 Good. Okay. So I modified the title a little bit. I just found out this morning that I need to leave this evening, and so I figured that I would give you my entire input in this 20-minute section since I won't have time for dialogue, and since I wasn't in the previous session because I was rearranging flights. Sorry about that. So just to summarize, I think a key issue at this juncture for us all in terms of implementation of genomic medicine is implementation science, and this one slide kind of summarizes the points that I'll make that really we need for major investments if we're to go from our current stage of pioneering genomic medicine into the early adoption phase. So this is where we're at under, you know, at least from my standpoint, that we have been innovators and now we're starting to recruit early adopters, but we're very, very far away from seeing generalization of genomic medicine in routine practice. And so the strategies that are necessary, business logic has taught us, are quite different as a pioneer and an innovator from an early adopter from getting into the large majority or the early majority. Different strategies and approaches are needed. I'm going to focus on one homogeneous setting where I think genomic medicine makes sense today, and it is timely to talk about scale, and that is NICU and PICU, neonatal and pediatric intensive care units. Essentially, those are homogeneous setting, acutely ill infants, many of whom will have genetic diseases. This is the outlook for California, 35 million residents, only a total of 47 units that would need to be converted to genomic medicine to deliver genomic medicine to that entire population in this setting and affecting about 30,000 kids per year. So not something that's outlandish in terms of could we potentially do this within the next year or two. It's actually quite tractable, only 47 units. The economics are very favorable for implementation of genomic medicine in this setting. These are charges that are quite outdated, 2009, so six years old, and reflect level two through level four care, and at least initially, genomic medicine for neonates would be limited to level three and level four, where charges are going to be on the high end of these numbers. So the idea of implementing genome sequencing or exome sequencing at scale really would be a rounding error. So let's just look at the data for a second. Nothing like data. This is a unique area of need for genomic medicine. I know that the president's focus is heavily on adulthood cancers, but really if you think about genetic disease burden, the greatest genetic risk, obviously, is that conception, and then it sort of erodes exponentially through birth, and then it has a massive uptick again, because suddenly this human being has to operate without mom's accessory organs. And so with this massive uptick at birth, again, of the burden of genetic disease, and then again an exponential decay until we hit somatic mutation time late in life. So this really is a place of huge genetic disease burden. So let's look at the data. There really are only two published papers that I'm aware of that really look at whether we are ready for prime time in terms of delivery of genomic medicine in this setting. The first, just from last month, genetics in medicine, a group out of Melbourne, Australia, 80 infants, singleton exome sequencing delivered to them, and then all standard conventional testing. And the diagnosis rate, 58% by singleton exome versus 14% by collapsed all conventional tests. That's cytogenetics, arrays, fish, sanger sequencing, that sort of stuff. However, the time to diagnosis was 134 days, which clearly is not relevant much for NICU care. But despite that, there was a 19% change in management, so somewhat paradoxical. Second study is ours, which was funded by NHGRI and NICHD, excuse me, 35 infants in a single level for NICU. Amazingly enough, exactly the same rate of diagnosis. We got there the hard way with trio genomes as opposed to singleton exomes. Still not sure how they managed to get that with their singleton exomes, but hey, they must just be smarter than us. 9% by all collapsed standard methods. So again, very clear-cut improvement in diagnosis. We had troubles getting timeliness as well, not so much, but it was taking about 26 days for us on average, or actually it's the median, to enroll a baby. So they had to go through this series of steps where the diagnosis was not apparent. It raised to the level of a genetics consult, and then there was dialogue and eventually ascertainment that this baby might benefit. Despite that, though, 37% change in treatment, that's overall, that's not 37% of the 57%, that's 37% of the 35 kids. And so really, I think that we're here today in terms of our technology, that I don't know exactly where we're at on the curve, but I do think that in terms of short read sequencing, we pretty much are good enough to think about deploying this very broadly. Let's just look a little bit at what clinical utility means for a day 49 return of result. And the critical thing for me is this column here, quality-adjusted life years added. So you can read the numbers as well as I can. 17% palliative care guidance, one life saved, one patient in whom a NICU stay was expedited by a month, and two patients in whom major morbidity was avoided. And then we can add some sort of estimates of what the yield is in terms of quality-adjusted life years. Sorry. But really, if we look at something which is possible probably today, if we had ascertainment at birth, and then let's say a seven-day turnaround test, which could be quite feasible, we would achieve 94% diagnosis prior to discharge instead of 37. And just looking at the numbers based on our historical data, a 31% palliative care guidance, unlikely that we would change that one life saved to more, but possibly, but certainly we would increase the number of folk who had a prolonged NICU stay. So I think even with the data at hand, although I can't do statistical analysis on this and show statistical significance, the data is starting to become compelling in terms of there's a mature opportunity here, which we ought to consider realizing. So I just took a new job. I'm seven months there. We're starting from scratch with a brand new research institute in 2016 that exists to deliver genomic medicine to this hospital, which is the sixth largest children's hospital in California. This is our catchment area. It's about 4 million people. And so I'm thinking about these bottlenecks and problems a lot incessantly, in fact. And from a practical standpoint, how do we implement things at scale that we know are highly likely to work? We've done some modeling in terms of, again, quality-adjusted life here saved. How many sequences we would need per year to get us there? And you can see the numbers are quite remarkable. This is based on fairly conservative estimates and based on the published data in terms of diagnostic rates and then the consequences of those for care. So some things we've already figured out and are ready to go. Others we haven't. So one is that we now have methods that can give us genome sequences, and that's from consent through return of result in a timely manner, even for acutely ill infants. Second of all, we have instruments now that are population scalable. I would point out that we have a huge problem here, which is that you can either have cost-effectiveness or speed. Today we don't have both. So we need to work on that. But there's lots of bits that we haven't figured out, and that's really what I want to focus in on. So first one is timely patient ascertainment. As I told you, in the studies that we've looked at today, it took about two weeks to three weeks to ascertain a patient and enroll them for genome sequencing. One of the problems is that no phenotypic feature seems to enrich for diagnosis. And so clinicians struggle to understand where you deploy this. They're not ready yet maybe to think about giving it to all babies of, say, greater than 36 weeks gestation. And so one innovation that I think I'm keen to see us work on is the idea of a surveillance system that would trigger alerts for clinicians that a baby might benefit from a genome or an exome sequence. In order to do this, we don't need very much. We need a data model. That exists. Then we need an algorithm that exists. And then we just need to parameterize that with enough cases so that we can look at the area under the receiver operating characteristic curve and understand the sensitivity and specificity of that. This exists for sepsis. If you're a CERNR hospital in the United States, you can opt into this. And this will run dynamically on all patients under care and will send two levels of alert, the first one to a nurse, the second one to a clinician to say Mr. Jones in bed six to get checked out for sepsis. So something like this for genomic medicine, I think could be quite feasible. Second major problem we have is that we've been very successful in discovering genetic diseases. The list now is over 8,000 long. The problem being that no human being can contain that information. And his presentations, inheritance patterns, treatments, and so on prognosis and complications. Furthermore, we only have 23 genetic, 2300 genetic counselors. And so we have a gap here. I don't have the numbers for medical geneticists. I imagine it's somewhat similar. We do have these tools, which are first generation tools that can bridge that gap. These are algorithms, software tools that will take a set of clinical features and marry those to the entire OMEM database and thereby give something of a rank ordered differential diagnosis. However, one major problem with all of these is that they're not really modeled on empiric data. They're modeled on simulated data. And so two things that we really need are data driven models of genetic disease topologies in terms of the range of presentations and their temporal evolution, especially in infants and automated dynamic clinical feature extraction, very much akin to what happens with the sepsis alert. Third gap where I think it would make sense to focus some effort made by a famous by Elaine Mardis is the fact that analysis and interpretation is a bottleneck. Very fortunately, the ACMG and CAP, or maybe it's AMP, came up with computerizable guidelines for giving variant pathogenicity. Unfortunately, to my knowledge, nobody has yet automated these. It seems that most of this could be pre-computed. And if it was automated, we would be married with some of these types of tools that do a pretty good job of semi-automated, pardon my spelling there, auto-moleted, variant filtering. We could have something that could greatly accelerate the bandwidth of lab directors and other analysts. Gap 4, obviously one that we all cry much about in our tea leaves is the fact that we don't get paid for our work. But frankly, historically, what's been necessary is to have high quality evidence, which typically has meant randomized controlled studies. And I know that that's a radical idea for the diagnostics world. But we live in radical times in terms of the constraints of payors. And the other thing I think we haven't done very well is to get strong MD support from powerful MD lobbies. Much as we love geneticists, we don't have the clout of, say, cardiologists or neonatologists. And so we need to educate and garner support from our colleagues. And I don't foresee that there will be a problem with reimbursement, sort of carve outs. Once we've established that this is clinic has clinical utility and in certain settings, cost effectiveness. So to summarize some challenges and some solutions, timely patient ascertainment could be solved by an alert system, genome cost versus timeliness. Some what we just have to wait, both for Moore's law and for market forces, we need to pray for a competitive landscape, comprehensive differential diagnosis, a huge bottleneck. And as I've tried to show you, I think that we could largely automate this, certainly at first pass. And bear in mind that, you know, if we look at genomic medicine and we're talking about democratizing this, we're not talking about a few quaternary referral centers across the United States. We're looking at putting this in the hands of many, many, many physicians in hospitals who may not have medical geneticists. And so this may seem absurd to you, but I think it's going to be incredibly necessary if we're really to reach all of the needy kids. We have too few staff, expert staff, and we need targeted education of generalists, both nurse practitioners, certainly in an intensive care unit setting and an outpatient setting. And we very much need electronic clinical decision support. What I mean by that is you don't hand them a typical genetic test report, which for many MDs is largely unintelligible. What we need is clinical decision support that explains and walks them through the management plan, the counseling plan, et cetera. The last one I've mentioned. So just to get back to where I started, at four in the morning, this cartoon made a lot of sense. For the life of me, I can't understand why. So adoption of genomic medicine will require investments in implementation science. I think this is a new world. I think we've gone from basic research to translational research. I think implementation science is even further downstream. These are the components as I see it. We've got these researchers down in the left-hand corner. Many of their efforts will relate to computation. So both building software, but much more modeling large data sets to build predictive tools. We need teams and process engineering and versioning to implement at scale. That's pretty obvious. I've just mentioned this education and engagement. A lot of engagement, clinical decision support tools, and then very much, I believe, randomized controlled trials where we measure success and failure. Thank you very much. Okay. And just like for the last panel, we'll take a few minutes of questions if there are any immediate questions for Steven before we move on to the next speaker. Yes. So I have one question. So for rare disorders, when we're looking at therapies, enzyme replacement, we've used natural history data and then how you modify it. So is there a way, without doing a randomized clinical trial, you can look back at outcomes, collect that data, and then use with the new technique, rather than having to do a randomized clinical trial? Yes, there definitely is the ability to use historical information. It has caveats. It's not the gold standard, but I think it's going to be necessary, I think, to quote my colleague, John Lantos, we have extreme fragility of equipoise. And what I mean by that is that some of our neonatology colleagues are not willing to randomize to the control arm for diagnostic tests. I do think we need to differentiate between what I would call precision medicine and genomic diagnosis. And I think today is a ripe opportunity to think about implementation at scale of genomic diagnosis. I do think implementation of protocols that deliver precision medicine is a longer term need and requires a lot of thought and planning. For example, a precision palliative care plan might make a lot of sense, but that's not something I think we can just do today. It will take quite a while to get our heads around what those are and then how to implement them. So you mentioned, Stephen, that it would be relatively straightforward if laborious tasks to develop an algorithm based on real data versus speculative or simulated data. Do you have a feel for how many patients, how many centers, et cetera, would be needed in order to develop really robust algorithms that could be used throughout the country? I think provided we look at homogeneous groups and provided we ask targeted questions, the numbers are not outlandish. And so in particular, obviously my focus is on the infant at birth and within the first three months of life. So that's a very focused group. And there, I think maybe 20,000 individuals could give us a really nice, a really nice group. Having said that, you know, the historic point is a good one where we can use historic data largely to model that people have been receiving genetic diagnoses for decades. And so that could be a retrospective study. So just in follow-up, I wasn't clear if that focused target group you were looking at was all neonates or neonates and ICUs, where I see the latter as being a much more compelling argument than perhaps all neonates. Yes. Right now I think we can make compelling arguments for neonates, newborns, infants in general, in intensive care units, both PICUs and level three and level four NICUs. I'm not so sure about level two NICUs. Is there one more before we move on? Okay. I wanted to know, in your presentation, you suggested that the NICU-PICU population are mostly affected by single mutations. And I was wondering what's the evidence that that's the case, as opposed to, you know, maybe more complex genetic models of disease? Well, the evidence to date is really the two studies that I've shown you. I know that other groups have reported large series of exome cases, but they have generally involved patients in many age groups, but certainly in the two published studies that are reviewed, those were the overwhelming groups that were tractable today for genomic medicine. That's not to say that there are many, many other things that genomic information can help in that setting. Let's say nutritional status or organ maturation or early detection of onset of sepsis. You know, omic data can certainly do that. But here now today, if we're to say, what could we flip the switch on and what might the benefits be? This is a very clear cut case where I think we can make a good argument that this is ready for prime time. Okay. Thank you very much.