1
00:00:00,948 --> 00:00:03,920
I have a tendency to assume the worst,

2
00:00:03,920 --> 00:00:06,854
and once in a while,
this habit plays tricks on me.

3
00:00:06,854 --> 00:00:09,988
For example, if I feel
unexpected pain in my body

4
00:00:09,988 --> 00:00:13,340
that I have not experienced before
and that I cannot attribute,

5
00:00:13,340 --> 00:00:18,234
then all of a sudden, my mind
my turn a tense back into heart disease

6
00:00:18,234 --> 00:00:21,909
or calf muscle pain into ??.

7
00:00:21,909 --> 00:00:25,406
But so far I haven't been diagnosed
with any deadly or incurable disease.

8
00:00:25,406 --> 00:00:28,559
Sometimes things just hurt
for no clear reason.

9
00:00:28,559 --> 00:00:32,180
But not everyone is as lucky as me.

10
00:00:32,180 --> 00:00:36,497
Every year, more than
50 million people die worldwide.

11
00:00:36,497 --> 00:00:40,054
Especially in high-income
economies like ours,

12
00:00:40,054 --> 00:00:44,490
a large fraction of deaths
is caused by slowly progressing diseases:

13
00:00:44,490 --> 00:00:49,165
heart disease, chronic lung disease,
cancer, Alzheimer's, diabetes,

14
00:00:49,165 --> 00:00:51,510
just to name a few.

15
00:00:51,510 --> 00:00:57,041
Now, humanity has made tremendous progress
in diagnosing and treating many of these,

16
00:00:57,041 --> 00:01:00,294
but we are at stage
where further advancement in health

17
00:01:00,294 --> 00:01:03,314
cannot be achieved only
by developing new treatments,

18
00:01:03,314 --> 00:01:05,852
and this becomes evident
when we look at one aspect

19
00:01:05,852 --> 00:01:09,388
that many of these
diseases have in common.

20
00:01:09,388 --> 00:01:12,480
The probability for successful treatment

21
00:01:12,480 --> 00:01:15,580
strongly depends on
when treatment is started,

22
00:01:15,580 --> 00:01:19,654
but a disease is typically only detected
once symptoms occur,

23
00:01:19,654 --> 00:01:24,259
and the problem here is that in fact
many disease can remain asymptomatic,

24
00:01:24,259 --> 00:01:28,007
hence undetected,
for a long period of time.

25
00:01:28,237 --> 00:01:34,011
Because of this, there is
a persisting need for new ways

26
00:01:34,011 --> 00:01:37,325
of detecting disease at early stage,
way before any symptoms occur.

27
00:01:37,325 --> 00:01:39,687
In health care, this is called screening,

28
00:01:39,687 --> 00:01:42,947
and as defined by
the World Health Organization,

29
00:01:42,947 --> 00:01:46,195
screening is the
presumptive identification

30
00:01:46,195 --> 00:01:48,516
of unrecognized disease

31
00:01:48,516 --> 00:01:51,212
in an apparently healthy person

32
00:01:51,212 --> 00:01:55,724
by means of tests that can be applied
rapidly and easily.

33
00:01:55,724 --> 00:01:58,412
That's a long definition,
so let me repeat it:

34
00:01:58,412 --> 00:02:02,060
identification of unrecognized disease

35
00:02:02,060 --> 00:02:04,614
in an apparently healthy person

36
00:02:04,614 --> 00:02:09,036
by means of tests that can be applied
both rapidly and easily.

37
00:02:09,036 --> 00:02:11,579
And I want to put special emphasis
on the words "rapidly" and "easily,"

38
00:02:11,579 --> 00:02:14,221
because many of
the existing screening methods

39
00:02:14,221 --> 00:02:16,399
are exactly the opposite.

40
00:02:16,399 --> 00:02:19,027
And those of you who have gone colonoscopy

41
00:02:19,027 --> 00:02:22,345
as part of a screening program
for colorectal cancer

42
00:02:22,345 --> 00:02:24,504
will know what I mean.

43
00:02:25,442 --> 00:02:28,519
Obviously, there's a variety
of medical tools available

44
00:02:28,519 --> 00:02:30,534
to perform screening tests.

45
00:02:30,534 --> 00:02:32,794
This ranges from imaging techniques

46
00:02:32,794 --> 00:02:36,372
such as radiography
or magnetic resonance imaging

47
00:02:36,372 --> 00:02:38,576
to the analysis of blood or tissue.

48
00:02:38,576 --> 00:02:41,622
We have all had such tests.

49
00:02:41,622 --> 00:02:45,884
But there's one medium
that for long has been overlooked,

50
00:02:45,884 --> 00:02:48,650
a medium that is easily accessible,

51
00:02:48,650 --> 00:02:50,766
basically non-depletable,

52
00:02:50,766 --> 00:02:55,278
and it holds tremendous promise
for medical analysis,

53
00:02:55,278 --> 00:02:58,773
and that is our breath.

54
00:02:58,969 --> 00:03:02,599
Human breath is essentially
composed of five components:

55
00:03:02,599 --> 00:03:09,598
nitrogen, oxygen,
carbon dioxide, water and argon,

56
00:03:09,598 --> 00:03:13,060
but besides these five, there are
hundreds of other components

57
00:03:13,060 --> 00:03:15,710
that are present in very low quantity.

58
00:03:15,710 --> 00:03:18,050
These are called volatile
organic compounds,

59
00:03:18,050 --> 00:03:20,224
and we release hundreds,
even thousands of them,

60
00:03:20,224 --> 00:03:23,409
every time we exhale.

61
00:03:23,409 --> 00:03:26,823
The analysis of these volatile
organic compounds in our breath

62
00:03:26,823 --> 00:03:28,876
is called breath analysis.

63
00:03:28,876 --> 00:03:33,461
In fact, I believe that many of you
have already experienced breath analysis.

64
00:03:33,461 --> 00:03:37,168
Imagine, you're driving home late at night

65
00:03:37,168 --> 00:03:39,990
when suddenly there's
a friendly police officer

66
00:03:39,990 --> 00:03:42,440
who asks you kindly but firmly

67
00:03:42,440 --> 00:03:46,404
to pull over and blow
into a device like this one.

68
00:03:47,759 --> 00:03:51,057
This is an alcohol breath tester

69
00:03:51,057 --> 00:03:54,773
that is used to measure
the ethanol concentration in your breath

70
00:03:54,773 --> 00:03:58,752
and determine whether driving
in your condition is a clever idea.

71
00:03:58,752 --> 00:04:01,680
Now I'd say my driving was pretty good,

72
00:04:01,680 --> 00:04:03,191
but let me check.

73
00:04:03,372 --> 00:04:05,299
(Beep)

74
00:04:08,655 --> 00:04:12,043
0.0, so nothing to worry about, all fine.

75
00:04:14,646 --> 00:04:17,193
Now, imagine a device like this one

76
00:04:17,193 --> 00:04:20,161
that does not only measure
alcohol levels in your breath,

77
00:04:20,161 --> 00:04:22,504
but that detects diseases

78
00:04:22,504 --> 00:04:26,467
like the ones I've shown you
and potentially many more.

79
00:04:27,145 --> 00:04:30,745
The concept of correlating
the smell of a person's breath

80
00:04:30,745 --> 00:04:33,044
with certain medical conditions

81
00:04:33,044 --> 00:04:36,148
in fact dates back to Ancient Greece,

82
00:04:36,148 --> 00:04:41,110
but only recently, research efforts
on breath analysis have skyrocketed,

83
00:04:41,110 --> 00:04:45,130
and what once was a dream
is now becoming reality.

84
00:04:45,130 --> 00:04:48,497
And let me pull up this list again
that I showed you earlier.

85
00:04:49,257 --> 00:04:52,762
For the majority of diseases listed here,

86
00:04:52,762 --> 00:04:56,465
there's substantial scientific evidence
suggesting that the disease

87
00:04:56,465 --> 00:05:00,174
could be detected by breath analysis.

88
00:05:00,174 --> 00:05:02,336
But how does it work exactly?

89
00:05:02,336 --> 00:05:05,433
The essential part is a sensor device

90
00:05:05,433 --> 00:05:09,691
that detects the volatile
organic compounds in our breath.

91
00:05:09,691 --> 00:05:12,655
Simply put, when exposed
to a breath sample,

92
00:05:12,655 --> 00:05:15,531
the sensor outputs a complex signature

93
00:05:15,531 --> 00:05:20,508
that results from the mixture of volatile
organic compounds that we exhale.

94
00:05:20,508 --> 00:05:22,228
Now, this signature represents
a fingerprint of your metabolism,

95
00:05:22,228 --> 00:05:25,776
your microbiome,

96
00:05:25,776 --> 00:05:28,253
and the biochemical processes
that occur in your body.

97
00:05:28,253 --> 00:05:31,926
If you have a disease,

98
00:05:31,926 --> 00:05:34,292
your organisms will change,

99
00:05:34,292 --> 00:05:37,867
and so will the composition
of your exhaled breath,

100
00:05:37,867 --> 00:05:40,212
and then the only thing that is left to do

101
00:05:40,212 --> 00:05:42,843
is to correlate a certain signature

102
00:05:42,843 --> 00:05:46,762
with the presence or absence
of certain medical conditions.

103
00:05:47,978 --> 00:05:52,002
The technology promises
several undeniable benefits.

104
00:05:52,002 --> 00:05:55,177
Firstly, the sensor can be miniaturized

105
00:05:55,177 --> 00:05:58,132
and integrated into small,
handheld devices

106
00:05:58,132 --> 00:06:00,890
like this alcohol breath tester.

107
00:06:00,890 --> 00:06:04,030
This would allow the test to be used
in many different settings

108
00:06:04,030 --> 00:06:05,617
and even at home,

109
00:06:05,617 --> 00:06:07,651
so that a visit at a doctor's office

110
00:06:07,651 --> 00:06:11,320
is not needed each time
a test shall be performed.

111
00:06:11,320 --> 00:06:14,692
Secondly, breath analysis is non-invasive

112
00:06:14,692 --> 00:06:18,207
and can be as simple as blowing
into an alcohol breath tester.

113
00:06:19,110 --> 00:06:23,495
Such simplicity and ease of use
would reduce patient burden

114
00:06:23,495 --> 00:06:27,843
and provide an incentive
for broad adoption of the technology.

115
00:06:27,843 --> 00:06:31,730
And thirdly, the technology is so flexible

116
00:06:31,730 --> 00:06:35,436
that the same device could be used

117
00:06:35,436 --> 00:06:38,164
to detect a broad range
of medical conditions.

118
00:06:38,164 --> 00:06:44,088
Breath analysis could be used to screen
for multiple diseases at the same time.

119
00:06:44,406 --> 00:06:48,760
Nowadays, each disease typically requires

120
00:06:48,760 --> 00:06:49,930
a different medical tool
to perform a screening test,

121
00:06:49,930 --> 00:06:54,058
but this means you can only find
what you're looking for.

122
00:06:54,058 --> 00:06:57,058
With all of these features,
breath analysis is predestined

123
00:06:57,058 --> 00:07:02,746
to deliver what many traditional
screening tests are lacking,

124
00:07:02,969 --> 00:07:06,355
and most importantly,
all of these features

125
00:07:06,355 --> 00:07:10,712
should eventually provide us
with a platform for medical analysis

126
00:07:10,712 --> 00:07:16,071
that can operate at attractively
low cost per test.

127
00:07:16,071 --> 00:07:19,284
On the contrary, existing medical tools

128
00:07:19,284 --> 00:07:22,214
often lead to rather high cost per test.

129
00:07:22,214 --> 00:07:24,535
Then, in order to keep costs down,

130
00:07:24,535 --> 00:07:27,856
the number of tests
needs to be restricted,

131
00:07:27,856 --> 00:07:30,714
and this means a) that the tests
can only be performed

132
00:07:30,714 --> 00:07:35,855
on a narrow part of the population,
for example the high-risk population,

133
00:07:35,855 --> 00:07:41,631
and b) that the number of tests per person
needs to be kept at a minimum.

134
00:07:41,857 --> 00:07:43,828
But wouldn't it actually be beneficial

135
00:07:43,828 --> 00:07:46,788
if the test was performed
on a larger group of people

136
00:07:46,788 --> 00:07:51,934
and more often and over a longer period
of time for each individual?

137
00:07:51,934 --> 00:07:56,025
Especially the latter would give access
to something very valuable

138
00:07:56,025 --> 00:07:58,895
that is called longitudinal data.

139
00:07:58,895 --> 00:08:03,232
Longitudinal data is a dataset
that tracks the same patient

140
00:08:03,232 --> 00:08:07,617
over the course of many months or years.

141
00:08:07,617 --> 00:08:10,316
Nowadays, medical decisions

142
00:08:10,316 --> 00:08:12,682
are often based on a limited dataset

143
00:08:12,682 --> 00:08:15,955
where over a glimpse
of a patient's medical history

144
00:08:15,955 --> 00:08:18,701
is available for decision-making.

145
00:08:18,701 --> 00:08:22,660
In such a case,

146
00:08:22,660 --> 00:08:23,613
abnormalities are typically detected

147
00:08:23,613 --> 00:08:27,812
by comparing a patient's health profile

148
00:08:27,812 --> 00:08:30,774
to the average health profile
of a reference population.

149
00:08:30,774 --> 00:08:34,571
Longitudinal data would
open up a new dimension

150
00:08:34,571 --> 00:08:37,487
and allow abnormalities to be detected

151
00:08:37,487 --> 00:08:41,105
based on a patient's own medical history.

152
00:08:41,105 --> 00:08:45,068
This will pave the way
for personalized treatment.

153
00:08:45,068 --> 00:08:47,084
Sounds pretty great, right?

154
00:08:47,084 --> 00:08:49,216
Now you will certainly have a question

155
00:08:49,216 --> 00:08:51,039
that is something like,

156
00:08:51,039 --> 00:08:55,596
"If the technology is as great as he says,
then why aren't they using it today?"

157
00:08:55,596 --> 00:08:57,799
And the only answer I can give you is,

158
00:08:57,799 --> 00:09:01,247
not everything is as easy as it sounds.

159
00:09:01,247 --> 00:09:03,456
There are technical challenges.

160
00:09:03,456 --> 00:09:07,865
For example, there's the need
for extremely reliable sensors

161
00:09:07,865 --> 00:09:11,075
that can detect mixtures
of volatile organic compounds

162
00:09:11,075 --> 00:09:13,840
with sufficient reproducibility.

163
00:09:13,840 --> 00:09:16,414
And another technical challenge is,

164
00:09:16,414 --> 00:09:21,445
how do you sample a person's breath
in a very defined manner

165
00:09:21,445 --> 00:09:23,680
so that the sampling process itself

166
00:09:23,680 --> 00:09:27,440
does not alter the result of the analysis?

167
00:09:27,440 --> 00:09:29,766
And there's the need for data.

168
00:09:29,766 --> 00:09:33,748
Breath analysis needs
to be validated in clinical trials,

169
00:09:33,748 --> 00:09:36,462
and enough data needs to be collected

170
00:09:36,462 --> 00:09:41,516
so that individual conditions

171
00:09:41,516 --> 00:09:44,504
can be measured against baselines.

172
00:09:44,504 --> 00:09:46,430
Breath analysis can only succeed

173
00:09:46,430 --> 00:09:47,113
if a large enough dataset can be generated

174
00:09:47,113 --> 00:09:51,443
and made available for broad use.

175
00:09:51,443 --> 00:09:55,047
If breath analysis
holds up to its promises,

176
00:09:55,047 --> 00:09:57,505
this is a technology
that could truly aid us

177
00:09:57,505 --> 00:10:00,059
to transform our health care system,

178
00:10:00,059 --> 00:10:03,798
transform it from a reactive system

179
00:10:03,798 --> 00:10:06,984
that treatment is triggered
by symptoms of disease

180
00:10:06,984 --> 00:10:08,778
to a proactive system

181
00:10:08,778 --> 00:10:12,373
where disease detection,
diagnosis and treatment

182
00:10:12,373 --> 00:10:14,362
can happen at early stage,

183
00:10:14,362 --> 00:10:17,901
way before any symptoms occur.

184
00:10:17,901 --> 00:10:19,835
Now this brings me to my last point,
and it's a fundamental one.

185
00:10:19,835 --> 00:10:24,631
What exactly is a disease?

186
00:10:24,631 --> 00:10:28,920
Imagine that breath analysis
can be commercialized as I describe it

187
00:10:28,920 --> 00:10:31,912
and early detection becomes routine.

188
00:10:32,627 --> 00:10:34,225
A problem that remains

189
00:10:34,225 --> 00:10:37,967
is in fact a problem that
any screening activity has to face,

190
00:10:37,967 --> 00:10:39,922
because for many diseases,

191
00:10:39,922 --> 00:10:43,748
it is often impossible to predict
with sufficient certainty

192
00:10:43,748 --> 00:10:46,686
whether the disease
would ever cause any symptoms

193
00:10:46,686 --> 00:10:49,200
or put a person's life at risk.

194
00:10:49,200 --> 00:10:51,783
This is called overdiagnosis,

195
00:10:51,783 --> 00:10:54,137
and it leads to a dilemma.

196
00:10:54,137 --> 00:10:56,370
If a disease is identified,

197
00:10:56,370 --> 00:10:58,313
you could decide not to treat it

198
00:10:58,313 --> 00:11:02,943
because there's a certain probability
that you would never suffer from it.

199
00:11:02,943 --> 00:11:04,688
But how much would you suffer

200
00:11:04,688 --> 00:11:08,165
just from knowing that you have
a potentially deadly disease?

201
00:11:08,165 --> 00:11:12,312
And wouldn't you actually regret that the
disease was detected in the first place?

202
00:11:13,604 --> 00:11:17,403
Your second option
is to undergo early treatment

203
00:11:17,403 --> 00:11:19,473
with the hope for curing it,

204
00:11:19,473 --> 00:11:23,491
but often this would not
come without side effects.

205
00:11:24,225 --> 00:11:27,961
To be precise,

206
00:11:27,961 --> 00:11:30,941
the bigger problem is not overdiagnosis,

207
00:11:30,941 --> 00:11:32,281
it's overtreatment,

208
00:11:32,281 --> 00:11:35,407
because not every disease
has to be treated immediately

209
00:11:35,407 --> 00:11:39,221
just because a treatment is available.

210
00:11:39,709 --> 00:11:42,711
The increasing adoption
of routine screening

211
00:11:42,711 --> 00:11:45,375
will raise the question:

212
00:11:45,375 --> 00:11:49,407
what do we call a disease
that can rationalize treatment,

213
00:11:49,407 --> 00:11:52,323
and what is just an abnormality
that should not be a source of concern?

214
00:11:52,323 --> 00:11:59,600
My hopes are that routine screening
using breath analysis

215
00:11:59,600 --> 00:12:03,056
can provide enough data and insight

216
00:12:03,056 --> 00:12:07,994
so that at some point,
we'll be able to break this dilemma

217
00:12:07,994 --> 00:12:10,895
and predict with sufficient certainty

218
00:12:10,895 --> 00:12:15,216
whether and when to treat at early stage.

219
00:12:15,216 --> 00:12:18,913
Our breath, and the mixture
of volatile organic compounds

220
00:12:18,913 --> 00:12:20,143
that we exhale,

221
00:12:20,143 --> 00:12:25,495
hold tremendous amounts of information
on our physiological condition.

222
00:12:25,655 --> 00:12:30,203
With what we know today,
we have only scratched the surface.

223
00:12:30,203 --> 00:12:34,523
As we collect more and more data
and breath profiles across the population,

224
00:12:34,523 --> 00:12:39,111
including all varieties of gender,
age, origin and lifestyle,

225
00:12:39,111 --> 00:12:42,407
the power of breath analysis
should increase,

226
00:12:42,407 --> 00:12:48,023
and eventually breath analysis
should provide us with a powerful tool

227
00:12:48,023 --> 00:12:52,449
not only to proactively detect
specific diseases

228
00:12:52,449 --> 00:12:57,392
but to predict and
ultimately prevent them,

229
00:12:57,392 --> 00:13:00,826
and this should be enough
motivation to embrace

230
00:13:00,826 --> 00:13:03,992
the opportunities and challenges

231
00:13:03,992 --> 00:13:06,654
that breath analysis can provide,

232
00:13:06,654 --> 00:13:10,391
even for people that are not
part-time hypochondriacs like me.

233
00:13:10,391 --> 00:13:12,740
Thank you.

234
00:13:12,740 --> 00:13:16,757
(Applause)