00:00hi everyone welcome to the a six in Z
00:01podcast I am sonal and I'm here today
00:03with our general partner for our bio
00:05friend Vijay Pandey and we have a
00:08special guest joining us we're just
00:09really talking a lot about machine
00:11learning and bio and all the trends and
00:12computer science meets bio right now
00:14it's just fun I'm excited and joining us
00:16today we have Jeff Katz who is the
00:19co-founder of a company called Q the
00:22website is Q bio they're building tech
00:24that measures digitizes and simulates
00:26human physiology so first of all how the
00:29heck do you guys do that and why does
00:31Aemon matter well I think we live in an
00:33era where the amount of information we
00:36actually collect about our physiology is
00:38like exceedingly small if you think
00:41about what the annual physical goal
00:43consists of and actually if you ask most
00:45physicians and most research there is
00:48actually no real correlation between
00:50outcomes and consistently doing any of
00:53physicals but at the same time look at
00:54something like dental care which with
00:56you're you're sick or not your mouths
00:58too sick or not twice a year you go and
01:01they measure a set of semi-quantitative
01:03standardized metrics and as a result we
01:05have these longitudinal profiles on
01:07hundreds of millions of Americans nobody
01:09goes dentist twice a year for the record
01:20you know I think if you look at the
01:23trends by most metrics even adjusted for
01:26inflation the cost of dental care is
01:28staying flat or going down but at the
01:31same time the quality of our dental
01:33health is going up which means there's
01:35an example of a system that is very
01:36personalized it's very preventative and
01:38it's really just based on a simple idea
01:41of tracking longitudinal changes in some
01:44standardized set of metrics over time in
01:46order to make forecasts about the
01:49general trajectory of your oral health
01:51yeah that's a great point because you
01:52think about you go to a dentist you get
01:54x-rays and you get them you know once
01:57every few years and you have this record
01:58over time and so you can see things
02:01develop where maybe it starts off small
02:03you don't know what it is but you can
02:04see it change and that added dimension
02:06of time I you know I think would be such
02:08a huge advantage I think we want to
02:10reinvent healthcare to be preventative
02:13believe that healthcare can be
02:15reinvented so that it gets better and
02:18cheaper over time and I think our vision
02:21for the world is really one where no one
02:24dies of a treatable disease in the scope
02:28of what is treatable is actually
02:29increasing in the scope of what is
02:30diagnosed what is creasing so we think
02:32that if we have the technology to treat
02:34something no one should die from it in
02:36medicine you know we talk about
02:38Diagnostics having a specific
02:39sensitivity and specificity and by
02:41Diagnostics you mean just tests that
02:43measure disease well that's actually I
02:45think an interesting question because we
02:46have specific opinions about what a
02:48diagnostic is but in general a
02:50diagnostic is a predictive model okay
02:52right and when we talk about machine
02:53learning if your signs like effectively
02:55you do a clinical study you come up with
02:57a variable and you say this variable has
02:59some predictive power and the way the
03:00medical community communicates this is
03:02in terms of sensitivity specificity and
03:04that tells you something about its true
03:05positive false positive or false
03:07negative rates but typically in practice
03:10the way that we apply that those
03:12clinical studies is to take a point
03:14measurement if you're typically if
03:16you're symptomatic and then compare it
03:17to some clinical study that was done
03:19potentially a long time ago on a
03:22quote-unquote representative population
03:23and if you were above or below some
03:26threshold some decision is made and
03:28that's what the sensitivity and besa
03:30facility is based on the reality is that
03:32we have a lot of Diagnostics that are
03:35recommended not to be used regularly
03:38because they don't have great
03:40sensitivity specificity as a point
03:42measurement PSA is a perfect example of
03:44this which might have in the PSA
03:46prostate antigen no it's it is the gold
03:49standard test for determining if a man
03:51has prostate cancer but it has a very
03:53high false positive rate now you know I
03:56think if you look at this as a time
03:58series right series you mean like more
04:01than one point in right then you're
04:03effectively understanding you're really
04:05looking at changes in time then and for
04:08us you know looking at changes in time
04:11is what personalized medicine is about
04:13there's a huge focus on genomics and we
04:15think genomics is important but the idea
04:18of a representative population I think
04:21is fundamentally flawed especially it's
04:23kind of ironic in the age where or you
04:26Racing the idea that everybody is unique
04:28genetic code but the reality is is that
04:30everybody's evolution of their
04:31physiology is unique to them even if
04:33you're a twin even if two people have
04:35the same genome they actually won't
04:38excessively express the same phenotypes
04:39which to us means that there's more
04:41information encoded in the evolution of
04:43your physiological state and the
04:45trajectories in your physiological state
04:47over time then your genetic code alone
04:49so that effectively from our perspective
04:51that means that human physiology is a
04:53long-tailed distribution so your genetic
04:55code cannot be the kind of you know
04:58silver bullet that is magically going to
04:59bring us to the edge of personalized
05:00medicine for us personalized medicine is
05:03much more about understanding the
05:05history of your physiology and how it
05:06changes over time and how doctors can
05:08use that to interpret that based on you
05:11know you as an individual and then your
05:13genome can add extra information about
05:16how to interpret you know whether you're
05:18not you have a high level of some
05:20biomarker based on a genetic variant you
05:23might have do you think doctors to some
05:25extent are doing this already we talked
05:27to more and more doctors who you know
05:28who in in honestly patients who say you
05:30know we would love to do this there's a
05:32few problems and the biggest one is the
05:36fundamental thing that governs our
05:37healthcare system right now is based on
05:38retro actively looking actuarial models
05:41and as healthcare costs increase that
05:44means that next year in order to cover
05:46the population premiums have to increase
05:48rather than thinking about how much does
05:51it cost to take care of a specific
05:53population because last year over the
05:56last ten years they've gotten sick at
05:57this rate and it cost us much for these
05:58drugs and these drugs are going up and
05:59so look at it more like well we know we
06:02have therapeutics for these certain
06:03diseases and that kill some percentage
06:05of people what would we have to measure
06:07and how sensitive would those
06:09measurements have to be and how
06:11predictive would the models have to be
06:13in order for us to save money in the
06:15long run rather than cut cost in the
06:18short run this is where the healthcare
06:20reimbursement model is kind of really
06:22the key thing when you reimburse for
06:25services you only think about doing that
06:28when you have a real reason and
06:30especially it puts the providers in this
06:33situation where they have to argue why
06:35this service is important to get the
06:36parish to do it and for something we
06:40it may give the impression that you're
06:42doing services that are not not
06:43necessary and the system is in the old
06:45model is really kind of stacked against
06:47this in the new model where you're
06:48trying to develop and give the best of
06:51value that's we're actually doing this
06:53could make a ton of sense what's the
06:55name of the new model so you know what's
06:57come up with the ACA and and other
06:59modifications to the healthcare laws
07:01we're being value-based versus just
07:03fee-for-service based it actually will
07:06create the opportunity where this
07:07actually makes a ton of sense because
07:08the total value to the patient the value
07:12to the payer is in making sure people
07:14don't get these diseases which are very
07:15expensive you know and add interestingly
07:17enough I almost an irony to the Nydia of
07:20value-based care is if you're going to
07:22have a value-based care model if you're
07:24not monitoring the population
07:26continuously and regularly you really
07:28don't have a way to know if you know you
07:31somebody's getting healthier or not so
07:33unless we have a way to quantify what it
07:35means to get healthier we can't even
07:37develop a metric for what value-based
07:40care is and so I think that not only
07:42does it does this make sense from it in
07:44terms of preventing illness and catching
07:47it early stages if we want to shift to a
07:49value-based model I think it's required
07:52in order of in order for us to develop
07:54metrics that would allow us to do that
07:55I'm hearing you guys say in reality
07:58right now we don't really have a way of
07:59really capturing a person's longitudinal
08:02trajectory not with clinical data like
08:04there's obviously an explosion in
08:05wearables but you know that obviously
08:07isn't necessarily always clinical
08:09quality data but or there some
08:11patchiness I mean you get a test here
08:13and there but it's it's just very
08:14anecdotal and what's what's in the
08:16record is that because they're different
08:17doctors taking the measurement or just
08:19you don't get every test on it all the
08:21right I mean you just get it when you're
08:22symptomatic typically and that's sort of
08:24when is already potentially getting to
08:26be late yeah I mean most of the most
08:29lethal diseases by the time you're
08:31symptomatic you're in an advanced stage
08:32so if we want that comes up over and
08:34over again yeah I I can't say that I
08:36know anybody who doesn't have some story
08:38of some relative or some friend going in
08:40for some unrelated test and finding out
08:44they had some terrible thing and if
08:46they're lucky it was caught early but if
08:48they weren't it was caught late and so I
08:50think one of the questions to ask is if
08:53neurology to detect these things early
08:55and we just sometimes get lucky for
08:58unrelated reasons and catch it why
09:00wouldn't we figure out a way to use that
09:02more frequently and and I think the
09:05answer actually is costs right and false
09:07positive problems two aspects are cost
09:09right so cost is one thing and false
09:11positives are another just even if it
09:13didn't have the false positives the cost
09:14could be quite and seems like it could
09:16be higher let's talk about that so let's
09:17first of all define what a false
09:19positive is and why it matters I
09:21remember them from my days of psychology
09:23studies yeah well yeah so the issue for
09:25odd cases is that you get a false
09:27positive you know where the doctor
09:28thinks you have the disease but you
09:30now would I even further segment to two
09:32levels where there's the risk once you
09:35have a false positive of having an
09:36adverse outcome because of the false
09:37pause because it because a lot of times
09:39you can have a false positive and then
09:41do another non-invasive test which
09:43eliminates the need to do something
09:45invasive the real problem is a subset of
09:48the false positives which is you know
09:50depending on you know the pathology and
09:53the diagnostic and you know what the
09:55well they're just for care it's it's
09:57different and so when we talk about the
09:59cost of a false positive you know
10:02there's the cost of any incremental test
10:04and that's one thing I think that's part
10:08is actually less concerning than the
10:10part that where a person's health is
10:12negatively affected because of the false
10:15positive false positive in terms of
10:17prostate cancer can mean a biopsy that
10:20is unnecessary in Beijing damage you
10:23know your prostate came with breast
10:25cancer and yep you know this is where
10:27times actually really interesting
10:28because if you're looking at PSA one
10:30shot you may or might not do well
10:32compared to the population but if you
10:34have the time that's a time series data
10:36you know that looks like it could be
10:38useful for improving the accuracy
10:39reducing the false positive right yeah
10:42and you know the reality is is that we
10:44don't think it's novel actually it we
10:45would argue that if you look at almost
10:47any discovery in any scientific
10:49discipline the approach almost built
10:52defined into the scientific method is
10:55you have a system you don't understand
10:57you have the ability to measure some
10:59parameters of the system you regularly
11:00measure it you build a model you see if
11:03that model predicts the next part of the
11:05time series if it doesn't you measure
11:07or you find a better model whoever it is
11:10and you iteratively get better and
11:11better and better and that's how you
11:13eliminate false positives well a classic
11:14example playing out right now and this
11:16is actually a very heated debate among a
11:17lot of all kinds of scientists doctors
11:20and various associations the
11:22recommendations for women in mammograms
11:24I think one of the issues with the tests
11:26on mammograms is the data suggests that
11:28it has too many false positives and that
11:31you think about both the cost to the
11:33system and just cost to these women that
11:35they think they have breast cancer but
11:37they don't that's been a huge motivation
11:40to deprecate the test for only women
11:43that are older but this is the
11:45opportunity we're having an improvement
11:47on the tests accuracy would be the key
11:50thing because if the test isn't accurate
11:52enough then we get into this problem
11:53where the medical system will naturally
11:56and I think correctly suggests that it
11:57doesn't make sense to get the data and
11:59this is maybe the irony of ironies that
12:01if you think of this test from a one-off
12:04it may look less useful but if you think
12:07of her from a longitudinal point of view
12:08or longitudal in my context of many
12:10other tests suddenly now this is
12:12actually potentially useful information
12:14but that's perhaps the case that has to
12:16be made and so on and so I think that's
12:17the challenge that lays ahead one of the
12:19issues I would just point out is
12:20mammograms aren't completely
12:22non-invasive yeah and so that's an
12:25example of something that I think would
12:27potentially be dangerous to recommend
12:29doing at using like a time series to
12:33increase the Sensodyne specificity you
12:34know I don't think that's essentially
12:35true for all of the imaging modalities
12:37but I think it's that is important going
12:40back to thinking about it from a data
12:42science perspective and you know built a
12:44number of businesses based on data
12:47science where you do these analysis of
12:50the predictive power or information
12:51value you might say of a single variable
12:53and I can tell you like when you talk
12:58about and I don't have the exact number
13:00with me but if mammograms have a
13:03predictive power information about in
13:04terms of defining or predicting whether
13:06or not somebody has breast cancer with a
13:0870% accuracy no data scientist in the
13:10world would throw out a variable that
13:12gave you that amount of information they
13:15would just say well we need to find the
13:16other ones to combine with this that
13:19have some orthogonal information
13:21exactly that will bring us to a much
13:23higher level and we don't have a way of
13:24doing that right now well I would argue
13:26we do we just don't do it that's a
13:28really interesting think we have to you
13:31know it's a combination of this idea of
13:32how can you increase the sensitivity and
13:35specificity of Diagnostics and when I
13:37say and I want to make one other point
13:39that's I think really critical is and
13:41one other issue we have with even the
13:44notion of a diagnostic in medicine is
13:45that you know when we talk about
13:48measuring a system and modeling it that
13:50requires D conflating two ideas that
13:53currently kind of bound together in
13:55medicine which is in each other just the
13:56simple idea of a diagnostic is actually
13:59two separate ideas that are jammed into
14:01one thing and that is a measurement and
14:03then an analysis know one of those
14:05things is immutable in some sense it's
14:08can be done once it doesn't even matter
14:10if you do it wrong you can't change it
14:12was just done an analysis can be rerun
14:15like right now in medicine a lot of
14:18times we store the result of a
14:20diagnostic but we don't keep the
14:22measurement which actually prohibits us
14:25from going back in building models that
14:27you know in reusing the measurement
14:29right to prevent door of data exactly I
14:33come from the world of developmental
14:34psychology and our bread and butter is
14:37longitudinal studies where you track a
14:39human beings progress over time that's
14:41only way you can get the information
14:42people would actually do these things to
14:44try to control for genomics by using
14:46twin studies in the interesting thing is
14:49is that there are fields that are
14:50actually biological that do this too so
14:52I you know sometimes I hear oh that
14:54works in physics and chemistry where
14:56there's you know there's not the
14:57biological or life element but ecology
15:00from predicting population growth or
15:01deforestation we use longitudinal data
15:05predict what the climate is going to be
15:06in a hundred years we use it to predict
15:08what the weather is gonna be like next
15:09week yeah for some reason this has
15:11completely been you know ignored for the
15:14purpose of healthcare and I think that's
15:16why and when we look at success of
15:19something like dental care we believe
15:22that it's possible if you step back and
15:24think about and kind of redesign the way
15:27health care is delivered especially at
15:28the primary care level and the
15:29relationship between your primary care
15:30doctor and you the health care can get
15:34time it's really about creating a
15:36positive information feedback loop you
15:38know not to go you know harp on this one
15:40dental care because we all got our teeth
15:44there is dentist even our trained now
15:46and know that they are effectively the
15:48frontline of health care there are
15:50papers and research studies that show
15:51that there's a lot of diseases you know
15:53that somebody can be you know have no
15:56symptoms for like cardiac disease that
15:58dentists get the first look at because
16:00that's correlated to gum disease so the
16:02next question to ask is well how do we
16:04know that you know gum disease can be an
16:06early indicator of cardiovascular
16:08disease you know we would argue that
16:10it's not because your gums are the best
16:12biomarker for determining whether your
16:14heart is healthy or not but we have the
16:16most longitudinal data on our mouths
16:18which makes it easy to correlate to an
16:20outcome so so our question is is what
16:23are the if that's not the first order
16:25biomarker what are they and why aren't
16:27we tracking those right and how many
16:29people could we save if we figure out
16:31what those first-order biomarkers are
16:32because they're likely inside your body
16:34not on your gums I mean part of the
16:37other challenge even beyond value versus
16:39fee-for-service is really demonstrating
16:41that this will affect value and this
16:43possibly is a chicken and egg problem
16:45right because you need the data to prove
16:47the value and you need the sense of
16:49value to get the data you know so so how
16:52does one deal with that I think that's
16:53more than anything an ethical issue hmm
16:55and you know that's something we spend a
16:57lot of time thinking about we
16:59fundamentally believe that there are a
17:02growing number of people and there's I
17:04think a lot of data that supports us not
17:06just patients but actual doctors who
17:10believe that a patient has a lot more
17:12rights to information about their body
17:13right now there is a to some degree
17:17there is a paternal position taken by
17:20you know organizations like the AMA that
17:22says there's certain information that is
17:24dangerous for you as a consumer to have
17:26about your own body we find that a
17:27little bit ironic considering we also
17:29live in a country where you're allowed
17:29to drink and smoke and do things that
17:31definitely damage your health but we're
17:34also we're at the same time told that
17:35there's certain information we're afraid
17:37to give you because you might misuse it
17:39so yeah I had a person of cigarettes
17:41talk about the chicken egg problem we
17:43think a big part of that is changing
17:45what a patient's rights people should
17:46own and controlling from
17:48about their own bodies and have a right
17:49to whatever information they want about
17:51it especially if it's can be
17:53non-invasively gathered and you know
17:55they're willing to pay for it that also
17:56suggests the obvious opportunity the
17:58ability for patients to better use the
18:01data and have something to do with the
18:02data I think might be part of the
18:03solution I'm still not clear though on
18:05where this data is coming from how is
18:08this data gonna get into the system this
18:10longitudinal tracking and how does it
18:12affect say a concrete case like prostate
18:14cancer hope this isn't too too long
18:16we're gonna answer which it's kind of
18:18taking it one step back this year is the
18:20200th year anniversary of the stuff is
18:21really 200 that is used in the annual
18:30physical what it's true and let's pause
18:33on that for a minute
18:33so the 200 year old stethoscope first of
18:37all the thing is 200 years old that's
18:38shocking and it's pretty much in the
18:40same form for the most part than it does
18:42it was 20 years ago there's been some
18:43modifications where you can now hook it
18:45to your iPhone and it is the most
18:51advanced piece of technology in a
18:53regular physical I don't know if it's
18:55the most advanced and at the light LED
18:58that's probably slightly more you know
19:01the the frequency of light coming on the
19:03LEDs not the point of an actual
19:06measurable but for the most part like
19:08those are kind of the cutting edge tools
19:11that we use to assess our general
19:13physical health if you think about when
19:15that was that we did not have the
19:17technology there tools to non-invasively
19:19measure anything about our physiology we
19:22didn't have the ability to look inside
19:23of our bodies at high resolution without
19:25ripping you open I have a feeling that
19:28the first dentist kind of said well what
19:31how are we going to assess somebody's
19:32health no said well we can just open
19:33their mouth and look you know doctors
19:35didn't have that when the physical was
19:37invented doctors couldn't just say well
19:38we're just gonna open your chest up and
19:40look now we have an unbelievable set of
19:43technologies that allows us to look at
19:46our physiology at everything from an
19:48atomic scale to at varying degrees of
19:50scales it's certainly this explosion in
19:53omics genomics is just the tip of the
19:54iceberg it's the tip of the iceberg
19:56actually just in terms of information
19:57content your genome only has about 10 to
20:00in it we think your physiological state
20:02at a point time has probably 10 to the
20:0418th more information than it which is a
20:08million logical state ferrous conjecture
20:10that the way you define the complexity
20:13of a system is in terms of the number of
20:14bits it takes to represent its state the
20:17complexity of the genome is on the order
20:18of 10 to the ninth you know the best
20:20estimates we have for a complete
20:22representation of your physiological
20:23state at a point in time is that it's a
20:26million trillion times more information
20:29Jesus like every voxel that describes
20:32you although you could argue it's really
20:34the diffs right so the genome the the
20:36you know maybe there's a million base
20:38pairs not a billion in the you know are
20:40actually a million between us and chimps
20:42so even less between you and I but then
20:44and then the voxel difference might be
20:45less but still it's huge the voxel you
20:47know is like a 3d PDP Majan you could
20:51break us down in our bodies into a whole
20:52bunch of 3d pixels what the information
20:54content of that is and that's you know
20:56that's about how your arteries are
20:58getting clogged that's about how you
21:00know you're getting tumors that's about
21:02you know all these things
21:03we now have genomics transcriptomics
21:05proteomics metabolomics microbiome x you
21:09know so you could imagine dividing your
21:11body up into you know as small volumes
21:13as you want and then say well I want to
21:16do a proteomics on just this point in
21:17space and if you continue to squeeze the
21:19size down and down and down down you can
21:21start to imagine the amount of
21:22information that's contained in your
21:23body and don't forget the fact that your
21:26genome while it is relatively static you
21:29know and you do have mutations in it the
21:31environment that your genome is in or
21:34things like methylation which actually
21:35changed the way your genetic code is
21:37interpreted but doesn't necessarily
21:39change your genetic code all those
21:41things are part of your physiological
21:42state that accumulate over time so and
21:45and we would argue that it really is
21:47part of the explanation so for why twins
21:49diverge you were constantly interacting
21:51with your environment in accumulating
21:53complexity we keep on thinking about DNA
21:55as being fixed but our DNA is changing
21:57over our lifetime and that a little
21:59little tags get put on that affects how
22:01the DNA behaves and this is where twins
22:03can start off with the same base DNA but
22:05the methylation can change very much
22:07over time your point is that even with
22:09that methylation there's just a
22:11physiological pattern that we express in
22:13that it's more indicative I know like
22:16when we talk about 10 to the ninth bits
22:18in your genome that doesn't include the
22:21complexity that's added by the number of
22:24ways methyl groups can attach to your
22:26DNA and how that that is different in
22:28every cell in your body and you have 5
22:30trillion cells and then another 5
22:32trillion that are microbiome that's only
22:34the genomic so start to see I think so
22:38if you actually think about the
22:39percentage of the complexity of your
22:41physiologically physiological state we
22:43measure on an annual basis I don't think
22:45it's any surprise that we have very poor
22:48ability to predict when we get slit
22:51here's the the the key part which
22:53connects it all together which is that
22:55it seems like a lot of bits but the
22:57fraction that's changing over time is
22:59relatively small and so would you go
23:01from is what may look like a lot of data
23:03but might be a lot of noise by having
23:06the change over time suddenly this huge
23:08amount of bits goes down to be
23:09relatively small I have a background in
23:10particle physics and that's you know
23:12what was all about is it's everything's
23:14relative there is no absolute doing a
23:16clinical study and coming up with an
23:18absolute threshold is inherently we know
23:22from a lot of other Sciences less
23:23sensitive than tracking deltas of in in
23:27the same system in the realities were
23:28all different systems and we should
23:30probably be treated that way I think
23:31there is also information about in the
23:33population analysis right but I think
23:35it's the kind of combination of being
23:37able to statistically analyze a person's
23:40changes with respect to the entire
23:43population but also within the context
23:45of their past healthy self and I think
23:48those are the kinds of things that we
23:50can do today to improve the sensitivity
23:54specificity zuv tests we have and which
23:57hopefully would reduce you know if we
24:00had perfect tests obviously people would
24:02be worried about doing them as much as
24:04we wanted and I think that that is you
24:06know a you know a you know a noble goal
24:09but you know I think we definitely know
24:11how to set up systems that have this
24:14positive information feedback we can get
24:16better and better I mean Facebook and
24:17Google almost every modern business
24:20that's made like large modern business
24:22and tech is based on the idea of
24:24creating a positive feedback loop right
24:27product gets better the more people use
24:28it and there's really no reason so just
24:32concretely it's a very simplified
24:33example my hemoglobin is low but there
24:36are other people who may have the same
24:37exact absolute level of hemoglobin for
24:39them it's yeah it's broadly speaking low
24:42but it's not as dramatically low as -
24:44because mine is significantly lower than
24:46its baseline that they've tracked over
24:48time because they're taking it down a
24:49blood does overtime and that's sort of
24:50like a simplified way of thinking about
24:51once you have a baseline down you can
24:53get a sense that you are getting into
24:55some danger zone without the level
24:57needing to get high in some absolute
25:00level I think that brings up the second
25:02part of this is so you can add time as
25:04another dimension to any specific
25:06biomarker and increase the sensitivity
25:08specificity the other thing you can do
25:10is have multivariate models for
25:12pathology it's kind of ironic that we
25:14live in a world where we use millions of
25:17variables to predict which add you're
25:18gonna click on whether or not you
25:19deserve to get a loan right what movie
25:22you might watch next but we want to
25:24reduce things like cardiovascular
25:25disease to one or two variables yeah
25:27that's so true it's a little insane that
25:31the most mundane and yes still important
25:33in your day to day routine things in
25:35your lives have so much data going into
25:37them and to your point multi multiple
25:39variables over time so combining all
25:42these things but for our physicals and
25:44our exams it's like - your hemoglobin
25:49being at a certain level it could be low
25:51for you and there's an enormous number
25:54of reasons why that could be I know the
25:55reason it might be measurement error but
25:57the point is is that we need to move
25:59beyond and take you know what has been
26:02tremendously successful in almost every
26:04other domains of science in terms of
26:06modeling you know measuring and
26:07iteratively modeling systems we need to
26:09apply that to our bodies I think that
26:11there's a little bit of reticence to do
26:13that because people think oh our bodies
26:16are so complex you know we'll never
26:18understand it and I think we have to
26:20decouple our thinking part of our body
26:22from the physiological state which is
26:24governed by the same laws of physics in
26:26all of things and also because it's
26:28complex is why we need all the data I
26:30think the separation of measurement and
26:34analysis is absolutely critical
26:37and you know if and I think it's one of
26:39the reasons that I think we have issues
26:41and I know there's a lot of comb you
26:42spend a lot of money throwing a lot of
26:45computational power at all of the
26:47medical data that we have or technically
26:50all the analyses we have exactly in but
26:52I think part of the problem is that we
26:53don't actually have the measurements
26:55right even when you get your blood
26:56glucose measured right the output that
26:59that little device gives you is an
27:02analysis it doesn't store the actual
27:04sensor readings that it used to compute
27:07that number we would be better served as
27:09a society if we recorded the actual
27:11sensor reading yeah and people some
27:13people might say oh it's expensive to
27:14store much information and if there's
27:16one thing that's like I'm confident of
27:18is it's that storage is getting freer
27:20every day it's more sophisticated by the
27:23way just a really interesting example of
27:25this I read this amazing article on
27:27Landsat and how over the last like 30
27:30years they've been you know obviously
27:31collecting all this GPS data all the
27:33satellite data but the sensors and they
27:36have all the information encoded from
27:37these sensors but they didn't have the
27:40technology to be able to read that data
27:42yet and so now they're going back and
27:45trying to figure out how to back analyze
27:4730 year old data and because they stored
27:50the information in its original form in
27:53the sensors they're able to now go back
27:55and do these weird things like bake the
27:57data at like literally bake it what's at
28:01the end of the day you know if if we can
28:03do this and transform the kind of way we
28:05capture information in the healthcare
28:08what we really need at the you know a
28:10lot of people are organ donors
28:12what we need is data donors right and we
28:15need ways that I can put on my license I
28:17am anonymously donating my health
28:20outcome what from whatever I die from as
28:22well as the longitudinal profiling of
28:25myself because if I don't need an organ
28:28I can save one person's life if I donate
28:30my data I can help save the lives of
28:32every person who ever is born ever after
28:34me and that's a network effect that's
28:37how you as a person can improve the
28:40health of every generation after you yep
28:43that's amazing and how do you then
28:44control for the fact that the sample
28:47because you're not getting to a
28:50is so small like you might have a few
28:52random people who might take on some
28:54project like donating their data how
28:56many people do you need 4 yeah exactly
28:58that's a really good question there will
29:01be a lot of debate on on what that is I
29:02mean just to give orders magton my gut
29:04feeling is a thousand would still be a
29:07I mean you compare that to people who
29:09donate organs it's a relatively small
29:11fraction you know I mean a million would
29:14be a dream but like thousand ten
29:17thousand or think about the people who
29:19donate their bodies to be cadavers what
29:21if if if those same people I would much
29:24rather donate my data I feel a little
29:26bit weird about during my body I would
29:28love to see the government sponsor a
29:31public data repository where people
29:33could donate data and they could put
29:36like at you go to the DMV and you'd say
29:38are you a data donor and your health
29:40care data if you die is released heavily
29:43anonymized we're talking it and about a
29:45feedback loop for Humanity it's amazing
29:48hopefully thank you for joining the
29:49ethics in the-- podcast yeah thank you