1
00:00:00,780 --> 00:00:04,350
Okay. Could you just start by saying your name and your title and department?

2
00:00:04,800 --> 00:00:12,060
Hi. My name's Andrew Sultan, and I'm a doctor, a medical doctor that is working at the John Radcliffe Hospital predominantly.

3
00:00:12,330 --> 00:00:19,800
And the research work we're talking about is with the Institute of Biomedical Engineering, which is part of the Department of Engineering Sciences.

4
00:00:20,100 --> 00:00:22,830
Okay, that's great. And what kind of doctor will you.

5
00:00:23,220 --> 00:00:29,550
I'm so I'm I'm a junior doctor training to be a cardiologist and and halfway halfway on the road to to that.

6
00:00:30,120 --> 00:00:35,280
Yeah. Let's develop that a bit further. So can you just tell me, I mean, we don't have time to do the whole thing,

7
00:00:35,490 --> 00:00:41,340
but can you just tell me a little bit about how you got to where you are now, starting from how you first got interested in medicine?

8
00:00:41,880 --> 00:00:51,000
And well, it say that that question stretches back a lot, but I think from a very young age, I was always interested in fundamental technology,

9
00:00:51,030 --> 00:00:57,840
science and biology and for various reasons, which will have been kind of specific at the time and evolved with time.

10
00:00:58,140 --> 00:01:00,120
And I decided to apply to medical school.

11
00:01:00,420 --> 00:01:06,960
And I think the reasons I'm very glad I'm practising today are very different from the reasons I applied to medical school all these years ago.

12
00:01:07,650 --> 00:01:13,530
But at the same time, I'm still glad for that decision. And so I studied for six years at Cambridge.

13
00:01:13,530 --> 00:01:21,900
I was at Christ's college and the course was set up with three years of undergraduate heart lecture based material,

14
00:01:22,500 --> 00:01:27,239
including a year in immunology and virology, which was fascinating and later proved to be relevant.

15
00:01:27,240 --> 00:01:33,420
And then three clinical years, which were based at Spectrum British Hospital and the surrounding hospitals like Ipswich,

16
00:01:33,420 --> 00:01:40,409
Bryson Edmunds, Stevenage, Peterborough, we did a lot of travelling and in the east of England and after that I came over to Oxford.

17
00:01:40,410 --> 00:01:46,200
It must have been 2018 and I've been here since then for four years now.

18
00:01:48,550 --> 00:01:52,330
And I quite enjoyed it here. I appreciate it.

19
00:01:52,330 --> 00:02:00,040
Slightly heresy. I've enjoyed it. And and before COVID came along, had you become engaged in research work in Oxford?

20
00:02:00,460 --> 00:02:07,380
Yes. So I'd met a postdoctoral researcher in David David Clifton's lab, and at a conference in Cambridge.

21
00:02:07,390 --> 00:02:12,190
And when I came across, I met with Hashim paintings. And she introduced me to David.

22
00:02:12,340 --> 00:02:15,370
And David and I got on very well.

23
00:02:15,370 --> 00:02:20,590
And. And he's in biomedical engineering. Yes. And David is a professor of clinical machine learning.

24
00:02:21,490 --> 00:02:24,819
And we had started work on on a small number of projects,

25
00:02:24,820 --> 00:02:32,440
all of which were machine learning based on the general theme of it is looking at the kinds of clinical data that are collected in routine practice,

26
00:02:32,440 --> 00:02:40,239
and that can be blood tests, vital signs, it can be textual medical records, and David's lab speciality is anything but imaging.

27
00:02:40,240 --> 00:02:42,100
They could actually be anything but medical imaging.

28
00:02:43,330 --> 00:02:54,640
And we, at the time pre-COVID, were working on a project looking at a disease which affects older men and called an abdominal aortic aneurysm.

29
00:02:54,730 --> 00:03:00,250
These are data. It's a key blood vessel at the back of your tummy and which can balloon out to a portion of men.

30
00:03:00,700 --> 00:03:07,930
And the difficulty with with patients who have this condition is a lot of people have a small amount of ballooning,

31
00:03:08,500 --> 00:03:12,579
but only a very small portion of those will progress for that balloon to get bigger and then eventually pop.

32
00:03:12,580 --> 00:03:18,340
And that would be life threatening. And but many will will live the rest of their life happily with a small amount of ballooning.

33
00:03:18,340 --> 00:03:19,840
And there would be no consequence to that.

34
00:03:20,500 --> 00:03:26,530
And one of the difficulties for surgeons is deciding who do you operate on which which of the patients are going to progress?

35
00:03:26,530 --> 00:03:30,190
And we will try to use routinely collected data as part of a project.

36
00:03:30,250 --> 00:03:32,680
I'm under. Mr. Leaves, not professionally.

37
00:03:33,040 --> 00:03:39,630
I'm in the now for Department of Surgical Sciences and to try and pick out which patients are going to progress more quickly,

38
00:03:39,640 --> 00:03:46,960
which patients we're going to progress slowly and and hopefully that would one day feed into a system to decide which patients to operate on.

39
00:03:47,590 --> 00:03:51,700
So tell me a bit about in medicine. So what do you what are you asking?

40
00:03:51,700 --> 00:03:58,900
So you're trying to write an algorithm essentially to take this routinely collected data, you tell me.

41
00:04:00,520 --> 00:04:04,719
And so I means a lot. It means a lot of things to a lot of people.

42
00:04:04,720 --> 00:04:13,450
And it's a very exciting field. I think it's broadly helpful to look at AI in perhaps two broader categories.

43
00:04:13,450 --> 00:04:21,040
One is what I would call the simple AI, which is an older approach, and what we've done actually falls into the simpler AI category.

44
00:04:21,430 --> 00:04:25,089
And there's been a category of what I would call hierarchy.

45
00:04:25,090 --> 00:04:30,579
This is the long term goal to design a machine or a system that is autonomous,

46
00:04:30,580 --> 00:04:35,110
truly autonomous, and can can work for multiple purposes, multiple functions.

47
00:04:35,560 --> 00:04:42,700
I think for now, the successes and innovations we've seen in artificial intelligence in the last, say, ten years, they've been in both fields.

48
00:04:42,700 --> 00:04:46,120
There have been some great breakthroughs, particularly from Google, DeepMind,

49
00:04:46,420 --> 00:04:51,040
and when it comes to the broader goal of generalised artificial intelligence,

50
00:04:51,040 --> 00:04:56,980
but in what we call in the field narrow AI, which relates to A.I., which does one thing, A.I., which does one specific thing.

51
00:04:57,430 --> 00:05:01,209
And that's that's the field we've been working in.

52
00:05:01,210 --> 00:05:08,410
And what what's the reason we work in that field is that it's it's a little bit it's quite a bit easier.

53
00:05:08,500 --> 00:05:13,780
And the methods that exist today already are quite good and that kind of AI.

54
00:05:14,770 --> 00:05:18,490
Performs a single task or single question. In all case, it makes a single decision.

55
00:05:18,550 --> 00:05:21,850
Does this patient have COVID or do they not have COVID and.

56
00:05:23,050 --> 00:05:30,970
That kind of challenge. I would broadly describe to people as almost, almost a fantasy type of pattern recognition.

57
00:05:31,000 --> 00:05:34,360
I mean, we use positions at medical school that taught me a lot of medicines,

58
00:05:34,360 --> 00:05:37,420
pattern recognition, you spot the patterns in the data and reach diagnosis.

59
00:05:37,690 --> 00:05:41,319
This was all about and getting a machine to spot the patterns.

60
00:05:41,320 --> 00:05:46,000
And because it's a machine doing it, not a human, the machine doesn't get tired, it doesn't get hungry, it doesn't get sleepy.

61
00:05:46,180 --> 00:05:50,050
And it can look at more things than humans cannot work reliably well for everyone.

62
00:05:50,440 --> 00:05:58,660
And whereas, it might take 20, 30 years for a physician to see two or 3000 cases, an algorithm can see that in a minute or two.

63
00:05:58,990 --> 00:06:02,440
And it's that kind of very narrow, very limited, very focussed.

64
00:06:02,440 --> 00:06:08,650
Hey, AI, where I think at the moment we're seeing tangible applications beginning to creep into clinical practice

65
00:06:09,010 --> 00:06:12,819
and there are many who are working on the broader goal of trying to get something which is generalised,

66
00:06:12,820 --> 00:06:18,820
more intelligent, which much more resembles a human. But I think it will be a little bit of time before we get to this.

67
00:06:18,820 --> 00:06:21,940
Just sticking with aortic aneurysms for a moment, how how is that going?

68
00:06:22,690 --> 00:06:26,860
And so the main challenge with that, as with many bits of research, is,

69
00:06:26,890 --> 00:06:35,230
was the data say we developed an algorithm which I picked out certain things and it may have predicted progression,

70
00:06:35,230 --> 00:06:37,780
but the challenge with A.I. as you really need to go and validate it,

71
00:06:37,780 --> 00:06:41,920
much like having a truck that works in the lab if I need to go out and run the clinical trial.

72
00:06:42,280 --> 00:06:45,339
And I think that the sticking point at the time was,

73
00:06:45,340 --> 00:06:52,510
was getting access to data from other centres and trust hospital groups outside of the UK to show that this works.

74
00:06:52,510 --> 00:06:56,860
And I think it's hard to know without that kind of validation.

75
00:06:56,860 --> 00:07:02,620
You ready? You really need to go out and try it and see if it works so that that that was where, where we got up to when when COVID came along.

76
00:07:02,680 --> 00:07:05,919
Right. Okay, so let's that we've now reached with coming along.

77
00:07:05,920 --> 00:07:13,809
Can you remember where you were when you first became aware that COVID was going to be something that was really quite serious?

78
00:07:13,810 --> 00:07:18,879
And and at one point you thought this was something that your region of research was.

79
00:07:18,880 --> 00:07:22,150
Area of research was something that might contribute. Hmm.

80
00:07:22,270 --> 00:07:28,600
I'm at the time and I was working as a junior doctor at the adult intensive care unit at the John Ratcliffe.

81
00:07:29,380 --> 00:07:33,520
And I, to be honest, I quite enjoyed it. I thought it was a it was a very lovely meditation.

82
00:07:33,520 --> 00:07:34,389
Everyone was very kind.

83
00:07:34,390 --> 00:07:43,780
And I remember in my last in my last month, we were ushered into a room for what they called high consequence infectious disease training.

84
00:07:43,990 --> 00:07:47,170
And at the time for us, COVID was something that month was this, do you think?

85
00:07:47,350 --> 00:07:50,380
And this must have been it must have been February 2020.

86
00:07:51,250 --> 00:07:58,360
And at the time, at the very beginning of February, perhaps at the time, COVID was something that was on the news and felt a long way away.

87
00:07:58,360 --> 00:08:06,129
And it didn't feel real for us just yet in the UK. And and I remember we went through our PPE training.

88
00:08:06,130 --> 00:08:07,450
This was personal protective equipment.

89
00:08:07,450 --> 00:08:16,060
And in the very first instance, and particularly for the intensive care unit, the protocol was was was very stringent.

90
00:08:16,360 --> 00:08:22,750
You had multiple layers of PPE, for example, three pairs of gloves and two lots of face coverings.

91
00:08:22,760 --> 00:08:26,890
And every every inch of you was covered at least twice. So it was quite something.

92
00:08:27,760 --> 00:08:34,300
It really was quite something. And I think that was the first time I thought, oh, gosh, this, this, this really is coming, this is going to happen.

93
00:08:34,870 --> 00:08:40,419
And and as part of that rotation we had, we used to cover the acute medical take.

94
00:08:40,420 --> 00:08:46,840
This is acutely among patients in the emergency department and the medical department at the front door.

95
00:08:48,100 --> 00:08:54,660
And I think the first time that that. The scale of the challenge that was coming really hit me was was.

96
00:08:55,730 --> 00:09:00,920
A couple of weeks before lockdown was announced. It was it must have been either the end of February or the start of March.

97
00:09:01,400 --> 00:09:06,920
And all these weapons, a junior doctor as part of a team run by one of the infectious diseases specialists.

98
00:09:07,460 --> 00:09:14,300
And this was shortly after the guidance had been issued about testing people coming into hospital who had respiratory symptoms.

99
00:09:14,570 --> 00:09:20,690
So initially it was based on if you travelled to certain regions, but then that evolved into anyone who had a cough, fever or shortness of breath.

100
00:09:22,040 --> 00:09:25,999
And the difficulty was that a very high proportion of our patients had a cough fever.

101
00:09:26,000 --> 00:09:30,799
So it was a breath, it's a hospital and the tests had to go off, I think,

102
00:09:30,800 --> 00:09:38,390
to Portsmouth at the time they were sent by road to a specialised laboratory and as a result, turnaround times were 3 to 4 days.

103
00:09:38,390 --> 00:09:41,900
And in that period of time, in those 3 to 4 days, you don't know.

104
00:09:41,900 --> 00:09:46,400
You've just you've just got someone who has a cough fever.

105
00:09:46,410 --> 00:09:49,549
We didn't really understand the clinical trajectory of COVID,

106
00:09:49,550 --> 00:09:55,070
so you couldn't even really use the clinical history and contact tracing was patchy at best at the time.

107
00:09:55,520 --> 00:10:01,340
And so we found ourselves very quickly on an evening shift with a lot of patients who needed

108
00:10:01,490 --> 00:10:06,260
a COVID result before they can be moved out of the acute part of the hospital down towards.

109
00:10:06,590 --> 00:10:13,430
But we knew that result wasn't coming for several days and I think that that's when it started.

110
00:10:13,430 --> 00:10:19,639
I was thinking we have a reasonable amount of routinely collected data, exactly the same types of data.

111
00:10:19,640 --> 00:10:27,469
We worked with the abdominal landing project and that data is already available within one hour, but pathway to do that is very well established.

112
00:10:27,470 --> 00:10:35,360
Every patient gets sits at the front door and whether we can use that data to predict a COVID test result many hours,

113
00:10:35,360 --> 00:10:38,690
many days in advance that back then when you when you got the results.

114
00:10:39,800 --> 00:10:46,070
So what were the the key pieces of data that get collected at the front door when a sick patient comes in?

115
00:10:46,700 --> 00:10:53,899
So when you come in, you you're initially met by by a member of my staff in the nursing staff, but sometimes sometimes medical staff,

116
00:10:53,900 --> 00:10:58,790
depending on how severe your condition is and some basic details will be taken, your name,

117
00:10:58,790 --> 00:11:03,260
address, contact details, and a very brief summary of your reason for coming to hospital.

118
00:11:03,560 --> 00:11:08,180
Then often patients quickly proceed to have their blood taken and their vital signs recorded.

119
00:11:08,570 --> 00:11:11,360
And these blood tests are in technical language.

120
00:11:11,360 --> 00:11:17,329
They have a full blood count, the urine crap and the electrolytes, liver function and C-reactive protein.

121
00:11:17,330 --> 00:11:22,370
These are kind of a very standard panel of blood tests that most patients,

122
00:11:22,370 --> 00:11:25,700
if not all patients, being admitted to every hospital in this country will end up having.

123
00:11:25,760 --> 00:11:29,089
And what do they tell you? They are a general health check-up.

124
00:11:29,090 --> 00:11:32,330
They tell you things like the oxygen carrying capacity of your blood.

125
00:11:32,600 --> 00:11:38,899
They look at your kidneys and how your kidneys are functioning, your liver and high levels functioning, and then some markers of inflammation.

126
00:11:38,900 --> 00:11:43,700
So is there any evidence of an angry immune response in your body?

127
00:11:44,150 --> 00:11:48,799
And in total, there are about 40 to 50 different data points within those.

128
00:11:48,800 --> 00:11:52,940
So there are kind of subtests. Each one of the subtests tells you something,

129
00:11:52,940 --> 00:11:57,589
and we know a little bit about each one of the subtests and then you go heart rate and that's where you come in to that.

130
00:11:57,590 --> 00:12:02,989
Yes. Your vital signs are that's your your heart rate, blood pressure, temperature, respiratory rate.

131
00:12:02,990 --> 00:12:08,360
And this, again, pretty standard set of measurements. And together, it's a lot of data points.

132
00:12:08,360 --> 00:12:14,540
And we as clinicians, we're trained to to skip over. Then we'll kind of look at the important ones and look for grossly abnormal values.

133
00:12:14,900 --> 00:12:18,110
But the normal range actually covers all manner of things.

134
00:12:18,110 --> 00:12:25,759
It includes a huge amount of variety and trying to unpick a signal within results that are normal for the general population,

135
00:12:25,760 --> 00:12:31,219
but not necessarily normal for that patient, I think is something difficult to do in a busy clinical shift.

136
00:12:31,220 --> 00:12:36,350
And this is really where the kind of advanced pattern recognition of of narrow AI comes into effect.

137
00:12:36,400 --> 00:12:43,640
It's where I can look at these data points in a much in a way that's much more granular than we can as humans.

138
00:12:44,450 --> 00:12:49,879
And and then often after these data have been collected, typically you'll have the blood test back in an hour,

139
00:12:49,880 --> 00:12:51,980
maybe an hour and a half if you're in a similar hospital,

140
00:12:52,670 --> 00:12:58,490
and your vital signs take about 1 to 2 minutes to collect, and after that, you would wait to see a doctor,

141
00:12:58,930 --> 00:13:04,759
you'd be assessed and a clinical decision would be reached about what your treatment plan would be.

142
00:13:04,760 --> 00:13:06,169
And that might include scans.

143
00:13:06,170 --> 00:13:16,639
It might include different tests, tracings of your heart, X-rays, admission to hospital, or perhaps going home, just depending on what the rest was.

144
00:13:16,640 --> 00:13:24,440
And I think the approach we were taking is, is there enough in that data, these bits of data collected in the first couple of hours,

145
00:13:24,440 --> 00:13:29,629
is there enough there for us to make a meaningful decision about your COVID status without needing

146
00:13:29,630 --> 00:13:35,959
to wait the three days at the time that unfortunately now it's more like 8 to 12 hours in hospital,

147
00:13:35,960 --> 00:13:40,130
but even 8 hours in an emergency department is a long time when an.

148
00:13:41,360 --> 00:13:46,519
But that that band, that was what we were trying to achieve. So you've used the royal way like people do all the time.

149
00:13:46,520 --> 00:13:51,800
But was it. Was it your was it your idea to to because you you were there at the front line.

150
00:13:51,810 --> 00:13:55,400
Was it was it you went to your colleagues and said, can we do something with this?

151
00:13:55,580 --> 00:14:02,420
Yes. I say, right, yeah. I remember sending an email tape now professionally about the the abdominal aortic aneurysm project.

152
00:14:02,440 --> 00:14:10,580
And David was copied in and I floated the idea and David immediately was it was immediately supportive as he always is brilliant.

153
00:14:10,850 --> 00:14:16,990
He immediately kind of jumped to help. And David Knight and David Clifton and I had a chat and and the rest went from that.

154
00:14:17,000 --> 00:14:21,680
And with David, we developed the idea. We applied for the ethics and we we took it for it.

155
00:14:22,160 --> 00:14:26,360
Did you have to apply for extra funding to work on it? And so initially, no.

156
00:14:26,360 --> 00:14:35,470
So I think we were in a position that there are already research databases within the university that straight data from the hospital.

157
00:14:35,480 --> 00:14:44,750
So we were able to sort of piggyback on on existing routes within the UK and David's lab already has the infrastructure in terms of computers

158
00:14:45,110 --> 00:14:53,570
and in times of time I already had part of my job delineated for research for that obviously with COVID that we've got slightly eroded.

159
00:14:53,990 --> 00:14:58,100
And so initially we were able to focus actually for next to no additional cost.

160
00:14:58,550 --> 00:15:02,660
And it's not it's not particularly taxing to develop an algorithm,

161
00:15:02,660 --> 00:15:08,000
really the challenges and then taking this algorithm and going out and checking it works at other hospitals.

162
00:15:08,210 --> 00:15:13,190
And that was the idea. I mean, does David have developers or software people in his lab?

163
00:15:13,550 --> 00:15:18,500
I mean, who actually writes the code? And everyone in David's life as a case, including myself, I chaired that.

164
00:15:18,530 --> 00:15:20,780
Yes. It's one of the things that I've done since I was a kid.

165
00:15:21,140 --> 00:15:29,090
So so I suppose there's a difference between someone who can code for research purposes and someone who can code for industry grade.

166
00:15:29,420 --> 00:15:33,840
You'd put this on an aeroplane purposes. We're all coders, but for scientific purposes.

167
00:15:34,220 --> 00:15:39,920
And if people are able to do a basic to meeting standard of code, but that's distinctly different from someone who,

168
00:15:39,920 --> 00:15:43,130
for example, might write the software that runs your hospital or your aeroplane.

169
00:15:43,550 --> 00:15:49,129
And, and, and all of the code was written in-house by by myself and by by others in the lab.

170
00:15:49,130 --> 00:15:53,740
And. And that that allowed us to build this very cheaply and very quickly.

171
00:15:54,370 --> 00:15:59,949
And we Dave and I applied for a grant fund from the It's Money,

172
00:15:59,950 --> 00:16:06,550
which comes from the Wellcome Trust and then cascades through the university called the Medical and Life Sciences Translational Fund.

173
00:16:06,970 --> 00:16:13,690
And that was the money that we then used to do the next steps of the study and pay for the validation and pay for the the deployment to the jail.

174
00:16:15,310 --> 00:16:18,850
So what was the strategy for the validation? Let's take it step by step.

175
00:16:19,480 --> 00:16:25,480
I mean, so I suppose the research questions are we we think we've developed an air test that works.

176
00:16:25,720 --> 00:16:29,980
We have prospectively tried it on data from Oxford and we've shown it works.

177
00:16:30,190 --> 00:16:34,360
But will it work in the second wave? Will it work in other hospitals in the UK?

178
00:16:34,690 --> 00:16:39,210
Will it work as well in hospitals in the UK? And can you make it better?

179
00:16:39,220 --> 00:16:43,150
Can you make it faster? And can we just. Sorry, go back to the beginning of that.

180
00:16:43,150 --> 00:16:48,040
You said you'd prospectively do it, so you were actually applying it to the patients that were coming into the job?

181
00:16:48,070 --> 00:16:52,870
Yes, but not not in a way that was initially not in a way that was being used for that clinical care.

182
00:16:52,960 --> 00:16:56,200
So initially we were doing it using the data collected for those patients,

183
00:16:56,200 --> 00:17:01,450
but not in that we weren't using the result to make a decision on that care.

184
00:17:01,480 --> 00:17:05,080
Right. So was it not in real time with this? Well, you know, initially it wasn't.

185
00:17:05,200 --> 00:17:10,150
It was data that had been previously collected when the patients came in to us. But you just worked with that data.

186
00:17:10,180 --> 00:17:15,670
Yeah. Okay. And we'd be then wanted to take it from there and kind of take it forwards.

187
00:17:15,690 --> 00:17:25,560
Yes. And so I think the first question to us is. Can we can we make it faster, say the kinds of data points we used, typically available in one hour,

188
00:17:25,570 --> 00:17:32,340
but some subsets of it maybe took an hour, 15 or an hour 20 and were done for nearly all patients instead of all patients.

189
00:17:32,670 --> 00:17:35,909
And so we cut out quite a few of the bits and bobs that went into this.

190
00:17:35,910 --> 00:17:41,520
And we we really cut it down to a very minimal set, have a blood test that everyone gets.

191
00:17:42,090 --> 00:17:47,310
And then we thought, actually, can we go even further? Can we cut it down to blood tests that don't even need a lab to get?

192
00:17:48,090 --> 00:17:51,989
So we started off with this big panel and we then took it down to your full blood count,

193
00:17:51,990 --> 00:17:55,740
your kidney function, liver function, inflammatory markers and your vitals.

194
00:17:56,070 --> 00:18:01,980
And then we tried taking up kidney function, liver function and inflammatory markers and really went for just your full lockdown vitals.

195
00:18:03,180 --> 00:18:08,879
And what was nice about that combination is you can get a device that's about the size of a shoebox.

196
00:18:08,880 --> 00:18:14,550
It's called the know. It's from an Israeli start-up and that we just started using in Oxford.

197
00:18:15,000 --> 00:18:17,360
And that will give you those results in 10 minutes.

198
00:18:17,400 --> 00:18:22,290
And so you went from a test that might take one hour to collect the data, and I can get the data in 10 minutes.

199
00:18:22,770 --> 00:18:27,840
And so we started developing sub fashions and each of which was meant to be faster than the last.

200
00:18:28,410 --> 00:18:32,489
And then we thought, okay, we've got our fashion, we've got some fashion. Let's go and see if this actually works.

201
00:18:32,490 --> 00:18:39,059
And we had we're very fortunate. We had collaborators and wonderful people elsewhere who were interested in helping.

202
00:18:39,060 --> 00:18:45,629
And we we worked with the university hospitals, Birmingham, Portsmouth hospitals and and Bedford as well as,

203
00:18:45,630 --> 00:18:50,460
of course, for the second wave here in Oxford, and then took our test and applied it on those data.

204
00:18:50,470 --> 00:18:55,830
And what we found is it performed just as well in all of these centres as it did for us, and that's a big thumbs up.

205
00:18:55,830 --> 00:19:01,410
It says we're not just detecting something that's Oxford specific or Oxford special or we're just doing something funny here.

206
00:19:01,620 --> 00:19:06,719
This is this is a signal that replicates. And that's really important for you to trust your algorithm.

207
00:19:06,720 --> 00:19:13,950
You need to know what's going to work when you try it elsewhere. And had you given it a name by this point?

208
00:19:14,010 --> 00:19:18,720
Yes. So can you explain the name? So we ended up calling it Curio.

209
00:19:18,750 --> 00:19:25,350
And the way it came around is a funny story. Back back in the very early days of the project, we had lots of paperwork to fill in, ethics,

210
00:19:25,350 --> 00:19:31,020
etc. and one of the boxes on the ethics form required you to give it a short

211
00:19:31,050 --> 00:19:36,240
title to something you can use throughout the phone to tape and to refer to it.

212
00:19:36,240 --> 00:19:45,540
And at the time it was called, I think, COVID 19 machine learning, undifferentiated illnesses would be quite a technical, quite a long fright.

213
00:19:45,540 --> 00:19:49,050
It's not copied and pasted it into one of these online acronym generators.

214
00:19:49,470 --> 00:19:54,299
And it came up with that and it selects it's not it's not a pure acronym and

215
00:19:54,300 --> 00:19:57,860
that some of the letters in the middle are selected from the middle words.

216
00:19:58,240 --> 00:20:05,190
And it's a little bit like a clinical trial sometimes is a bit of a tenuous acronym, but I thought it's a nice word,

217
00:20:05,310 --> 00:20:12,390
it reads easily and it's stuck and it has connotations of something to do with the Catholic Church, doesn't it?

218
00:20:12,420 --> 00:20:23,400
Curia and I found this out. I found this house about a year later when we realised that curial com we were looking at web domain names.

219
00:20:23,790 --> 00:20:32,159
The web domain was owned by the European Court of Justice and it was the name of an app they used as a filing system for legal documents.

220
00:20:32,160 --> 00:20:35,280
But we didn't we didn't do that.

221
00:20:35,280 --> 00:20:40,589
But everybody back in 2020 anyway, sorry, I interrupted.

222
00:20:40,590 --> 00:20:40,920
So yeah.

223
00:20:40,930 --> 00:20:51,120
So you gather data again doing this using existing data that had been gathered, but not using it in anger as it were, to, to make the diagnoses.

224
00:20:51,420 --> 00:20:59,170
Yeah. In these other hospitals say so to use it to make a diagnosis, you would need to get some sort of approval from the regulatory authorities.

225
00:20:59,280 --> 00:21:07,439
So pre-COVID it would have been CE marking with Brexit, it was going to be Ukca marking and it's quite an expensive process to go through.

226
00:21:07,440 --> 00:21:10,710
It only takes a year to two years.

227
00:21:10,950 --> 00:21:18,300
I think if you're quick and in the heat of the moment, we didn't have the resources or the funding or the time to go through that.

228
00:21:18,600 --> 00:21:23,879
Now, the only other thing you need, of course, is proof that it works fast.

229
00:21:23,880 --> 00:21:31,860
So you have to do the study before you can before you can do that. And the MHRA have got some special pathways in place.

230
00:21:31,860 --> 00:21:40,830
The MHRA are the licensing authority for this in the UK and these special pathways allow you with some data to to get an expedited approval for COVID.

231
00:21:41,190 --> 00:21:46,409
And a lot of the original COVID tests that came into use and on the market came through this pathway.

232
00:21:46,410 --> 00:21:53,250
So lateral flows, for example, initially came via this pathway and only later picked up their full regulatory approval.

233
00:21:54,320 --> 00:22:01,320
And so to us that that was an important part of actually being able to use it in the anger of the moment.

234
00:22:01,320 --> 00:22:06,210
And that's why initially we we had to do this with the retrospective data festival, get the proof of works then,

235
00:22:06,600 --> 00:22:11,430
then try and try and get some sort of authority approval to use it in the real world.

236
00:22:13,280 --> 00:22:16,950
And that's that has happened. So the next step.

237
00:22:16,970 --> 00:22:24,380
Yes, the next step was this. And we had some very wonderful colleagues who work in the emergency department.

238
00:22:24,980 --> 00:22:31,010
And it's a group of consultant emergency physicians with a particular interest in research.

239
00:22:31,010 --> 00:22:34,690
And they run a research organisation in Oxford called M Rocks.

240
00:22:36,380 --> 00:22:42,530
How do you spell that E M or X emergency medicine research in which is not in the acronym Oxford?

241
00:22:42,620 --> 00:22:48,919
Yes. So another one of these slightly tenuous acronyms. And and they were they were wonderful.

242
00:22:48,920 --> 00:22:57,220
They're very supportive. They wanted to see this news. And we we got local approval in Oxford to put one of these machines into practice.

243
00:22:57,230 --> 00:23:04,010
And this is the shoebox sized machine into practice. And to put the algorithm into practice and then run from the results in real time.

244
00:23:04,010 --> 00:23:09,079
And we when we did the study, we needed a bit of help.

245
00:23:09,080 --> 00:23:13,670
We wanted people on the ground to help run these point of care samples to give you the quickest results.

246
00:23:14,090 --> 00:23:20,330
And for that, we called on the wonderful medical students of the University of Oxford who were brilliant,

247
00:23:20,350 --> 00:23:25,489
they worked hard, they were enthusiastic, and they were helping out the clinical staff as well.

248
00:23:25,490 --> 00:23:33,080
So they would, for example, when a patient would come in, they would help with taking the blood, which is one job for the clinical teams to do.

249
00:23:33,350 --> 00:23:37,340
They would take the blood, run the sample, and then send off the blood to the lab.

250
00:23:39,870 --> 00:23:44,459
And we then had to look at various metrics and we compared it to lateral flows.

251
00:23:44,460 --> 00:23:49,710
And what we were able to show is that with this box on a desk and the algorithm,

252
00:23:49,980 --> 00:23:55,320
we could get results in 45 minutes instead of an hour and one minute from the front door for lack of light.

253
00:23:55,410 --> 00:24:03,000
Now I appreciate lateral takes, take 30 minutes, but this is including the time from you first setting foot in the department up to having a result.

254
00:24:03,000 --> 00:24:10,079
So that's the time for you to find a bed and for the swap to actually be taken for the swap to then be to then be

255
00:24:10,080 --> 00:24:17,460
run and we were all test was able to do in 46 minutes lateral place and one hour one minute from door to result.

256
00:24:17,850 --> 00:24:24,650
And that that was a big change because the lot of the PCR test still took another 8 hours to give you a result.

257
00:24:24,660 --> 00:24:28,260
So in the heat of the moment, that's very helpful. What's it say?

258
00:24:28,770 --> 00:24:31,890
And it just helps you move the hospital a bit quicker.

259
00:24:32,190 --> 00:24:37,950
And I think the worry with lateral places that that there are some concerns about how sensitive they are.

260
00:24:38,490 --> 00:24:46,860
And I appreciate it's one of these very controversial topics of the day and with various groups saying X, but a group saying why?

261
00:24:47,250 --> 00:24:51,540
And we had to look at how lateral flow is performed at our hospitals in Oxford.

262
00:24:51,540 --> 00:24:55,169
And we found that they they were 56.9% sensitive.

263
00:24:55,170 --> 00:25:00,209
And what that means is they miss 43 out of every 100 PCR positive COVID patients.

264
00:25:00,210 --> 00:25:04,260
We admitted that while that might be alright in the community,

265
00:25:04,500 --> 00:25:09,750
in a hospital where you've got the most vulnerable patients, that that's not not really good enough.

266
00:25:10,200 --> 00:25:14,219
And we wanted to see can all test give you more confident negative results.

267
00:25:14,220 --> 00:25:18,900
Can we be sure that an a negative from all tests means you don't have COVID?

268
00:25:18,900 --> 00:25:22,140
And fortunately the data showed that and we were very pleased with that.

269
00:25:22,410 --> 00:25:27,090
So they were all these patients were getting PCR tests as well. So you had something to check against?

270
00:25:27,090 --> 00:25:31,140
Yes. And by the time we got to this, everyone was getting PCR as they would done in Oxford.

271
00:25:31,350 --> 00:25:35,080
And so it was taking 8 to 12 hours instead of three days as it wasn't.

272
00:25:35,310 --> 00:25:38,510
And in 2020. And what?

273
00:25:38,580 --> 00:25:42,210
Sorry. I think you may have just said it and I missed it, so I'll just ask you to say it again.

274
00:25:42,570 --> 00:25:48,300
How did your application compare with PCR for accuracy?

275
00:25:48,660 --> 00:25:56,880
And so we were comparing our results to PCR, which means we rather the PCR was all kind of auteurist that was our gold standard.

276
00:25:57,210 --> 00:26:01,050
So we could only benchmark it against the PCR to compare it with the PCR,

277
00:26:01,500 --> 00:26:09,300
but we compared it with the with the lateral flow test and we showed that we were we were more sensitive in that using a combination of the two,

278
00:26:09,570 --> 00:26:12,630
we could cut missed COVID cases by over 70%.

279
00:26:13,440 --> 00:26:18,780
So for us, we saw our competitors the lateral flow test. Right. And then then the PCR tests.

280
00:26:18,790 --> 00:26:28,439
Yeah. And where what we're kind of thinking in the longer term is we're now moving to a phase where we're starting to live with COVID and and testing.

281
00:26:28,440 --> 00:26:32,129
Everyone who comes into hospital by PCR is actually quite expensive.

282
00:26:32,130 --> 00:26:34,380
It's quite labour intensive, it's extra clinical work.

283
00:26:35,490 --> 00:26:44,760
And we're now thinking about how can we use our algorithm on everyone who comes in to try and pick out which populations we should be testing.

284
00:26:45,120 --> 00:26:50,160
So rather than testing everyone and then getting back, maybe 2% of our results are positive.

285
00:26:50,580 --> 00:26:59,370
Can we pick out a population of five or 10% and say we think 50% of this subgroup will test positive, so why don't we test these people as a priority?

286
00:26:59,640 --> 00:27:04,860
And then if the results announced, we can then test the rest. But we think very few of them, if any, will test positive.

287
00:27:06,530 --> 00:27:12,620
So what's what's and I mean, I've noticed looking at your paper that you've got various sort of sub names of curios,

288
00:27:12,620 --> 00:27:17,300
this one Korean repeat in carrier lab. So what, what are those three different things.

289
00:27:17,510 --> 00:27:24,110
I say Korea. One is the the original Korea. This was the first version of Caro from from March 2020.

290
00:27:24,140 --> 00:27:28,820
It was probably it must have been made by the time it was ready to go.

291
00:27:29,120 --> 00:27:33,209
So that was Korea, one that used all of the data that we can get.

292
00:27:33,210 --> 00:27:38,960
So that was a full panel of plots and your vital signs. Korean lab was then taking out the blood test.

293
00:27:38,960 --> 00:27:47,080
But not everyone not quite everyone got. And the ones that took slightly longer say you're coagulation, for example, typically takes an extra 15,

294
00:27:47,090 --> 00:27:56,660
20 minutes and then cure repeat is the version which just uses you for blood count and your vitals and it can run on this box at the box office.

295
00:27:58,160 --> 00:28:03,500
And so what's the position at the moment? What's and is it still being used in the hospital.

296
00:28:03,620 --> 00:28:11,480
So position today and for us to get this wider embedded into the NHS, we need to get some sort of regulatory,

297
00:28:11,930 --> 00:28:14,540
full regulatory approval and that's going to take quite a bit of funding.

298
00:28:14,540 --> 00:28:22,630
So David Clifton and I are speaking with the university's Tech Transfer Office, that's Oxford University Innovation Hub,

299
00:28:22,700 --> 00:28:27,109
very kindly putting us in touch with investors and we're looking at forming a Start-Up Company out of the

300
00:28:27,110 --> 00:28:35,000
university and trying to take the quality of technology into every hospital in this country and hopefully beyond,

301
00:28:35,630 --> 00:28:40,610
and be whether we can use our methods and use our approach for something slightly different.

302
00:28:40,610 --> 00:28:44,179
So, for example, we've done this for COVID. Why can't we do this for free?

303
00:28:44,180 --> 00:28:49,280
Why can't we do this for sepsis? And the idea would be to take the same same approach, same technology, same A.I.,

304
00:28:49,520 --> 00:28:55,190
but train it differently for flu and then start providing flu screening.

305
00:28:55,200 --> 00:29:01,669
So that that's kind of the longer term vision. But that's going to need it will need a team funding resources.

306
00:29:01,670 --> 00:29:07,850
And we're hoping to be able to do that by and by looking at the start up mechanism.

307
00:29:07,970 --> 00:29:11,840
Hmm hmm hmm hmm. Well, that's all very exciting.

308
00:29:12,020 --> 00:29:16,400
But you you did sort of mention you are a junior doctor working in a hospital.

309
00:29:16,670 --> 00:29:18,740
So all the time, all this development was going on,

310
00:29:19,130 --> 00:29:25,580
presumably you had responsibilities as a clinician in a hospital that was coping with the pandemic.

311
00:29:26,420 --> 00:29:31,100
That's right. It's been busy. It's been a busy couple of years.

312
00:29:31,370 --> 00:29:35,120
I think there are several reasons why it's worked out.

313
00:29:35,120 --> 00:29:41,959
So, first of all, Oxford is brilliant. We've had so much support and my myself, my job as a is at clinical academic roles.

314
00:29:41,960 --> 00:29:46,820
So we have a portion of our time safeguarded for academics, even during the pandemic.

315
00:29:47,060 --> 00:29:48,770
Well, not not during the height of COVID,

316
00:29:48,770 --> 00:29:55,430
but but in the kind of the have been some lulls in the COVID workplace during those levels, you've been able to get a bit of time.

317
00:29:56,300 --> 00:30:05,180
Secondly, we've been very fortunate in Oxford as a whole in that we're we're a hospital with we have our own clinical staff and through brilliant,

318
00:30:05,180 --> 00:30:13,669
through exceptionally good. But we also have a lot of researchers who are clinically trained and are working in research topics of all varieties.

319
00:30:13,670 --> 00:30:15,140
And so when when the pandemic hit,

320
00:30:15,530 --> 00:30:22,939
a good number of researchers who were clinically trained could come back into clinical practice here, and they were wonderful support.

321
00:30:22,940 --> 00:30:26,090
So we had more numbers here than we did elsewhere, I think.

322
00:30:27,570 --> 00:30:33,030
On balance, we were very fortunate here. We had we had two people and and that that meant that.

323
00:30:34,350 --> 00:30:42,150
We relatively things relatively weren't as bad in Oxford as they were in, say, deprived parts of London or in other inner cities.

324
00:30:43,700 --> 00:30:45,499
That question I'm going to ask earlier.

325
00:30:45,500 --> 00:30:57,620
I mean, are there are there any risks associated with shifting to an AI approach as opposed to the 30 year experience looking at the data?

326
00:30:58,100 --> 00:31:04,249
So yes, and what a large part of our work is looking at how do we mitigate those risks?

327
00:31:04,250 --> 00:31:08,570
How do we how do we make sure that we're adding value without causing harm?

328
00:31:09,050 --> 00:31:15,500
And every medical test has a risk. So every medical test is imperfectly sensitive or is imperfectly specific.

329
00:31:15,500 --> 00:31:20,420
It's not it either misses cases or it picks up too many or it takes too long, or it just doesn't work.

330
00:31:20,900 --> 00:31:26,900
And I think the best that you can do is to compare your test to what the best thing in use out there is.

331
00:31:26,900 --> 00:31:30,920
What's your competitor? What are you actually trying to replace? And in all cases,

332
00:31:31,850 --> 00:31:38,299
COVID is known as that can give you no symptoms at all or can give you very mild symptoms or any any one over a long list of symptoms.

333
00:31:38,300 --> 00:31:43,400
And so your ability as a clinician to say a patient doesn't have COVID is actually quite slim.

334
00:31:43,850 --> 00:31:50,120
And during the height of, say, the second wave one in maybe, you know,

335
00:31:50,120 --> 00:31:53,690
as much as one in ten of your patients will have had COVID and you may not know about it.

336
00:31:53,900 --> 00:32:00,560
Whereas in the lull of the summer beforehand, COVID was actually quite uncommon, and maybe one in 200 of your patients will have had COVID.

337
00:32:00,830 --> 00:32:03,940
And so clinical symptoms alone isn't enough.

338
00:32:03,950 --> 00:32:08,659
You need something more because people can have no symptoms. And that's where our test result comes in.

339
00:32:08,660 --> 00:32:12,940
Very helpful. And you can design how you use a test to make sure you don't do harm.

340
00:32:12,950 --> 00:32:19,910
Say, for example, what we what we said in some part of our study is if if curious as you haven't got COVID,

341
00:32:20,270 --> 00:32:24,710
but a lateral flow test as you do, then we think you should act as though that person does have COVID.

342
00:32:25,010 --> 00:32:32,820
And in doing that, you make sure that there's you can't possibly be more dangerous than than the current standard of care, if that makes sense.

343
00:32:32,840 --> 00:32:34,430
So, yes, there are risks,

344
00:32:34,430 --> 00:32:40,670
but a lot of clinical data is about mitigating those risks and trying to at least show that you're better than what we have in use today.

345
00:32:42,320 --> 00:32:56,809
Thank you. Yes, I'm glad I remembered to get that. And is I mean, is this an approach unique to do Oxford at the moment,

346
00:32:56,810 --> 00:33:04,520
or are other similar approaches being developed elsewhere that you're aware of in the world or or in the country?

347
00:33:04,850 --> 00:33:08,479
And so climate change is very hot topic.

348
00:33:08,480 --> 00:33:10,310
Everyone's interested in it, very important.

349
00:33:11,140 --> 00:33:20,120
And broadly you've got that the Asian approaches to COVID and the non-elite, the majority and on the lateral place, the most common of the approaches.

350
00:33:20,360 --> 00:33:28,309
There are other groups out there who have published similar things to our study say so I the kind of in the years since our initial study,

351
00:33:28,310 --> 00:33:32,960
there have been a handful of papers that have tried to replicate it and they've all been able to replicate it.

352
00:33:33,200 --> 00:33:37,040
And in science, that's a good thing. It suggests that we're on to something here.

353
00:33:37,460 --> 00:33:42,440
And in terms of actually getting it into clinical use, I think we're the furthest ahead by some distance.

354
00:33:42,440 --> 00:33:47,180
I think the other groups at least have not stated an intention of trying to get this into practice.

355
00:33:48,920 --> 00:33:53,809
And I think that that's part of our case for they're trying to spin this out as a company.

356
00:33:53,810 --> 00:34:01,730
We think we think we're on to something. We think we're well, we hope you know, we hope it's it's it's going to work.

357
00:34:01,730 --> 00:34:05,090
And we hope that we've got a bit of a competitive advantage.

358
00:34:05,090 --> 00:34:09,919
And that's why we're hoping that a start up might might allow us to to put this into ideas.

359
00:34:09,920 --> 00:34:13,940
But no, there aren't there aren't others. So using something like this clinically.

360
00:34:14,080 --> 00:34:20,630
Mm hmm. Good. Right. So I'm going to go to sort of more slightly more personal questions.

361
00:34:23,250 --> 00:34:25,139
Well, I've got you know, I think we've covered that.

362
00:34:25,140 --> 00:34:29,340
I was going to say, how did the first lockdown impact on what you're able to do, which is a question I've been asking everybody.

363
00:34:29,340 --> 00:34:37,200
But you're a doctor working in a hospital, right? Basically, it meant putting on three layers of PPE and getting on with it, presumably.

364
00:34:37,470 --> 00:34:42,660
Yes. I think, again, this comes back to us having, you know, excellent clinical leadership in Oxford.

365
00:34:42,660 --> 00:34:47,340
The guys who were running the show in Oxford made sure we had the PPE.

366
00:34:47,340 --> 00:34:50,720
They you know, everyone was trying very hard. It was it was difficult.

367
00:34:50,730 --> 00:34:54,150
But in Oxford, we do all that.

368
00:34:54,150 --> 00:34:57,900
We do. I you know, I definitely felt that we were protected in Oxford.

369
00:34:57,900 --> 00:35:02,459
I felt like we had we had the staff when we needed them. And sure, there were times when things were tough.

370
00:35:02,460 --> 00:35:07,890
But I can only think of people who were elsewhere who didn't have the kind of resource that we had here.

371
00:35:08,910 --> 00:35:16,470
Did you see colleagues falling sick and did that I mean, to what extent did you feel personally threatened by the virus or maybe you've had it?

372
00:35:16,670 --> 00:35:23,999
I mean, so I felt to prove that I've had it, I think from a statistic, just looking at it statistically,

373
00:35:24,000 --> 00:35:29,069
I think it's nearly, nearly impossible that I've not had it, but I think more than adequately exposed.

374
00:35:29,070 --> 00:35:33,030
And and I think I think we're looking at a very high percentage of the population having had it.

375
00:35:33,270 --> 00:35:37,040
I mean, of course, we were amongst the first to get vaccines in Oxford because because it's Oxford.

376
00:35:37,440 --> 00:35:42,630
And but yes, we did look after colleagues who were sick and.

377
00:35:43,960 --> 00:35:49,540
That though, there was some some who died. And I think that there were moments which were tough.

378
00:35:51,070 --> 00:35:55,059
And I'm mindful that there's a lot I can't say.

379
00:35:55,060 --> 00:36:00,580
But but there were moments where you when you were affected because there were people who you would

380
00:36:00,580 --> 00:36:04,059
have worked with or people you would have walked past in the car just with a shattered canteen with,

381
00:36:04,060 --> 00:36:14,080
you know, this was your trust, your staff. I think, fortunately, the majority of, you know, I'm young and so therefore I'm in a lower risk category.

382
00:36:14,080 --> 00:36:17,920
And we got vaccines early and we we were well protected here.

383
00:36:17,920 --> 00:36:26,799
But no, it was it was tough. And there's obviously the the kind of psychological impact of this, the constant uncertainty in some ways.

384
00:36:26,800 --> 00:36:34,959
Do you think that the fact that you had this part of the project to work on that you sound very positive about?

385
00:36:34,960 --> 00:36:38,770
It seems to have been quite optimistic as it went along.

386
00:36:39,160 --> 00:36:43,809
Do you think that helped to support your own wellbeing that you had that that focus 100%?

387
00:36:43,810 --> 00:36:49,020
And I think it's nice to feel like you're doing something and that you're doing something that can make a difference.

388
00:36:49,030 --> 00:36:55,560
And I think this was my way of lockdown amusement when when I wasn't at work and couldn't get out.

389
00:36:55,570 --> 00:36:59,740
Well, I had something to do. Yes.

390
00:36:59,740 --> 00:37:04,480
You've talked earlier about how your reason for going into medicine was probably different from what it turned out to be.

391
00:37:05,170 --> 00:37:09,700
And I wondered if that. I mean, most people seem to go into medicine because they want to help people.

392
00:37:09,940 --> 00:37:14,200
But is that is that what sort of came around full circle in the end?

393
00:37:14,680 --> 00:37:22,239
I think so. I think that I think what I'm realising more and more is that trying to there is a lot of that which there's a lot of things.

394
00:37:22,240 --> 00:37:26,889
But how much of it is in our day to day clinical use, how much do I actually use it in my day job?

395
00:37:26,890 --> 00:37:35,950
Well, maybe not. And as someone who is technically inclined, who loves tech and is medically trained,

396
00:37:36,430 --> 00:37:40,509
you find yourself in a position where you're actually you're really able to make a difference.

397
00:37:40,510 --> 00:37:47,530
You're really able to take this tech, which people who work in David's lab, you know, top of their game, they know their stuff, they make cool things.

398
00:37:47,800 --> 00:37:52,330
But getting it from cool thing in a lab to cool thing in clinical practice is is a mighty leap.

399
00:37:52,330 --> 00:37:59,530
And I think that's where being where I'm very glad to be a doctor because it helps me work on that leap from lab to real world.

400
00:37:59,560 --> 00:38:08,920
Mm hmm. So that's a very good point. Yeah. So I think I'm drawing to a close so to use partly and also this I think.

401
00:38:08,920 --> 00:38:13,930
But as the work you've done on Curiel raised new questions that you'd like to work on in the future.

402
00:38:15,050 --> 00:38:17,870
Oh, lots. I think I've learned to know mightier mountains.

403
00:38:17,870 --> 00:38:21,820
There are things you learned that you knew you were going to learn and things you learn that you would never.

404
00:38:22,100 --> 00:38:26,360
You know, I did with things such as, ah, you could you could be.

405
00:38:26,510 --> 00:38:36,530
Oh, so. So I think it's I've definitely engaged in the field of ethics and the practicalities of I deployment differently to how I would before,

406
00:38:36,530 --> 00:38:41,179
I think before I might have seen it as a little bit restrictive,

407
00:38:41,180 --> 00:38:44,570
something which makes it very difficult for you to get stuff into practise, which is good.

408
00:38:44,840 --> 00:38:47,239
I think part of doing this work is,

409
00:38:47,240 --> 00:38:53,780
is actually meeting the people who are interested in the field and you learn that actually it's a much more nuanced approach.

410
00:38:54,040 --> 00:38:58,910
This is a kind of very well thought out reasons for why things are the way they are,

411
00:38:59,420 --> 00:39:06,620
and learning about those ages has made me interested in how can you make this process work as well as possible between,

412
00:39:06,620 --> 00:39:12,230
you know, doing the bridges and and B, how can you engage with the process and actually genuinely deliver something that's better for everyone?

413
00:39:13,940 --> 00:39:17,540
I think I sort of want to do it again.

414
00:39:17,690 --> 00:39:28,100
I want to knock over it. I mean I mean, take take a clinical problem, apply some sort of machine learning technique can solve the clinical problem.

415
00:39:28,110 --> 00:39:31,639
And, um, yes, yes, yes.

416
00:39:31,640 --> 00:39:35,360
With, with the, the benefit of the experience that you've already got.

417
00:39:35,720 --> 00:39:46,640
Mm hmm. And how has the experience changed your attitude or your approach to your work and all the things you'd like to see change in the future?

418
00:39:46,790 --> 00:39:51,829
I mean, actually, one question I didn't ask, which I should have asked more specifically is about collaboration,

419
00:39:51,830 --> 00:40:02,059
because really it's a multidisciplinary approach. Did you find that this project involved wider collaboration than than might normally be the case?

420
00:40:02,060 --> 00:40:07,490
I mean, I know academic life can be tremendously competitive, but in this instance,

421
00:40:07,490 --> 00:40:11,960
a lot of people were trying to solve a problem very quickly and actually competition can get in the way of that.

422
00:40:12,350 --> 00:40:17,089
Yes. So what was unique about the circumstances in February-March?

423
00:40:17,090 --> 00:40:21,139
I think the regulatory authorities were taking an extremely pragmatic view.

424
00:40:21,140 --> 00:40:26,330
So for example, we didn't need to go through to the full six month ethical approval cycle.

425
00:40:26,330 --> 00:40:32,080
In fact, I remember one morning we submitted the ethics application to the MHRA in the afternoon

426
00:40:32,080 --> 00:40:35,890
and I had a phone call from a gentleman called Kevin who was very friendly and said, Right, we get to go.

427
00:40:36,380 --> 00:40:40,310
And which is that's is not what happens in standard times. It's often many months.

428
00:40:40,310 --> 00:40:49,160
And that that was that was because of the pandemic I think work like this needs you to collaborate with technical experts, be with clinicians,

429
00:40:49,160 --> 00:40:51,680
see with clinicians elsewhere to get data from elsewhere,

430
00:40:51,890 --> 00:40:56,600
and then deal with an almighty number of people in research services and university support roles.

431
00:40:56,840 --> 00:41:01,010
And without that kind of infrastructure, without all those people, you can't do it.

432
00:41:01,010 --> 00:41:03,200
You need everyone and everyone plays that role.

433
00:41:03,200 --> 00:41:09,740
And I think what, what sort of sometimes upsets me is sometimes people who you've worked with have been very helpful.

434
00:41:11,480 --> 00:41:15,890
They never find out. But because of what they did, this whole project has managed to work out in the end.

435
00:41:16,850 --> 00:41:23,389
And I think it probably certainly when I'm a junior doctor, I do a little, little role.

436
00:41:23,390 --> 00:41:29,090
For example, I can operate a computer or do some, some, quite some, some.

437
00:41:30,090 --> 00:41:34,139
Paperwork. And that paperwork may feel mundane at the time.

438
00:41:34,140 --> 00:41:39,930
It might feel like a small role. But in the bigger picture, it's important and it's a small piece of a big puzzle.

439
00:41:39,960 --> 00:41:43,590
And I think that's that's how I describe this. Everything is the small piece of the puzzle.

440
00:41:45,210 --> 00:41:48,090
But do you think actually there's a case for that?

441
00:41:51,980 --> 00:42:01,700
I don't know, Will to to get things working quickly or to survive after we are no longer facing the pressures of the pandemic.

442
00:42:02,210 --> 00:42:08,570
Yes. And this is a bit difficult in that I think.

443
00:42:10,440 --> 00:42:19,910
As someone who wants to see A.I. in clinical practice, a lot of the structures and barriers to adoption that they're very well described there.

444
00:42:20,100 --> 00:42:24,000
People have written quite long and well-thought-out pieces about what kinds of barriers you look at.

445
00:42:24,750 --> 00:42:28,590
Many of those barriers represent real, practical problems that you need to think about.

446
00:42:28,980 --> 00:42:33,719
But COVID has shown some very good examples of where when, when those barriers are waived.

447
00:42:33,720 --> 00:42:38,000
People can can bring innovation into the real world much more quickly.

448
00:42:38,010 --> 00:42:42,479
So I hope I hope you'll get to meet the team behind the recovery trial.

449
00:42:42,480 --> 00:42:45,660
But that that's a fantastic example of how something is dynamism.

450
00:42:45,660 --> 00:42:53,430
I think it's a fantastic example of how something has come into clinical practice so quickly and has made such a difference and saved so many lives.

451
00:42:53,910 --> 00:42:57,290
And undoubtedly a lower burden of regulation.

452
00:42:57,300 --> 00:43:04,320
A lower burden of of. Bureaucracy and paperwork would have allowed them to do that in that timeframe.

453
00:43:05,070 --> 00:43:11,370
And to what extent those the lessons from that are going to persist, I don't know.

454
00:43:11,610 --> 00:43:20,220
And whether we're going to see a systematic reduction in red tape looking at the successes from COVID or whether will double down and and.

455
00:43:22,440 --> 00:43:26,009
Maybe. I think it's hard to know. It's hard to know.

456
00:43:26,010 --> 00:43:31,260
But I'm sure there'll be lots of very bright people who are very interested in this and study

457
00:43:31,260 --> 00:43:35,850
it and will hopefully come up with some conclusions for us that we can we can take forwards.

458
00:43:36,960 --> 00:43:38,640
Brilliant. Thanks very much.