteen months before the impair claims of symptoms when no scans in those days
00:49:34
in this way screen a high risk groups could
00:49:38
be to significantly earlier diagnosis that was forty years ago
00:49:45
so acknowledgements to all the wonderful collaborators all around the well
00:49:51
particularly japan a particular the to the rest of my team there's only three of us this is sarah holly who
00:49:59
does all the statistics on my wife was here today who looks after all the logistics of this on rudy group
00:50:08
and here's the conclusion
00:50:11
i thought was another acknowledgement others another acknowledgements light yeah just to say i'm
00:50:16
chaining brown it the three d. decision boundary david hughes did the a. f. c.
00:50:22
a risk stratification and this is the people who did the um
00:50:28
original long digital data analysis this is the ones who did it when they put the children down
00:50:32
when they get it the steel rule very little so this is our door to emily
00:50:36
and this is a husband who's a mathematically columnist and did all the programming so conclusions
00:50:44
i've lost it but the conclusions are but um as you can tell
00:50:49
that using plot test possible to make significantly early diagnosis i think we're getting close to
00:51:06
sure uh_huh
00:51:14
yeah good idea yeah yeah
00:51:18
yeah
00:51:27
on and i own i yes
00:51:36
you can never tell you know sears that you call tell because you don't know who because you'll
00:51:44
positive they might not be false positives you see it might be the scanned the is
00:51:49
the false negative 'cause i still too small to be see what you do definitely half
00:51:56
is false negatives
00:51:59
that is people who the model says do not have h. c. c. but you have a cheesy and there's a small number of those
00:52:06
and there is a wonderful to 'em up some of you may have heard a bit fuzzy logic we using this
00:52:14
type of analysis fuzzy logic to sort these people out there is only a small number sony is small
00:52:24
i wish him the e. e.
00:52:32
yes which e. e. c.
00:52:40
very good very good question and i do not know the answer to that yet and
00:52:46
how will we get the paper back from the reviewers up no doubt they
00:52:49
will ask us what you just put in one value you get an answer
00:52:53
but i don't know what the confidence limits are all that so
00:52:56
that if the this last thing eyes showed you the q. some
00:53:00
i did that much to briefly but you definitely have to get the baseline that to see the deviation from the baseline but
00:53:07
i would think it's probably something like three months three points
00:53:12
within a year or something like that but i honestly don't know
00:53:16
good point
00:53:18
yeah
00:53:25
i e. e. e.
00:53:30
yes yes yes yes
00:53:35
i'm using the e. m. which is you
00:53:40
o. h. m. m. o. g. e. age sure he sure i and v. h.
00:53:51
sorry more uh_huh uh_huh oh yeah he's
00:53:58
your mike i again i don't know the answer but you're absolutely right
00:54:03
be sort of models that we're making at the
00:54:06
moment rely on baseline data that is buried twentieth century
00:54:12
the future is as you say is going to be looking at changes
00:54:16
but then the maths is much more difficult and getting the data
00:54:23
is much more difficult only it seems japan has the
00:54:27
serial data so if you set off to do this prospectively
00:54:31
you've got an awful long wait you're absolutely right there real and stuff
00:54:36
the next generation is really looking at changes looking at kind ethics
00:54:41
over time we we we e. we do
00:54:51
but jeff was just telling me that he
00:54:53
has think has perspective data with serum stored so
00:55:02
oh they've already got h. yeah okay okay okay okay okay okay you can yeah
00:55:11
oh yeah right yeah oh well there you go where okay very good to look at it

Share this talk: 


Conference program

Very Early Diagnosis of HCC : Focus on NASH
Prof. Philip Johnson, Institute of Translational Medicine, University of Liverpool
21 Feb. 2019 · 6:05 p.m.
Antidiabetic Therapies and HCC
21 Feb. 2019 · 7:02 p.m.