1
00:00:00,180 --> 00:00:07,509
I told her. But today I thank you and thank you, Emily and Margaret, for the invitation.

2
00:00:07,510 --> 00:00:12,730
Can everybody hear me? Great. So, yeah, really glad to be here.

3
00:00:13,450 --> 00:00:20,500
Really interesting how big data has taken off since I started do my stay over ten years ago now.

4
00:00:20,530 --> 00:00:29,920
So I'd like to spend the next 45 minutes just talking to you about what I've learned over that time about heart failure in primary care,

5
00:00:30,070 --> 00:00:34,570
in the context of those big data studies that we've done as part of our group.

6
00:00:34,600 --> 00:00:40,389
So as Emily said, I'm a GP, so I'm a GP in the Midlands.

7
00:00:40,390 --> 00:00:45,310
In a suburban practice we look after about 10,000 patients in our parts.

8
00:00:46,330 --> 00:00:49,569
But I'm also a researcher as well at the University of Oxford.

9
00:00:49,570 --> 00:00:53,320
So just around the corner in the Nuffield Department of Primary Care Health Sciences,

10
00:00:53,920 --> 00:01:01,840
I'm also a member of the nice Chronic Heart Failure Guideline Committee and the Quality Standards Committee.

11
00:01:01,840 --> 00:01:07,719
So that guideline was updated in 2018 and the quality standard is being updated at the moment.

12
00:01:07,720 --> 00:01:13,330
So that gives me a slight angle on policy as well as my research and clinical work.

13
00:01:13,540 --> 00:01:21,160
And hopefully I'll bring that all together in the talk to show how big data can be very powerful for us to improve patient care.

14
00:01:25,040 --> 00:01:30,440
So my research focuses on big data as well as other methodologies.

15
00:01:30,890 --> 00:01:34,610
As Emily mentioned, I think it's important to put that into context.

16
00:01:35,450 --> 00:01:45,920
So big data, what do we mean by that? Well, when I'm in surgery as a GP, I'll be inputting clinical codes about my patient, about diagnosis.

17
00:01:46,490 --> 00:01:51,140
I'll put treatments are prescribed treatments through the electronic medical record.

18
00:01:51,470 --> 00:01:55,220
If a patient has a test that will appear in their electronic medical record.

19
00:01:55,460 --> 00:01:59,120
So there's lots of data that we input as part of routine clinical care.

20
00:01:59,660 --> 00:02:06,800
And as researchers, we're very fortunate to be able to access this data in large primary care datasets.

21
00:02:07,160 --> 00:02:14,149
So things like the clinical practice research, datalink key research or Kids, the Health Improvement Network,

22
00:02:14,150 --> 00:02:19,850
these are all examples of large primary care datasets that we can use for research purposes.

23
00:02:20,330 --> 00:02:27,210
Our research group has predominantly over the years, and the three examples I give today are all from CPR.

24
00:02:30,810 --> 00:02:33,980
But as a GP I see patients, I see real people.

25
00:02:33,990 --> 00:02:38,280
So the research databases are anonymized. We don't know who these people are.

26
00:02:38,280 --> 00:02:41,100
It's at a population level. We see the numbers.

27
00:02:41,430 --> 00:02:50,280
But when I'm in practice, in a clinic in in the town where I live, I see patients, all types of patients.

28
00:02:50,280 --> 00:02:57,479
So we serve people from young babies through to very elderly and people reaching the end of their lives.

29
00:02:57,480 --> 00:03:01,320
So we provide holistic care to the community which we serve.

30
00:03:01,830 --> 00:03:05,219
And as part of that, we see patients with heart failure,

31
00:03:05,220 --> 00:03:09,540
people with newly diagnosed heart failure, people that don't know they've got heart failure yet,

32
00:03:09,780 --> 00:03:14,730
people that have been living with heart failure for years and people that are dying of heart failure.

33
00:03:15,240 --> 00:03:25,620
So that gives me a real insight into what it's like being diagnosed and living with the condition, and that really helps to inform my research.

34
00:03:28,630 --> 00:03:33,340
So let's talk a bit about heart failure. So just the basics.

35
00:03:33,340 --> 00:03:40,750
Heart failure occurs when the heart is struggling to pump enough blood around the body to meet the metabolic needs of the organs.

36
00:03:41,290 --> 00:03:49,119
And that manifests for the patients in fluid building up in the lungs, causing them to feel really breathless and often increasingly breathless.

37
00:03:49,120 --> 00:03:58,480
Over time, they can also get really swollen legs and they can get fatigue, almost exhaustion, because their organs aren't being perfused properly.

38
00:03:58,960 --> 00:04:05,170
And there the symptoms that patients get, they can have one of those symptoms, they can have all of those symptoms.

39
00:04:05,380 --> 00:04:12,010
But there's around a million people in the UK living with heart failure at the moment and there'll be around 200,000

40
00:04:12,010 --> 00:04:21,190
people diagnosed each year with a new diagnosis of the condition it accounts for a large proportion of NHS expenditure.

41
00:04:21,190 --> 00:04:27,639
So 3 to 4% of the total budget goes on heart failure and that's largely due to hospital costs.

42
00:04:27,640 --> 00:04:33,760
So A&E attendances heart failure cuts around 5% of emergency hospital admissions.

43
00:04:34,120 --> 00:04:41,920
Patients tend to have a longer stay if they've got heart failure compared to other conditions, and they're frequently readmitted to hospital.

44
00:04:43,180 --> 00:04:49,329
And we find that many patients get their first diagnosis of heart failure on an emergency admission.

45
00:04:49,330 --> 00:04:53,050
So they may have had a gradual onset of symptoms up to that point.

46
00:04:53,410 --> 00:04:59,110
But many patients actually get admitted through a prior to their first diagnosis.

47
00:04:59,110 --> 00:05:03,790
And that's something I'm particularly interested in and I'll talk more about in the in the lecture.

48
00:05:04,540 --> 00:05:06,960
And these patients do have a poor prognosis.

49
00:05:06,970 --> 00:05:14,410
So the outlook for people with heart failure is similar to the outlook for people with with common types of cancer.

50
00:05:14,920 --> 00:05:20,340
But it's not all doom and gloom. It's a very treatable and increasingly treatable condition.

51
00:05:20,740 --> 00:05:25,959
We have a large number of drugs which we know can improve quality of life for patients,

52
00:05:25,960 --> 00:05:31,780
control their symptoms, prevent them from going into hospital and improve their survival.

53
00:05:32,260 --> 00:05:37,239
And this new drugs just come on the market, which are really improving outlook for these patients.

54
00:05:37,240 --> 00:05:44,709
So getting a diagnosis of heart failure is absolutely key to accessing these evidence based treatments,

55
00:05:44,710 --> 00:05:48,160
which we know can improve both quantity and quality of life.

56
00:05:51,450 --> 00:05:55,120
So that's the burden of heart failure for the population in the health care system.

57
00:05:55,140 --> 00:05:57,330
What about the patient and their family? Well,

58
00:05:57,360 --> 00:06:06,750
I did some qualitative work a couple of years ago interviewing people with a new diagnosis and ask them just to describe their diagnostic journey.

59
00:06:07,320 --> 00:06:11,520
And they found the term heart failure incredibly frightening. It sounds frightening, doesn't it?

60
00:06:11,550 --> 00:06:18,450
Heart failure. A lot of people felt it was a sort of imminent demise and this one patient just really hadn't

61
00:06:18,450 --> 00:06:23,410
realised what his symptoms meant or recognised that there was something seriously wrong.

62
00:06:23,430 --> 00:06:26,490
He said, I just thought it was probably my age.

63
00:06:26,820 --> 00:06:33,270
I don't think I thought it was the heart to start with. No, I just thought it was because I was walking a long distance and needed to stop.

64
00:06:33,690 --> 00:06:37,239
So we often find this in the qualitative research we've done.

65
00:06:37,240 --> 00:06:41,370
These patients normalise their symptoms and don't necessarily recognise heart

66
00:06:41,370 --> 00:06:45,720
failure in a way they might recognise other conditions where awareness is better.

67
00:06:47,780 --> 00:06:51,679
So during the talk I'd like to give three examples from our work.

68
00:06:51,680 --> 00:06:55,510
So I should say I work with a fantastic group in the department.

69
00:06:55,520 --> 00:06:59,540
So I co-lead the heart failure research team with Professor Richard Hobbs,

70
00:06:59,870 --> 00:07:05,240
and we've worked together for 15 years looking at the primary care aspects of heart failure in different ways.

71
00:07:05,660 --> 00:07:13,520
And I'd just like to give three examples where we use big data to answer important clinical questions and show how it can translate

72
00:07:13,520 --> 00:07:19,310
from these millions of numbers that you're seeing your spreadsheet to actually answering important questions for patients.

73
00:07:20,480 --> 00:07:27,050
And the first one I'd like to start with this survival. So survival is tricky.

74
00:07:27,320 --> 00:07:34,760
When somebody is diagnosed with a condition, it can be challenging to know what they want to know, what they don't want to know.

75
00:07:35,270 --> 00:07:39,170
But it's important that we have answers to questions patients might ask us.

76
00:07:39,980 --> 00:07:45,350
And as a heart failure group, we have a fantastic peer group, so patient and public involvement.

77
00:07:47,300 --> 00:07:51,290
They're a group of people that have got experience of heart failure and they give us

78
00:07:51,290 --> 00:07:55,340
honest answers when we ask them about the types of research we're thinking about doing.

79
00:07:55,460 --> 00:08:01,640
And then we're involved in throughout the research process through to dissemination of our findings at the end.

80
00:08:01,820 --> 00:08:08,420
And it's really fascinating, actually, our patients felt that they did want to have discussions around survival.

81
00:08:08,600 --> 00:08:13,180
They felt the discussions they had had in their own experience were often inadequate.

82
00:08:14,150 --> 00:08:18,800
One said, I like doctors to have the facts, and I felt that they just hadn't.

83
00:08:19,250 --> 00:08:27,290
And also this idea of this frightening term and imminent demise needed to be addressed by the clinician rather than ignored.

84
00:08:27,950 --> 00:08:31,790
So we decided to do a study called Survive H.S.

85
00:08:34,170 --> 00:08:36,879
To answer these specific type of questions.

86
00:08:36,880 --> 00:08:43,350
So a patient might ask me as a GP if I tell them they've got heart failure or the specialist has told them and they come back to me,

87
00:08:43,350 --> 00:08:47,910
So what is that and what does that mean? So how long will I survive?

88
00:08:48,570 --> 00:08:53,160
What will I die from? Or is outlook for people like me improving?

89
00:08:53,190 --> 00:08:57,840
Now, not every patient will want to know the answers to these questions, but some people might.

90
00:08:58,110 --> 00:09:03,120
And as a GP sitting in front of that patient, I want to know the answers.

91
00:09:03,120 --> 00:09:05,609
And you might think, Well, don't we know this already?

92
00:09:05,610 --> 00:09:12,839
There's a lot of work in cancer epidemiology where these these answers are very well known, but not in heart failure so much.

93
00:09:12,840 --> 00:09:23,639
So a lot of the data on survival comes from large screening studies, often from the US, which is a different population to the UK or from hospitals,

94
00:09:23,640 --> 00:09:29,190
patients and people in hospital with heart failure are quite different to people living in the community with heart failure.

95
00:09:30,120 --> 00:09:37,470
So we felt we were in a good place to answer this question using big data and this was the end result.

96
00:09:37,740 --> 00:09:45,720
So this was a paper published in the BMJ on Valentine's Day in 2019, which seemed very appropriate and timely.

97
00:09:45,750 --> 00:09:47,190
I think it was just coincidence,

98
00:09:47,190 --> 00:09:55,560
but we looked at trends in survival after a diagnosis of heart failure in the United Kingdom and over a long time period.

99
00:09:55,920 --> 00:10:02,550
So from the millennium, up until the most recent data when we started the study in 2017.

100
00:10:06,060 --> 00:10:15,000
So this was a population based cohort study using the clinical practice research datalink which you'll, you'll have heard of already on the module.

101
00:10:15,450 --> 00:10:20,489
So a large anonymized primary care dataset and we used a long time period.

102
00:10:20,490 --> 00:10:28,820
So we went from the 1st of January 2000 up until the end of December 2017, and we were doing the data in 2018.

103
00:10:28,830 --> 00:10:31,830
So we went all the way to the end of the previous year.

104
00:10:32,100 --> 00:10:40,260
The data completeness. One of the values of big data is that we can link to other datasets.

105
00:10:40,710 --> 00:10:43,470
So we linked to hospital episode statistics.

106
00:10:43,470 --> 00:10:51,270
So that gives us an idea of what happens to these patients in hospital when they go in, what they go in for, how long they're there.

107
00:10:52,230 --> 00:10:59,340
And we also had linkages to the Office for National Statistics Mortality Registry.

108
00:10:59,340 --> 00:11:04,590
So that tells us when patients have died and also what they've died of.

109
00:11:04,590 --> 00:11:10,380
And that information comes from the death certificate completed by the doctor who is last involved with their care.

110
00:11:10,920 --> 00:11:17,850
So these are quite powerful combinations because we can really get an insight into what's happening with these patients.

111
00:11:21,900 --> 00:11:30,840
So in survive. Tess We had 395 practices records for over 2 million people,

112
00:11:31,080 --> 00:11:39,690
so no cohort study or screening study where you actively recruit patients who would achieve anywhere like this number of people.

113
00:11:39,870 --> 00:11:48,599
You just wouldn't have the funding to do it. We chose people over the age of 45 because they're people that are likely to

114
00:11:48,600 --> 00:11:51,629
develop heart failure and those who get heart failure when they're younger,

115
00:11:51,630 --> 00:11:59,160
it tends to be a congenital cause. And that's a different condition with with different treatment options.

116
00:12:00,240 --> 00:12:08,840
So within this 2.4 million people, we found maybe 56,000 people with a new diagnosis of heart failure.

117
00:12:08,850 --> 00:12:13,710
So that's the new diagnostic code within their primary care records.

118
00:12:14,670 --> 00:12:23,630
Average age of diagnosis was 77, which is similar to what you find in other epidemiologic studies in heart failure.

119
00:12:24,510 --> 00:12:28,709
And we looked with the linkage with hospital less episodes statistics what

120
00:12:28,710 --> 00:12:34,500
proportion of people had been admitted to hospital around the time of diagnosis?

121
00:12:34,500 --> 00:12:41,190
And we found it was around 43% in our cohort, although some studies have reported that to be even higher.

122
00:12:43,950 --> 00:12:49,709
So we did survival analysis looking at age, sex and practice, match controls.

123
00:12:49,710 --> 00:12:53,790
So we had the people with heart failure diagnostic code and then we matched them by age,

124
00:12:53,790 --> 00:13:01,140
sex and practice to people that just didn't have a code that just to see the difference in survival and they saw a kaplan-meier curve.

125
00:13:02,100 --> 00:13:09,720
So the blue line is those without heart failure in their records and the dotted yellow line is those with heart failure.

126
00:13:09,960 --> 00:13:16,650
And as you can see, there's a dramatic difference. So X-axis, this time y axis is survival probability,

127
00:13:16,920 --> 00:13:23,430
and there's a real reduction in survival in those with heart failure compared to their peers who don't have heart failure.

128
00:13:24,570 --> 00:13:31,110
And just looking at that risk table, I think that's also the power of big data is just the sheer numbers.

129
00:13:31,120 --> 00:13:40,890
So each time point to five years, at ten years, even at 15 years, we've got lots of people in the cohort to look at and compare.

130
00:13:40,900 --> 00:13:47,550
So we can really see what's happening within that population and what the the trend looks like.

131
00:13:49,530 --> 00:13:58,679
And we also dig a little bit deeper. So if we want to answer the question, what is is the survival rate for people at one year, five, ten years?

132
00:13:58,680 --> 00:14:05,370
We can calculate that from the survival analysis, but we can also look at it by gender, but also by age.

133
00:14:06,480 --> 00:14:14,820
And as you can see at a younger age, your chance of surviving a year at a 90% if you're between 45 and 54.

134
00:14:15,750 --> 00:14:23,220
But as you go down in the older age groups, when you get to 75 to 84, it's it's 76.5%.

135
00:14:24,090 --> 00:14:32,760
So if you're talking to the patient in front of you, it's helpful not only to have an overall idea of survival rates in the general population,

136
00:14:33,120 --> 00:14:36,749
but it's also important to put it into context of the patient.

137
00:14:36,750 --> 00:14:45,600
Demographics and particularly age is very influential on survival, as you would expect, and I think that's important for us to report in our papers.

138
00:14:45,600 --> 00:14:52,800
When the BMJ paper came out, there was a lot of fear, I think, because overall the survival did look pretty bad.

139
00:14:53,070 --> 00:14:55,170
But actually if you look at younger patients,

140
00:14:55,170 --> 00:15:03,000
the survival wasn't as bad as the overall cohort because there are plenty of treatments available to improve prognosis.

141
00:15:05,520 --> 00:15:12,780
So we also looked at cause of death and we had over 30,000 deaths within the cohort.

142
00:15:12,780 --> 00:15:15,990
So a large proportion of the cohort died during the follow up.

143
00:15:16,980 --> 00:15:20,010
And this allowed us to look at what the causes were.

144
00:15:20,010 --> 00:15:23,999
So patients with a diagnosis of heart failure, what do they actually die from?

145
00:15:24,000 --> 00:15:27,570
Do they die from heart failure or do they die of something else?

146
00:15:28,050 --> 00:15:32,070
And as you can see on the so on a death certificate, you will put the primary cause of death.

147
00:15:32,080 --> 00:15:37,709
So the main thing that has led to death in that person and then you would put secondary

148
00:15:37,710 --> 00:15:42,240
causes which are sort of contributing factors but not the main thing they died from.

149
00:15:42,240 --> 00:15:51,690
So heart failure was the main cause in 7.2%, but it was actually influential in around half of patients.

150
00:15:52,410 --> 00:15:58,830
So that's important to know. But also that means 50% of people with heart failure actually died of something else.

151
00:15:59,430 --> 00:16:06,749
And that's important as a GP to bear in mind because these patients are usually living with

152
00:16:06,750 --> 00:16:11,010
several other long term conditions and we don't want to just focus on the heart failure,

153
00:16:11,010 --> 00:16:16,890
we want to focus on managing all of those conditions to give them the best quality of life and the best survival chances.

154
00:16:17,640 --> 00:16:20,049
So we also looked at the top three causes.

155
00:16:20,050 --> 00:16:27,390
So severe cardiovascular disease as high as you would expect, but then the next one down, respiratory and also cancer.

156
00:16:28,230 --> 00:16:31,470
So useful information for us to to have.

157
00:16:35,810 --> 00:16:39,930
And we also looked at trends over time. So what we want to see is improvement.

158
00:16:39,960 --> 00:16:44,060
So medical science improving survival rates for people with conditions.

159
00:16:44,210 --> 00:16:51,740
And we've seen this in cancer say cancer research. UK say that survival rates have doubled in cancer overall over the past 40 years.

160
00:16:52,070 --> 00:16:57,080
Now, unfortunately, we just don't see the same improvement in heart failure.

161
00:16:57,470 --> 00:17:04,940
So as you can see on the x axis, this the the year of diagnosis and then the Y axis is survival percentage.

162
00:17:05,270 --> 00:17:13,070
And there's a very slight increase in one year, five year and ten year survival rates over that time period.

163
00:17:13,910 --> 00:17:20,430
But if we look at the the differences from the start of the study to the end, it's really quite small numbers.

164
00:17:20,450 --> 00:17:26,210
So between six and 7% improvements in the one, five and ten year survival rates.

165
00:17:27,470 --> 00:17:33,560
Whereas as I said, with cancer, there's been a doubling. So 200% improvement in a 40 year period.

166
00:17:33,590 --> 00:17:39,920
So we still have a lot of work to do to improve patient's chances once they're diagnosed.

167
00:17:42,180 --> 00:17:46,229
And with the linkage we were also able to look at hospitalisations.

168
00:17:46,230 --> 00:17:54,510
So patients hospitalised around the time of diagnosis compared to those who didn't end up going into hospital around the time of diagnosis.

169
00:17:54,930 --> 00:18:01,710
And you can see the yellow line are those that weren't admitted and the pink cast on those were admitted.

170
00:18:01,980 --> 00:18:08,280
So patients admitted to hospital around the time of diagnosis had a worse survival than those who weren't.

171
00:18:08,610 --> 00:18:14,309
Now that that seems to make sense, but that might be important for us in primary care,

172
00:18:14,310 --> 00:18:20,340
because if we can actually keep people out of hospital by diagnosing and treating them a bit earlier,

173
00:18:20,640 --> 00:18:24,480
that we might actually improve the survival chances potentially.

174
00:18:24,540 --> 00:18:27,540
So there's a definite different scene with the data.

175
00:18:30,380 --> 00:18:37,430
And we also looked at both hospitalised and non-hospitalized patients in terms of the improvements on the left.

176
00:18:37,430 --> 00:18:42,799
It's those that were in the community, not in hospital and their survival rates.

177
00:18:42,800 --> 00:18:47,390
There has been an improvement compared to those who were hospitalised.

178
00:18:47,390 --> 00:18:53,240
So we really haven't seen much improvement at all in people that get hospitalised around that time of diagnosis,

179
00:18:53,240 --> 00:19:00,170
whereas people in the community there has been more improvement than looking at the the cohort as a whole.

180
00:19:04,210 --> 00:19:08,350
So there's strengths and limitations with big datasets.

181
00:19:08,680 --> 00:19:14,680
You get a large number of people and that is a very powerful thing for research.

182
00:19:14,680 --> 00:19:21,730
So we can look at trends over time. We can compare survival in those that do or don't have the condition.

183
00:19:21,940 --> 00:19:27,130
We can calculate survival rates based on age, based on gender.

184
00:19:28,990 --> 00:19:34,660
The downside of datasets, as I'm sure you've covered already, is that, as I said at the beginning,

185
00:19:34,660 --> 00:19:39,820
we use the primary care record predominantly to record our clinical care.

186
00:19:40,120 --> 00:19:45,729
So when I'm sitting in my surgery, seeing patients every 10 minutes at the forefront of my mind,

187
00:19:45,730 --> 00:19:50,530
isn't that this clinical code might be used by a researcher in Oxford at some point in the future.

188
00:19:50,530 --> 00:19:57,340
So I need to get it right. I mean, it does it does make me very aware of the need to code properly.

189
00:19:57,880 --> 00:20:02,200
But this is a snapshot of what's happening in real life clinical practice.

190
00:20:02,500 --> 00:20:07,120
Now, CPR data very good at looking at the quality of coding within their practices.

191
00:20:07,330 --> 00:20:11,040
So they do make sure that the practices are up to standard.

192
00:20:11,050 --> 00:20:21,580
So what we're seeing is reflective of real life primary care, but it is predominantly for routine clinical care.

193
00:20:21,580 --> 00:20:28,299
I think that is a strength as well because it does show real life recruiting patients into research studies.

194
00:20:28,300 --> 00:20:33,400
Does involved involve a certain type of bias, you know, who is willing to participate,

195
00:20:33,400 --> 00:20:37,930
etc., whereas this is showing what is happening every day in the community.

196
00:20:37,930 --> 00:20:40,210
A million patients a day see their GP.

197
00:20:42,220 --> 00:20:51,310
Clinical coding is something that we acknowledge is a limitation in the discussion of all our papers, particularly in heart failure.

198
00:20:51,310 --> 00:20:53,980
It's relevant for the type of heart failure.

199
00:20:53,980 --> 00:21:01,540
So two types of heart failure, heart failure with reduced ejection fraction and heart failure with preserved ejection fraction,

200
00:21:01,660 --> 00:21:08,080
and the treatments for those are quite different. So it's important at diagnosis to differentiate between the two,

201
00:21:08,410 --> 00:21:16,180
but that may not necessarily be recorded in the primary care record and we see in these codes increasingly use more recently.

202
00:21:16,420 --> 00:21:19,299
But if we go back a few years, they weren't really being used before.

203
00:21:19,300 --> 00:21:24,070
So we can't see what type of heart failure it is that we're looking at, unfortunately.

204
00:21:24,070 --> 00:21:26,980
But in the future we are likely to be able to see that.

205
00:21:28,210 --> 00:21:35,950
The other thing which review is sometimes say to us is around the reliability of death reporting in the UK because we don't do many post-mortems,

206
00:21:35,950 --> 00:21:42,040
it relies on the doctor looking after the patient during their last illness and what they put on the death certificate.

207
00:21:43,690 --> 00:21:48,340
But as a reporting system, it is one of the most robust in the world.

208
00:21:49,000 --> 00:21:54,399
And we have to go on what is written on the death certificate because there is no other way and

209
00:21:54,400 --> 00:22:00,040
the fact that it's actively recorded by S and then we can link it with the primary care record.

210
00:22:00,370 --> 00:22:05,800
That's that's quite a powerful linkage that isn't available in many other countries.

211
00:22:06,160 --> 00:22:12,280
So we always push back on that one because I think this is particularly looking at the cause of deaths.

212
00:22:12,550 --> 00:22:16,720
I think is is really important just to give us an idea of what people are dying from and when.

213
00:22:22,540 --> 00:22:33,370
So that was our paper. In 2019, we decided to publish the gender statistics in slightly more detail in the European Journal of Heart Failure.

214
00:22:33,700 --> 00:22:37,990
So we reported national trends in heart failure, survival for men and women.

215
00:22:39,190 --> 00:22:42,250
And and that was published the following year.

216
00:22:43,150 --> 00:22:50,799
And I think, again, because the dataset so big, you get very robust estimates for survival rates, splitting it by by gender.

217
00:22:50,800 --> 00:22:59,140
And what we found was that actually patients so female patients compared to male patients are diagnosed on average five years later.

218
00:23:00,820 --> 00:23:07,720
But then if we adjust, right, their survival is their mortality is actually lower.

219
00:23:07,900 --> 00:23:10,060
So their survival is better than men.

220
00:23:10,900 --> 00:23:18,880
These are infographics which we've started to produce for all of our papers because most people don't read a whole paper,

221
00:23:18,910 --> 00:23:24,760
unfortunately, cover to cover. But what we do is a visual representation of the study.

222
00:23:24,760 --> 00:23:31,480
So aim at the top, the key characteristics, the numbers that we're looking at, and then the key results in bright colours.

223
00:23:31,780 --> 00:23:36,910
And that just helps to get the message out from our research to a wider audience.

224
00:23:37,270 --> 00:23:47,139
So we put these on our website, we tweet them, we put them on Facebook, and we do blogs and things on on the mic part of the website,

225
00:23:47,140 --> 00:23:52,900
which is around diagnostics, just to try and increase the dissemination of our work.

226
00:23:52,900 --> 00:23:55,540
Because if we're going to spend years doing all this research,

227
00:23:55,540 --> 00:24:04,030
we want people to actually engage with it and see what we're doing because hopefully it's relevant to the community population.

228
00:24:06,200 --> 00:24:15,500
So what does this mean for my practice? So as I said, I work in a sort of suburban primary care.

229
00:24:15,560 --> 00:24:20,270
General practice is ATP, so we cover about 10,000 patients.

230
00:24:20,750 --> 00:24:26,450
So that means statistically we'll have about 20 patients newly diagnosed with heart

231
00:24:26,450 --> 00:24:30,890
failure each year and about 100 people on our list living with heart failure.

232
00:24:31,460 --> 00:24:34,880
So in that 20, that will be newly diagnosed.

233
00:24:34,910 --> 00:24:42,890
What our research shows is that around four of them, so 20% will die within the first year.

234
00:24:43,130 --> 00:24:50,720
And their patients that we need to identify early, we need to have them on our palliative care registers and gold standards framework,

235
00:24:50,720 --> 00:25:02,270
which is how we assess end of life care and identify people that are more likely to need end of life care because often these patients do get missed.

236
00:25:03,020 --> 00:25:09,020
Most of our registries focus on cancer diagnoses and cancer deaths,

237
00:25:09,200 --> 00:25:15,260
but actually people with heart failure also need high quality palliative care and end of life care.

238
00:25:16,670 --> 00:25:26,329
So at that 24, we're losing a year, but ten will survive the next five years and actually six will still be alive ten years later.

239
00:25:26,330 --> 00:25:38,360
So those six are people whose heart failure is potentially treatable, manageable, and it may be something else that causes them a problem in the end.

240
00:25:38,360 --> 00:25:41,599
So that's the difficulty in general practice.

241
00:25:41,600 --> 00:25:47,300
But also the joy of it is that you look at the patient overall, you don't just look at the one diagnosis,

242
00:25:47,660 --> 00:25:55,550
you look at what is wrong with them and how we can manage that individual optimally to optimise their pace,

243
00:25:55,760 --> 00:25:58,940
their quality of life as as well as their survival.

244
00:26:01,380 --> 00:26:06,570
So that's looking at our survival analysis and how we use big data for that.

245
00:26:06,870 --> 00:26:15,120
And I'd just like to spend the second half of the talk looking at diagnosis, which is what we're focusing our research on at the moment.

246
00:26:15,150 --> 00:26:19,370
So what does Big Data tell us about diagnosis?

247
00:26:21,830 --> 00:26:24,700
So just a little bit about the heart failure pathway.

248
00:26:24,710 --> 00:26:35,930
So as I said, patients with heart failure develop shortness of breath, ankles swelling, leg swelling, and they get incredibly fatigued and exhausted.

249
00:26:37,250 --> 00:26:45,649
And the diagnostic pathway relies on patients recognising the symptoms and seeking help for them.

250
00:26:45,650 --> 00:26:53,959
And as I said from my qualitative research, patients are often quite stoical at the beginning and didn't necessarily always

251
00:26:53,960 --> 00:26:58,870
appreciate that their symptoms could be linked to something like heart failure.

252
00:26:58,880 --> 00:27:04,910
So patients would say to me that they had put their breathlessness perhaps down to another condition.

253
00:27:04,910 --> 00:27:10,309
If they got lung disease, they put it down to that or almost normalise the symptom saying,

254
00:27:10,310 --> 00:27:14,810
Oh, it's just getting older, or put it down to all the medication that they were on.

255
00:27:15,080 --> 00:27:22,549
So there is in most of my interviews, there was a gap between patients getting symptomatic and actually seeing their GP.

256
00:27:22,550 --> 00:27:30,380
And I think there's an issue there about awareness and we need to raise awareness of heart failure more generally amongst the public.

257
00:27:31,040 --> 00:27:36,590
So when a patient comes to see the GP, it's then our responsibility to think of heart failure.

258
00:27:36,770 --> 00:27:41,930
So as you said, these patients often have several long term conditions or on lots of medications.

259
00:27:42,320 --> 00:27:50,120
So heart failure may be one of many things going through our minds when we're looking at why is this patient breathless?

260
00:27:50,780 --> 00:27:59,389
And we know from some research there is a delay between patients coming to see their GP and actually getting a diagnosis of heart failure because

261
00:27:59,390 --> 00:28:08,570
we can go down other diagnostic routes before we think about heart failure and then when we've considered heart failure is a possibility,

262
00:28:08,570 --> 00:28:12,350
we do a blood test which shows if the heart is under strain or not.

263
00:28:12,350 --> 00:28:21,139
And I'll talk about that in a second. And if that blood test is raised, then we need to refer the patient to get a definitive diagnosis.

264
00:28:21,140 --> 00:28:28,700
So heart failure is the diagnosis we make in primary care. If we think somebody's got heart failure, we need to refer for further imaging.

265
00:28:28,700 --> 00:28:38,210
So a heart scan called an echocardiogram and also a specialist assessment, and it's usually a cardiologist that would make the final diagnosis.

266
00:28:41,700 --> 00:28:46,190
So this is the nice guidance on testing for heart failure.

267
00:28:46,200 --> 00:28:52,110
So there's a blood test which goes up when the heart is under strain called not genetic peptide.

268
00:28:53,130 --> 00:28:56,630
There's two types this NT, Probnp and BNP.

269
00:28:56,640 --> 00:29:01,650
There are two parts of the same molecule and they're both not terrific peptides.

270
00:29:02,040 --> 00:29:05,680
The one that tends to be used more commonly now is anti probnp.

271
00:29:05,910 --> 00:29:12,250
And in the nice guideline, there's clear guidance for what we should do depending on the results of this blood test.

272
00:29:12,250 --> 00:29:17,610
So it's a simple blood test that we take along with blood tests we would do for kidney function,

273
00:29:18,570 --> 00:29:23,990
blood count, etc. if the level of treated peptides very high,

274
00:29:24,000 --> 00:29:34,049
so above 2025 with where to refer urgently for a patient to have a heart scan and a specialist assessment within two weeks.

275
00:29:34,050 --> 00:29:37,320
So similar to the two week wait cancer pathways.

276
00:29:37,950 --> 00:29:44,759
If it's between 402,000, we need especially assessment and echo within six weeks.

277
00:29:44,760 --> 00:29:52,620
But if it's below 400, then that means heart failure is unlikely and we should think about alternative conditions.

278
00:29:52,620 --> 00:30:00,150
So it can be a very helpful test to rule out heart failure if if you just want to make sure it's not that and it's probably something else.

279
00:30:01,780 --> 00:30:07,900
So not sure. I think peptides are released by the ventricles in response to pressure or fluid overload.

280
00:30:08,350 --> 00:30:12,460
They relax vascular smooth muscle and also induce a decrease.

281
00:30:12,610 --> 00:30:16,360
So they act like a lot of the drugs we use to treat heart failure.

282
00:30:19,570 --> 00:30:27,969
And the two types of not drastic peptide, a equally reliable intake gnosis and T Pro is slightly more stable.

283
00:30:27,970 --> 00:30:34,930
So if the blood test is sitting in a general practice for hours before they go off to the lab, that won't affect the anti pro results.

284
00:30:35,860 --> 00:30:40,149
And the thresholds are different though, so you need to know which one you're doing.

285
00:30:40,150 --> 00:30:50,410
So BNP less than 200 or anti pro less than 400 makes heart failure unlikely, but it doesn't differentiate between the two types of heart failure.

286
00:30:50,410 --> 00:30:52,060
So you need an echo for that.

287
00:30:54,660 --> 00:31:04,860
So this is a study where we looked at the association between the naturalistic peptide level at diagnosis and subsequently what happens to patients.

288
00:31:04,860 --> 00:31:09,290
So how likely they are to be hospitalised and how likely they are to die.

289
00:31:09,300 --> 00:31:13,360
And this was published in the journal Heart last year.

290
00:31:13,380 --> 00:31:18,990
So that's a journal that's read predominantly by cardiologists in the UK.

291
00:31:21,280 --> 00:31:27,999
So as we've said, heart failure is a malignant condition requiring urgent treatment, and the guidelines recommend that we do not treat it.

292
00:31:28,000 --> 00:31:32,320
Peptide Testing to prioritise referral for diagnosis.

293
00:31:35,490 --> 00:31:39,870
And we use CPR again between 2004 and 2018.

294
00:31:40,080 --> 00:31:45,900
And we had over 40,000 people with a new diagnosis of heart failure,

295
00:31:46,470 --> 00:31:52,050
half of whom were men and slightly older, 78 and a half was the main age at diagnosis.

296
00:31:52,410 --> 00:32:03,450
Again, we linked to the hospital episode statistics dataset and also to the OS civil mortality data to get a holistic view of what was happening,

297
00:32:03,450 --> 00:32:08,940
and particularly to answer this research question about who's going into hospital and who's dying of heart failure.

298
00:32:09,690 --> 00:32:17,490
And we looked at one year hospitalisation. So from a diagnostic code, how likely are people to go into hospital during that time?

299
00:32:17,760 --> 00:32:26,310
And also the one, five and ten year survival rates as we had with survival, but specifically in relation to the AMP level at diagnosis.

300
00:32:29,260 --> 00:32:37,560
So. As you can see from the survival curves, we've got time on the x axis, survival probability on the Y axis.

301
00:32:38,010 --> 00:32:43,560
And we looked at the different categories. So as we said, less than 400, it's unlikely to be heart failure.

302
00:32:43,680 --> 00:32:48,809
I think something else, 400, 2000 is what we call moderately raised.

303
00:32:48,810 --> 00:32:57,720
And then above 2000 is is a high level and they're the people that need referral and see within two weeks.

304
00:32:58,380 --> 00:33:03,840
And this survival curve just sort of confirms that the guidance is correct in

305
00:33:03,840 --> 00:33:09,389
that the sort of dotted blue line is those people that have an anti pro level

306
00:33:09,390 --> 00:33:14,790
above 2000 at diagnosis and they're the people that are much more likely to die

307
00:33:14,790 --> 00:33:20,610
of the condition than those who have a moderate or a normal anti probnp level.

308
00:33:25,190 --> 00:33:34,020
So overall in the cohort of 40,000 we found over half were hospitalised in the year following diagnosis, a huge burden for the health service.

309
00:33:34,170 --> 00:33:43,579
If you've got 50% chance of going into hospital a year after diagnosis and it was those people who had a very high anti pro BNP,

310
00:33:43,580 --> 00:33:52,250
so above 2000 that were more likely to have a heart failure related hospitalisation compared to those with a moderate natural tick peptide.

311
00:33:52,260 --> 00:33:55,760
So these are very high risk people and we need to get them seen,

312
00:33:55,760 --> 00:34:01,580
diagnosed and treated very quickly in order to prevent them going into hospital or having worse outcome.

313
00:34:03,560 --> 00:34:09,290
So we also looked at overall mortality rates between those with high or moderate MP.

314
00:34:09,290 --> 00:34:13,550
And as you can see, again, the mortality rates at one, one year,

315
00:34:13,580 --> 00:34:22,310
five years and ten years are all higher in that high anti probnp group compared to the moderate and TPO BNP group.

316
00:34:23,210 --> 00:34:27,850
We did a competing risk model to look at the risk of heart failure related deaths.

317
00:34:27,850 --> 00:34:31,790
So are people dying from heart failure or something else?

318
00:34:31,790 --> 00:34:42,620
And those with a high anti pro BNP had a 50% higher rate of heart failure related deaths compared to those with moderate.

319
00:34:42,630 --> 00:34:47,420
So these are people going into hospital with heart failure, dying of that heart failure.

320
00:34:48,740 --> 00:34:58,910
And interestingly, the median time between anti so natural peptide test and heart failure diagnosis was over 100 days.

321
00:34:59,210 --> 00:35:04,190
So nice recommend that patients following their test should be seen and diagnosed

322
00:35:04,190 --> 00:35:09,410
within two weeks if it's both 2000 but within six weeks for everyone else.

323
00:35:09,740 --> 00:35:17,750
And this is way outside of that. So patients are waiting a long time and I think that's an area where we need to shine

324
00:35:17,750 --> 00:35:24,680
a light and that's an area that requires improvement because if in primary care,

325
00:35:24,710 --> 00:35:29,720
I see a patient that might have heart failure. I do. And that genetic peptide test and it's high.

326
00:35:30,050 --> 00:35:40,700
If it's at both 2000, I need that patient to have the echo and see a specialist quickly because otherwise and what I have seen in clinical

327
00:35:40,700 --> 00:35:46,939
practice is patients have had their MP test and while they're waiting to get their scan and their right patient appointment,

328
00:35:46,940 --> 00:35:51,889
they're admitted to hospital because they're too unwell to to be in the community.

329
00:35:51,890 --> 00:35:57,110
So they go into hospital, have intravenous diuretics, and that's when they get their diagnosis.

330
00:35:57,350 --> 00:36:05,089
If we could make that diagnosis sooner and initiate treatment more quickly, we can maybe have prevented that hospital admission.

331
00:36:05,090 --> 00:36:11,960
And I think that's what they see this data show, is that this is what's going on and this is where we need some improvements.

332
00:36:13,130 --> 00:36:19,850
It's particularly important when hospitals are under such strain from COVID and trying to catch up after COVID.

333
00:36:20,600 --> 00:36:23,870
We need to keep people out of hospital unless they need to be there.

334
00:36:24,350 --> 00:36:28,970
And I think what we need is more capacity in cardiac imaging.

335
00:36:29,270 --> 00:36:32,510
Get in an echo, takes weeks and weeks where I work.

336
00:36:33,590 --> 00:36:38,990
So there is a real shortage of echo technicians to be able to do these specialist scans.

337
00:36:39,350 --> 00:36:48,889
So there's several points in the pathway where we need increased capacity in the system in order for the diagnostic process to work properly.

338
00:36:48,890 --> 00:36:55,310
And I think that's what these big data can show us and that's what we can take to policymakers to try and effect change.

339
00:36:59,380 --> 00:37:06,970
And my final example is our most recent paper looking at long term trends in nutritive peptide testing.

340
00:37:08,770 --> 00:37:19,990
So we've seen survival for heart failure is pretty poor 50%, five year survival rates that haven't really improved over time.

341
00:37:20,440 --> 00:37:28,599
And we've seen from the nice guidance that transit peptide testing is absolutely key to getting a diagnosis

342
00:37:28,600 --> 00:37:33,730
and getting on these effective treatments that we know can improve quality and quantity of life.

343
00:37:35,020 --> 00:37:43,930
So in this study, we wanted to look at what's happening with naturalistic peptide testing and what's what's the sort of trend with it over time.

344
00:37:44,320 --> 00:37:51,090
Because certainly when I qualified, we weren't using not tested peptide testing in the community.

345
00:37:51,100 --> 00:37:54,280
I didn't have access to it. And now we are.

346
00:37:56,190 --> 00:38:04,620
So the overall aim was to report the trends in testing and look at subsequent heart failure diagnosis over time.

347
00:38:04,890 --> 00:38:08,660
Again, we chose the same data as the previous study.

348
00:38:08,670 --> 00:38:13,950
It was all part of a CPD data code to answer three different questions.

349
00:38:13,950 --> 00:38:19,469
And the third question I won't talk about today, but will be at BGP in January.

350
00:38:19,470 --> 00:38:23,940
So that's our last bit to this, this big piece of work.

351
00:38:24,960 --> 00:38:33,420
But again, 14 year period, long period of time linked to the hospital episode statistics and OS mortality data.

352
00:38:33,720 --> 00:38:43,980
We had a bigger dataset this time, so 7.2 million patients we looked for not just peptide testing codes within their primary care records,

353
00:38:43,980 --> 00:38:53,850
and we had over a thousand practices in this. So a significant proportion of the UK population, probably 10 to 15% covered within this study.

354
00:38:54,570 --> 00:38:57,840
And what what has been happening with that peptide testing?

355
00:38:57,840 --> 00:39:04,590
So as you can see from 2004, we weren't really doing much testing because most of us didn't have access to it.

356
00:39:06,540 --> 00:39:16,229
I qualified in 2009. We got access in my practice in about 2011, and that really represents the national picture.

357
00:39:16,230 --> 00:39:26,460
So there's this sudden uplift in 2010, real increase in testing, both BNP, which is the blue line and anti pro BNP, which is the red line.

358
00:39:27,330 --> 00:39:30,510
But as you can see, and Tipos become much more the dominant test.

359
00:39:30,510 --> 00:39:38,159
So it depends what your lab offer as to what you can order, but anti pro is more stable over time.

360
00:39:38,160 --> 00:39:41,670
It's not metabolised by the newer drugs as well.

361
00:39:42,480 --> 00:39:52,620
So nice now recommending anti pro BNP testing if available, but obviously if your hospital lab is set up for BNP then that's fine too.

362
00:39:53,670 --> 00:39:58,559
And 2010 is the important year because that's when nice updated guidance previously.

363
00:39:58,560 --> 00:40:05,129
So I was on the 2018 update committee, but the previous update had been in 2010.

364
00:40:05,130 --> 00:40:11,490
And in that document they really strengthen the role of not treating peptide testing in the diagnostic pathway.

365
00:40:11,490 --> 00:40:15,000
And it's interesting to see this uplift in testing since then.

366
00:40:17,700 --> 00:40:22,799
We also looked at men and women again and in different age groups.

367
00:40:22,800 --> 00:40:26,580
So Blue is the oldest age group and this is appropriate.

368
00:40:26,580 --> 00:40:35,400
So it's by age. So there's more testing in older people and that's what you'd expect because the mean age at diagnosis is 76, 77.

369
00:40:35,640 --> 00:40:40,080
We should be testing more in older people because they're more likely to have the condition.

370
00:40:42,950 --> 00:40:51,490
And similarly, we looked at ethnicity. So not much difference between ethnic groups and deprivation.

371
00:40:51,500 --> 00:40:58,280
So it's good to say that in the most deprived group, that's where the most testing is.

372
00:40:58,700 --> 00:41:03,559
And heart failure prevalence is higher in deprived populations.

373
00:41:03,560 --> 00:41:09,680
But sometimes you see health inequalities where actually the least deprived, they're getting the most testing.

374
00:41:09,830 --> 00:41:18,530
But actually, from this big data, that wasn't the case. And we are testing the people that need to be tested, which is good.

375
00:41:20,980 --> 00:41:29,980
And we we get a number. So the natural peptide level within the big data so we can calculate the median level at

376
00:41:29,980 --> 00:41:35,080
time of heart failure diagnosis and we can look at that for each year of the 14 years,

377
00:41:35,080 --> 00:41:40,110
15 years in the study. And as you can see, it's pretty consistent.

378
00:41:40,120 --> 00:41:46,659
So the top graph, so those with BNP tests with and with that heart failure and the bottom is

379
00:41:46,660 --> 00:41:52,809
anti pro with heart failure without and there hasn't really been much change.

380
00:41:52,810 --> 00:42:04,120
So the median level of anti pro BNP at diagnosis is around 1200, which is actually quite high because the threshold for referral is only 400.

381
00:42:04,510 --> 00:42:13,360
So that means that we are still getting people that have more advanced heart failure then being close to that threshold.

382
00:42:13,420 --> 00:42:20,799
So that was an interesting finding because perhaps we need to be testing people a bit earlier and we need

383
00:42:20,800 --> 00:42:27,879
to think about how we might be able to do that in order to catch people sooner in the disease process,

384
00:42:27,880 --> 00:42:37,420
to be able to treat them before they land open. And because they're too breathless and they need intravenous diuretics to to make them feel better.

385
00:42:39,390 --> 00:42:44,910
So these were the main findings that we concluded. So there has been this real increase in testing.

386
00:42:44,910 --> 00:42:48,060
So we are using more natural peptide tests in primary care.

387
00:42:48,390 --> 00:42:53,400
We using them in appropriate populations. So older people, more deprived people are getting tested.

388
00:42:54,330 --> 00:42:59,129
We found the heart fake fairly detection rate was around constant, so around 10%.

389
00:42:59,130 --> 00:43:04,590
So one in ten people who have enough to take a test will get a diagnosis of heart failure.

390
00:43:05,010 --> 00:43:09,540
And that hadn't really changed. Whereas in cancer we're testing a lot more people.

391
00:43:10,020 --> 00:43:16,919
So the detection rates are much, much lower because most people tested on having a diagnosis of cancer,

392
00:43:16,920 --> 00:43:22,320
whereas in Hertfordshire that has stayed constant. So perhaps we need to be thinking about testing more people.

393
00:43:22,890 --> 00:43:30,090
And surprisingly, in the last period of the study, still up to 2017, so pre-pandemic,

394
00:43:30,090 --> 00:43:38,070
but recently only one in four people had a natural peptide test prior to that heart failure diagnosis.

395
00:43:38,610 --> 00:43:44,220
So three quarters of people who were diagnosed with heart failure hadn't had a electrolytic peptide test beforehand.

396
00:43:45,540 --> 00:43:48,870
So even though there has been this really big increase in testing,

397
00:43:49,470 --> 00:43:54,810
we probably need to look at more testing in primary care to try and facilitate a

398
00:43:54,810 --> 00:44:00,480
diagnosis at an earlier and more treatable stage before people end up in hospital.

399
00:44:03,620 --> 00:44:13,069
So overall, my lessons from Big Data, I think as I've said, it's a really powerful research tool to be used.

400
00:44:13,070 --> 00:44:16,549
I reflects better than any other methodology,

401
00:44:16,550 --> 00:44:23,900
really real life practice and what is happening in real life in the million consultations in general practice each day in the UK,

402
00:44:24,710 --> 00:44:30,110
I think the downsides are, as I've said, we use it for routine clinical care.

403
00:44:30,110 --> 00:44:35,719
It's not collected in the same way as we would for robust or qualitative work.

404
00:44:35,720 --> 00:44:41,760
It's just not using the same research methods, but coding standards are looked at and.

405
00:44:41,960 --> 00:44:49,400
ROPER So I think it's still a reliable data source, but we need to acknowledge the limitations whenever we report this work.

406
00:44:50,180 --> 00:44:57,919
I think the really key thing for me is always bringing the question back to the patient, because these data sets are so big,

407
00:44:57,920 --> 00:45:05,590
you can look at so many things and get so many numbers, but actually what what's important and what matters.

408
00:45:05,930 --> 00:45:15,890
So all three examples I gave the research question idea came from patients that I had talked to, either in our peer group or in my clinical practice.

409
00:45:16,250 --> 00:45:22,370
And that's what then drove the big data analysis and the write up and the reporting of the results.

410
00:45:22,370 --> 00:45:26,690
And I think without that focus we can get a bit lost in the big data.