 that the incidence of traumatic brain injury is four times population growth because of industrialization and because of road traffic used increasing. This has a global and enduring impact. There's enormous burden of late morbidity in survivors, and even mild TBI can have prolonged symptoms. There's concerns that disability may be progressive in a proportion of patients. About 20% of patients don't get better. They don't just not get better, they actually get worse. It may be triggering some kind of a chronic process, and it's an important epigenetic risk factor for late dementia, and that increased risk is somewhere between 1.5 and 10 times the baseline risk, depending on the severity of injury and also on your underlying genotype. So we currently classify traumatic brain injury very crudely. We use the Glasgow Coma Scale, and it's the 40th anniversary a couple of years ago, and basically it looks at whether you open your eyes spontaneously to speech, to pain or not, verbal response where you're oriented in time, place and person, confused and appropriate, incomprehensible sounds or none, motor response, you obey commands, localize pain. If you get poked somewhere, you go to that point, you withdraw from pain, you have abnormal posturing, or you don't. So the point about this is the two points to make. One is that even a chair has a Glasgow Coma Scale of three, you can't get down to zero. And the second thing is that this is for an adult. Obviously if you're a teenager, the highest you get to is you don't open your eyes to anything before 12 o'clock in the morning, you never obey commands, and the best responses you can get are incomprehensible grunts until sort of late evening. So it's age specific, so that's important. But then we classify a traumatic brain injury based on this as mild TBI where the Glasgow Coma Scale is 13 to 15, and typically these people are conscious but maybe confused. But even with this, somewhere between 10 and 30 percent of patients are disabled, cognitively disabled at three to six months, moderate TBI with the Glasgow Coma Scale of nine to 12, and 50 percent of these people are disabled at six months, and severe TBI where people don't respond to commands and are essentially in coma after the traumatic brain injury. There's 75 percent death or disability at six months. So it's a bad disease, but you can make it better. We know that simple intensive care can improve these outcomes substantially. The point is we want to find out what the processes are and be able to match treatments to patients so that as I'll show you later, we're using more precision medicine. And this is the problem. So these are patients with CAT scans of the brain, all of whom have been diagnosed as having severe TBI, and that's based on the fact that you say lift up your arm and they don't lift up their arm. And they're not opening their eyes spontaneously and they're perhaps just grunting. So they have the same behavioral response, but they have very different underlying pathologies. This is what's called an extradural hematoma. It typically is associated with the fracture of the skull and the tear of an artery, very little underlying brain injury. So if you take out the blood clot, the person will do dramatically well because they don't have underlying brain injury. These are contusions with an extradural on the other side. This is diffuse brain swelling on this side. This is diffuse brain swelling with a subdural hematoma on the surface of the brain that's usually associated with brain injury. This swelling here is not something that we have a good treatment for. We've got a whole load of medical treatments which are in our algorithm here. But we don't understand the pathophysiology well. When we hear about cancer doctors going and saying we're going to look at the genomics of the tumor and treat patients appropriately, we are hugely envious because for us the genome is the host genome. It's a response to injury that's different. And this classification takes no account of those variations in either the disease or the host, both of which are very important. And it combines variable prognosis, very different pathophysiology and different treatment needs. And as a result, though we have multiple therapies in use we don't have really high quality evidence for any of those. And what we would like to do is to do this to practice precision medicine where we match patients to the most appropriate therapy. But what we're currently doing is a bit more like this. It's from our one size fits none line. So we use all of the treatments and all of the people not based on whether they're most precise for that patient but because we know that they're the least toxic or the least side effects. But that may be the wrong thing because if the disease is really severe then maybe we should be using some of those more powerful and potentially toxic treatments with more side effects earlier so that you limit the progress of pathophysiology. So for this reason we asked for funding for center TBI and my partner in crime for this is Andrew Maas who's actually the senior coordinator. I'm the co-coordinator. So if anything goes wrong, his next on the block. So we got this and essentially it's precision medicine and comparative effectiveness research program which involves over 80 scientists from 43 institutions and INCF is up at that corner for a reason you'll see in a second because we also have over 17 investigating centers which are recruiting patients to these studies and one of the things we need to do is to find all the data bring it to one place and analyze it and see the patterns that would allow us to differentiate patient subgroups in order to be able to exercise precision medicine. And INCF have been the key people in that context with generous funding from one mind and we've been trying to make this happen. We're in the process where we're two-thirds towards recruiting all of our patients and then we have to follow up and curate the data but we're making substantial and rapid progress. So this is the study. It collects data from 5,400 patients, 1,800 in each stratum so not by mild moderate or severe we're also convinced that it's not just specific therapies but systems of care that make a difference. So if you have patients who are rapidly picked up from the roadside come to the emergency department quickly or triage rapidly and appropriately have surgery or go to intensive care as they need very quickly even if you're not doing anything different the fact that patients move slickly through that pipeline of care means that the outcomes are much better. So we want to see how patients are treated in different areas and find out whether for example having a trauma team arrive at the emergency department makes a difference and that may be much more important than having some specific treatment or not. So we've got the stratum from the ER, the admission and the ICU stratum but we're also collecting data on the centres so we have centre profiling which provides the context of care the healthcare systems, the institution and unit characteristics the population serve, the protocols and practice so those broad brush questions and it's really important because for example there's a suggestion that if you have really good general intensive care for the severely injured patient who turns up then your traumatic brain injury outcomes are likely to be better but also if you have specialist neuro intensive care experience that may improve outcomes and we need to understand how those two balance out. In the background we're also collecting data from patients who are not consented but only administrative data that's providing registry based data which doesn't need regulatory approval which needs regulatory approval but we have approval not to get consent and then we have sub-studies so this is very broad brush data, no great detail and sub-studies where we're collecting really highly dense data very granular data so there's an MR sub-study which is acquiring data from about one third of the patients at three time points high resolution ICU data really only in about 10% of the overall population probably closer to 20% of the ICU patient where we're collecting high resolution up to 50 hertz data advanced hemostasis in 20 centres electrocorticography in 5 centres which are putting in dwelling electrodes and EEG in unspecified number this funding that's going forward to try and get that and then we have a whole load of embedded and associated studies so these are our research targets better characterization of disease severity, disease process and outcome looking at the effectiveness of care and treatment evidence to underpin clinical guidelines and knowledge transfer and this is how we have it this is the core data set of about 5,000 patients the study registries outside that 20 to 30,000 and then we have collaborating national registries from the UK and Germany in particular international comparisons which I'll come to later and then these enhanced data sets in about a third of the patients so what we can do because these are all if you like like Russian dolls one inside the other as we can make sure that the really high density data we collect in the MR and ICU sub studies are represented or representative of the broader populations that our core data set is representative of the study registry and participating centres are representative of national or regional registries and then we can compare they don't have to be but then we can compare different national registries and we can compare international comparisons as well and this will lead us to undertaking a variety of research outputs there's evidence synthesis which is a key part of what ANCF is contributing and there's evidence communication and evidence translation and very importantly there's legacy because we have clinical data the neuroimaging, genomics, biomarkers, high resolution ICU data clinical and patient assessed outcomes and this is the intellectual capital that will derive from the Center TBI study and will provide dividends and the people who are banking that intellectual capital for us at ANCF so I told you there are international studies the Center TBI study is one part of a whole load of international studies which form the INTBA collaborative which is International Initiative for Traumatic Brain Injury Research with all of these participants, the European Union, National Institutes of Health, CIHR from Canada, one mine I understand now that the Ontario Brain Institute is also part of this as a funding partner and what this does is to fund data collection with similar hopefully identical but never going to be identical platforms but very similar which will allow aggregation from studies in Europe, USA, Canada and what we hope is we'll be able to bring together all of these studies the direct and leverage funding thus far is probably now touching about 100 million dollars we have data on over 2500 patients 10,000 of whom will be fully characterised with genomics biomarkers and so on the involvement of Chinese, Indian sites and now South Africa is getting interested will provide a generational opportunity for global TBI research our hope is that INTBA will come to be talked of rather like the Framingham studies talked of now for cardiovascular research it will provide that kick start in the repository for us to understand TBI as a disease but using the eyes of science so these are the opportunities fantastic opportunities, global collaboration but the proof of the pudding is in the implementing and the problem is that TBI research is not an average day at the office if you're running Alzheimer's disease you say I'm going to see 1000 patients with mild cognitive impairment when am I next in clinical, right down in the diary they'll come and send an appointment, they turn up the patient sits there quietly, gives you a history, you take a blood sample and they go away and that evening you go out and have a glass of wine and dinner and say, I've reached my 1000th patient our data collection starts here this is when the patient has multiple traumas so this is one of my colleagues, Ari Erkole who also works in pre-hospital medicine took this at somewhere he went to to start off the subject in this context was head down, the cab of this lorry had flipped over and the first two hours of this whole disease activity was getting this patient out now that may seem like a needless detail what has that got to do with the research it's really important because when this guy is upside down he's got really abnormal physiology and that impacts an outcome and keeping this guy upside down with inappropriate blood pressures may completely negate anything that we do over the next, I don't know, 30, 40 days acutely and in rehab, now we can't do anything this guy is going to turn up, he's an upside down patient the only way we'd be able to get him straight if he was in Australia I suppose but the point is these are confounders that we do need to account for the patient then pitches up in the emergency department and this is a real world, this is not set up this is from Jeff Manley's centre at UCSF and you're trying to save lives at the same time as trying to do research and it is a tremendous challenge because many of the early pathological indicators are the ones that are going to drive disease and provide the most useful biomarkers for understanding disease and doing that at the same time as everything else is happening is a problem and then they come to the ICU, this is our ICU and it looks fine but the problem then is that the amount of data multiplies this is the physiological data display system that we have ICM plus derived, devised by Peter Smileski and Marek Chosnick and we get a huge amount of imaging data originally the disadvantage was that it was only CT the advantage was that it was only CT we had only one modality now we've got MR, we undertake PET studies in these patients as well from the acute point onwards and the difficulty is that these are over a thousand data points per day this is multi-episode, multi-modality, multi-sequence and highly variable from one subject to the other and then for the sorts of studies we're doing when we get to the point where we're following these patients up it's another challenge because in a European study as opposed to an American study you have many more languages so one of the first things we had to do in Center TBI was to translate all of the outcome instruments in these 22 languages and the only reason there's no, there's a red cross across Russian there is because they're not recruiting pediatric patients so they didn't have a pediatric Glasgow outcome scale so enormous number of challenges we have a huge heterogeneity of data with fuzzy definitions but broadly we've got cross-sectional data demographics, mechanism outcome neural data, small number over time generally to do with trends these are IID data these are independent identically distributed data and traditional statistics will hold and then we've got time series data which is higher sampling typically regular but up to 50 hertz temporal substructure with data within data not IID because lots of autocorrelation and these traditional statistical assumptions that we've had have been violated by those data sets imaging data provided different kind of challenge the large data sets with high spatial resolution but relatively low temporal resolution serial imaging with multiple modalities moving from CT to MR even in a research study if someone just had an MR you wouldn't take them to CT because these are sick patients you don't want to transfer them unless it's absolutely essential so you have to find some way to interleave modalities and transfer information from one modality to the other and so we have low resolution high spatial or temporal resolution as a spectrum what about the data we're collecting at the bedside so as an example starter we collect data on blood pressure and we collect data on intracranial pressure you have a probe put into the head which is measuring intracranial pressure and a severely injured patient why is that important? it's important because the difference between these two drives blood flow to the brain the cerebral perfusion pressure so mean arterial pressure minus intracranial pressure and the reason you want that blood flow to the brain is to provide glucose and oxygen to provide energy but in the patient who's had a traumatic brain injury the intracranial pressure goes up and when the intracranial pressure goes up this difference comes down and as a consequence of that you don't have oxygen and the glucose gets converted to lactic acid so this is fundamental biochemistry but it's what's happening in the brain but the great advantages we can actually interrogate this and we do, we use it as part of our clinical management in these patients so patients have with severe TBI in Cambridge and many other places will have an intracranial pressure bolt placed in Cambridge we put in a triple lumen bolt which also has a brain tissue PO2 which is telling us about oxygen levels in the brain and a microdialysis catheter which allows us to look at various metabolites in the brain and also mediators in research studies like cytokines so we're looking at all of this physiology and we can set thresholds based on I'll have to laugh or spit as I say expert opinion but we have some ideas about what are likely to be effective but we don't know what dose of abnormal physiology is responsible for outcome is it a peak is it the area under the curve is it some combination of those and similarly cerebral perfusion pressure we have certain targets that we aim for but we need to be able to empirically define how much intracranial pressure elevation and what patterns is bad so that we don't use our treatments inappropriately none of our treatments are risk free all of the treatments we use in intensive care particularly in traumatic brain g have risks of their own so there's a quote from Hamlet from Shakespeare which says diseases desperate grown by desperate measures are relieved or not at all but if you're going to use a desperate measure so one of the interventions is to take the top of the skull off because the intracranial pressure is very high a decompressive craniectomy if you do that as a first go every time the ICP goes up we know that it's not just not good we know it's bad there's a trial run from Australia which shows that it was bad we've just got the results of the rescue ICP coming out next week I couldn't possibly tell you what the results are but it's coming out next week hopefully and we will be able to find out exactly where we should be able to put decompressive craniectomy but we need to be able to quantify the physiology we need to be able to quantify that burden of abnormal physiology to do this in a rational way so we have as I said these thresholds and they're based on associations but we need to be able to do this more sophisticated and the important point is that we can interrogate the biology in more detail but that's not all we've got the intracranial pressure all the biochemistry really we've got those three lumens and it depends on what sensors you put down so we're in collaboration with iMac for example to try and develop multi parameter sensors directly but we can also correlate the different parameters I talked to you about intracranial pressure and cerebral perfusion pressure but we know that the relation between those two tells us whether the brain is auto-regulating or not and if the brain is auto-regulating that it can cope with small reductions or increases in blood pressure without having the cerebral blood flow suffer that's a huge advantage and we can find the range the sweet spot where the brain an individual brain and an individual patient is doing that if we can analyze the data better and what we also have to do by looking at these relations here between mean arterial pressure and ICP is to find ways of displaying the data which again is a really interesting thought in the context of the previous session so Marcel Aries and Ari Ercol have been using this kind of display which has still not been validated what we want to do is to make it easy for clinicians to say this is the target that we should be aiming for how are we doing, what more do we need to do to get to that target and then there's a huge amount of detail that we can explore so here we've got 1500 seconds what's that 30 minutes ish of data and what you have here is the mean arterial pressure, the blood pressure on top and the intracranial pressure below and you can see what's happened is that the intracranial pressure has jumped up here and you've got a change in blood pressure there as well you don't know which one is primary and which one is secondary but being able to look at the patterns at least asks you to ask the question we can then go in and try and find out what's happening at a slightly greater detail this is 300 seconds, 5 minutes worth of data and now you're seeing that there are fluctuations in the intracranial pressure waveform not just from beat to beat but also over time which may reflect respiratory or cardiovascular systemic variation or you can go to 5 seconds and now the waveform itself, the high resolution waveform at 50 hertz which is being sampled gives you additional information that can tell you about compliance in the intracranial cavity because when you think about it every time the heart beats it's chucking some proportion of blood so the stroke volume from the heart is 50 mils about 25% of that goes into the head, the head is a closed box so if you chuck in 12 and a half mils of blood into the head which is a closed box if the intracranial cavity is compliant the brain is lax then you'll have very small rise in intracranial pressure if it's a tight head with lots of swelling you'll have a big rise in intracranial pressure so those relations will allow us not just to tell clinicians about how bad the situation is but how bad it might become in a little while so this kind of predictive algorithm would be a really interesting thing to pursue and then we can look at more details with looking at the frequency spectrum and other behaviors both linear and non-linear and we take for this a lot of stuff from looking at time variant data in time series so the quiz that we often show is which of these is a healthy patient you guys are all informed which of these do you think is a healthy patient sorry the second one yeah so the point is this so if you look that sleep apnea this is heart failure this is preset in cardiac death and the point is this we keep talking about homeostasis as the healthy systems try and hold on to a golden mean but that's not true what you have is homeokinesis that it allows the system to flex and extend as is needed to make the system work and disease generally results in de-complexification if you look at this so one of the things we want to explore in this data is homeokinesis rather than homeostasis and that noise and someone said this earlier during the discussion that noise is not really noise there's a lot of detail that might be able to get information for so this is Ari Erkel whom I've been talking about who leads a lot of this he cheated before he came to do medicine he did a PhD in condensed matter physics in Cambridge I think that's unfair but he's really interested he's really interested in data and has been using non-linear analysis wavelet analysis and looking at things like the holder coefficient to understand the complexity of data and has been collaborating very closely with David Nelson in Stockholm through INCF and one of the great advantages of synthetic TBI is that it's brought people like Ari and the people in INCF and David Nelson together the same thing applies to imaging as well this is conventional CT imaging with an extradural hematoma CT is great because it tells you whether you want to do an operation or not now we wouldn't be able to decide that and what operation unless you have the CT scans but the problem is it's very sensitive to some kinds of pathology these are near contemporaneous CT and MR in the same patient from our institution you can see first of all that the brain stem lesion even in retrospect it's difficult to see but there's a clear brain stem lesion in the dorsolateral quadrant of the midbrain which is prognostically hugely important and then this medial temporal contusion which is sort of there has a lot more detail to do with that why is that important because the brain stem keeps you alive the hippocampus and the medial temporal lobe are going to hold your memories and cognitive capacities and a great deal of detail and understanding precisely what's happening would be really important so just to go back one this lesion here is an example of what we call traumatic axonal injury the lay person thinks that if you get bashed on the head you'll get an injury underneath it does happen sometimes but much more commonly what happens are indirect forces so first of all the brain moves inside the skull and both accelerates and decelerates more quickly so the impact on the inside of the skull results in injury because we are a gyrancephalic species you have lots of folds in the brain and in order to hold them snugly in place particularly in the anterior temporal and inferior frontal lobes you have sharp ridges in the skull and if you're trying to slide the brain or bash into those you tend to get a lot of bruising of the brain you get contusions the other thing that happens is when it accelerates and decelerates very quickly with rapid say with a motor vehicle accident the brain flicks and all of the white matter fibres that are going up into the brain are fanning out and when you flick it or when there are rotational injuries they tear and that type of injury, traumatic axonal injury is probably in patients who survive the biggest driver of the quality functional outcome and is almost not detected at all by CT scanning what we do see sometimes the blood vessels that are going alongside those radiating nerve fibres are sometimes torn and you see micro hemorrhages but that's quite crude and not very sensitive and not very specific because it may have different important effects and we need to try and understand all of that better so as an example of that what we have is here a specific sequence called the susceptibility weighted imaging this is from Siemens but other companies have their own versions and you can see even on a conventional flare sequence which is what most people would use to look for pathology you've got a little bit of lesion here but that's because of the ICP bolt having gone down and it's caused a big hole there which is an artifact from the susceptibility but you can see all the micro hemorrhages much more clearly because the susceptibility weighted imaging is exquisitely sensitive to those micro hemorrhages so this is a more sensitive but indirect measure of traumatic axonal injury but we can use diffusion weighted imaging to actually image that white matter abnormality with traumatic axonal injury and try and quantify it better to find out what the burden now of disconnection is that's leading to disability and you can look also at the evolution of pathology so here this is 14 hours after injury a shearing contusion here because the different layers of the brain have moved at different rates and essentially there's been a tear in the fabric of the brain there and you can see that it's laid out in a way which suggests that there were rotational forces involved so that's the anatomy this is a pet several blood flow scan also performed at between 12 and 16 hours after the injury and you can see the abnormal pathophysiology goes far beyond the local anatomy and this is a diffusion weighted imaging map so this is work from Virginia Newcomb who's a postdoctoral clinical scientist in our group but the point to make is this that if you look at the 14 hour flare image here around this contusion in the front you've got a core and around that you've got an area of pericontusional edema but around that tissue looks normal but when you look at diffusion weighted imaging and diffusion tensor imaging the characteristics there are of tissue that has cytotoxic edema that's on the verge of dying and you can see that that's likely to be because of abnormal physiology we follow the patients up the lesion expands so this is what we would call the traumatic penumbra an intervention that prevented this from going and developing into this would be really useful and we have here a biomarker that would allow us to interrogate it but quantifying this trying to find automated methods to actually look at this is not going to be easy at a later time point we might want to look at the disconnections this is again work from Virginia as part of her PhD with me what we have here is whole brain tractography and you're looking at all of the white matter face on corpus callosum there and then the radiating fibers into the temporal and frontal lobes and down here the brainstem and this is a normal subject one of our neurosurgeons so about as normal as you get so then I'm going to show you one patient out of our study at two days, one week, six weeks, six months and one year after the injury and what you can see is that there's loss of white matter from two days up to one week and six weeks look at the corpus callosum that's where it's most obvious but what is really interesting is that there's continued white matter losses you go on from six weeks to six months to one year even between six months and a year this doesn't happen in everyone it happens in ten, twenty percent of patients and there are pathological studies which suggest that this white matter loss is real and very importantly it correlates with the trajectory of recovery or with the trajectory of worsening that you see in these small numbers of patients suggesting that it's not just an imaging artifact but it's biologically and clinically important and finally we can try and find this is not part of center TBI but center TBI will provide the platform for designing these studies ask what is the biological process that's driving that this is group work from our group which is imaging amyloid deposition in the brain after traumatic brain injury patients deposit amyloid in the brain even young patients teenagers, children within hours of traumatic brain injury why does that happen? Well you've got all this axonal tearing, amyloid precursor protein gets spilt out and it's chewed up by processing enzymes by secretases and the clearance mechanisms are simply overwhelmed and as a consequence of that you get build up of amyloid about a month or so after that it goes away and then by the time you look at about a year it's all cleared up but what Willie Stewart whom we collaborated with here has shown from his group in Glasgow is that if patients die many years after TBI 20, 30 years they show age related amyloid plaque deposition but it's accelerated by 10 to 20 years so the biological burden, the history that these brains have suffered in some ways has made them more susceptible to more accelerated age related loss and what we've shown here is not something new as a finding but shown that the same pet ligands that are used for imaging amyloid in patients with Alzheimer's disease can image the amyloid in patients with TBI there's a slight side story to this this control is me and I was the first person to be imaged in Cambridge with PIB with amyloid and I saw all this and I thought what does that mean have I got Alzheimer's or is this just normal luckily it was just normal though I'm scared to have another one in case things have got much worse there we go so this is all great but the question with all of this imaging is this too much of a good thing potentially sent to TBI is likely to provide over 7000 CT images up to 5000 MR images we can't manually analyze this the problem is when we did the contusion study and the DTI study with Virginia she sat down drew the ROIs we can't do that with 5000 patients 5000 scans and also it wouldn't be objective because we need something that's much more objective there has been automated analysis in other settings for example Alzheimer's disease and stroke but we have a particular problem with TBI there's no prediction by pathology we know that the BRAC classification says that the sequence of events in Alzheimer's disease follows a certain pattern in stroke we know it's a middle several artery stroke this is the bit that's going to be at risk with TBI it's distributed across the brain the mechanical insult is unpredictable and the key driver of disability is very subtle we need advanced imaging to look at it TBI affects the whole brain it's very unusual to find any bit of the brain that's normal and that's being recognized now in animal models as well there's substantial distortion of the anatomy at late stages so co-registering these brains is a big problem this makes anatomical segmentation and volumetry very difficult so some examples these are the intensity profiles from lesion and healthy tissue and you can see the huge overlap for T1 T2 flare and diffusion weighted imaging lesions vary in size and shape this if you like is a map across 110 patients where lesions are most likely to be for conclusions not for traumatic axonal injury and you can see that the frontal and temporal parts of the brain are most common and hand engineering features for detecting these is going to be very difficult so the person who's come to our rescue there is a collaboration with the bio-media group at Imperial College led by Daniel Rookert and the person who's leading this part is Ben Glocker who used to be at Microsoft Research and is now a lecturer at Imperial College and what he's been using is deep convolutional neural network processes to try and understand to try and provide platforms that allow us to identify and segment lesions and also in the early layers of the neural network to try and find out what are the the various features that are being detected so we can try and relate them to how radiologists and clinicians look at it and they've been doing a grand job to the extent now that when we want to undertake further validation what I've said the cost is you do it run it through the neural network and come back and if we've got anything wrong we'll tell you there will be some places where there's a mismatch what it doesn't have is the prior knowledge that looks at susceptibility artifact from bone or from some other part of the anatomy that it's not aware of and this is an example of how it does that's the flare image on top this is the manual outlining of lesions this is Costas's convolutional neural network and this is another technique Random Forest which we had originally started with and we found that the convolutional neural network is much more accurate in doing what we're doing and then we have to look at anatomy because we want to look at whether there's volume loss or not and you only have to look at this brain coronally showing that it's completely screwed up it would be virtually impossible to take this and put this into normal core registration system to normalize it and what Justin Laidig who was an imperial has now gone on to elsewhere did was to use an expectation maximization approach to improve this it's still not perfect there are still problems but the important thing for us is to apply it to all 5000 data sets and say which are the 100 or even the 500 in which it's failed in which we can tweak it manually but all of those require a very complicated pipeline this is just for one sequence and Steph Ann Winsek Wisnek and Maria Courier from Cambridge have been putting that together and hopefully we'll have in fact we have now a pipeline for mild TBI and we're just applying it to moderate and severe TBI which is more of a challenge now a lot of what we want to do has been helped enormously by what's called the common data elements project which was run by the NIH NINDS and the Department of Defense about five to six years ago where there was a consensus on how we would describe different clinical variables and it's made a huge difference that was led by Andrew Maas and several of us who are participating in intubate and what we had were clear definitions but though that's made things better it hasn't made things perfect so for example there are ambiguities in implementation so the Glasgow Outcomes Code is our commonest way of describing disability but when we started talking to each other we found out that there was no clear consensus on whether when you describe this disability you try to parcelate out what you thought was the TBI related disability from that caused by something else so give me an extreme example if someone has a traumatic brain injury and a spinal cord injury and is paraplegic and you look at their total disability it's going to be substantial so the sensible thing there is to actually say well we're going to get rid of the paraplegia as a consequence of the spinal cord injury and quantify the disability otherwise but the flip side of that is that if the lesion is not that clear cut if you've got open fractures of the tibia, a fracture of the pelvis and the patient is not able to move about very well do you parcelate what proportion of it to the traumatic brain injury and what to the peripheral injuries so there's no right answer but we need to know how we deal with this and then most recently we've been trying to pull together the genetics of this there's a pragmatic consortium of the intubar partners and some historic collections we've got minimum phenotype and a common genotyping platform that's going to be across three centres we've taken this to the welcome trust they've approved our outland application and we've got more and more people joining so we now have 12,000 patients with this basic phenotype the final application goes in on the 11th of November if any of you are reviewers for the welcome or know anyone who's reviewers for welcome please tell them how wonderful the study is so we have these repositories and resources and centre TBI the clinical data the physiological data neuroimaging DNA circulating biomarkers and the outcome data and very importantly on a country by country basis patient identifiers looking at the long term outcome whether there's an impact on late neurodegeneration and finally we've set up research networks which is why I think in this context it's important to make the point that INCF has played a key role in this area it's put together people in Europe who've got common interest it's put together people across the globe through the interaction with one mind which I can't speak of highly enough in order to make people talk to each other more because it's in their common interest as researchers we tend to be quite selfish and it's made us realise that talking and sharing stuff is important so it's more than the sum of its parts there's the logic of common data platforms the opportunities provided by treatment variations the strength in numbers and the power of networks and I made the point about the dividends that this capital would yield and the fact that for us in centre TBI INCF are the bankers but this cost data sharing in the future costs and some of the ethical and practical issues of data sharing was something that was discussed in the FEN's symposium a couple of months ago but if you go to ADNI and you talk to the ADNI investigators they say that the data sharing that they've undertaken to provide is 10 to 15% of personal and project costs that's a lot of money and ADNI is very richly endowed they have got probably about 60 million for studying 6000 patients with diary you come and come on Tuesday oh I can't make it on Tuesday, well Wednesday I've got a party on Tuesday whereas we are having these difficult phenotypes to collect but we simply don't have the money so when we come to data share it's really important for us to find ways of resourcing it so I think this is my last slide which talks more generally but also about data integration analysis and sharing there are substantial opportunities the harmonization of data acquisition is a continuing challenge for imaging a lot of this could be done at source just like we insist that when you do a biochemical assay that you say I want the blood urea in millimoles per liter there's no reason why regulatory authorities can't impose on imaging vendors the requirement to create more coverage however even with that there would be difficulty in imaging pipelines in a severely injured brain we need to develop physiological methods analysis methods of physiological data and robust statistical frameworks for analysis I mean when we are reporting studies if you are drawing 20 conclusions from the same data using a similar analysis we are born from only corrected but if you think about the ADNI data there are probably hundreds of very similar conclusions which have interrogated the data in the same way and no arrangement has been made to correct from multiple comparisons there should you or shouldn't you, we don't know there's some basic statistics that's important we need to get around regulatory barriers the emerging EU data legislation is going to be a problem you won't even be able to send the data to the UK because it's outside the EU what with Brexit but this is really important this is why it's so important for me to come and talk to this audience there's a huge shortage of appropriately trained individuals we need data scientists to develop implement and use big data there are likely to be 4.4 million IT jobs in big data by 2020 it's been estimated and the current training pipelines will only deliver one third of these so I'm in competition not just for people to come and work with us in research but also to train clinicians and provide them with literacy to communicate with people like that and current academic credits undervalue these people and it's really important but the biggest challenge of course is you've got lots of big beasts and the whole process in centre TVI is like herding cats but there is a book that tells you about that so I'll stop there I've spoken on behalf of a lot of people seeming to my mind to pretend to be knowledgeable about things that I'm not really about but if you email me I'll direct your questions to the right people thank you very much