Chapters Transcript Video Rethinking Low Back Pain: Biology, Biomechanics and Behavior Gwendolyn Sowa, M.D., Ph.D., presents at the Johns Hopkins Department of PM&R’s Grand Rounds on October 19, 2021. thanks again um for the time today. And so I'm going to talk today about how I think we need to rethink how we're approaching low back pain and so for those of you that care for patients um with low back pain, which is probably the vast majority of you um since it is such a ubiquitous problem, um how we need to take the learning that we've had in the past and combine it in novel ways so that we can actually move the field forward. And I hope that more than give you some information today, I hope that what really I'll do is provoke um some some new directions for you to be thinking about things because I think what we really need is more people working on this very complex and complicated problem. So just starting with my disclosures which are essentially related to research funding, uh government funding as well as some foundation funding as you see here and our objectives for today first will be to look at the risk benefit of current biologic approaches for dis degeneration, how mechanical loading impacts biology and responses in the disk and then finally to appreciate really the complex nature of both biomarkers and behaviors, which is quite complex in influencing outcomes for patients with low back pain. So why has progress been underwhelming? And I say this as a clinician and scientist in this space because I do think progress has been underwhelming um this low back pain affects about 90% of people at some point in their life despite that our costs um and interventions continue to rise despite the fact that our outcomes are not improving. So there's something missing here. Um And and I think that as I'll explain to you today, I think a lot of this is related to us losing um the forest for the trees if you will. And so what do I mean by that? Well, there's a lot of trees in the process of low back pain. So if you look at what happens to the spine, just with the accumulation of wear and tear changes over time as we age, it really affects every structure of the spine, interestingly, each of these structures can serve as pain generators um and do in some instances. So as we see the discs start to show changes with collapsing degeneration, The bone changes. Facet joint changes, ligament exchanges muscular changes. Um In addition to the behavioral biomechanical and biologic and inflammatory changes that we see, the question becomes, What's our target? What's our treatment target here? And how do we actually move the needle? And how more importantly do we define with the patient in front of us? What should be the target of our treatment when all of these different pathologies can actually be contributing to and driving the experience of low back pain um which in and of itself is essentially a syndrome rather than a diagnosis. Well, what is the most common diagnosis? Most common diagnosis in low back pain is just a generation starts in your thirties. We start to see wear and tear changes of the inter vertebral disc, we start to see loss of hydration of the nucleus pulp, Asus, thereby loss of its shock absorbing capacity. We start to see changes in the melody of the annual is fibrosis um with tears um which can lead to obviously herniation and other problems in the disk. And we see changes in the end plate sclerosis, which impacts the ability of the disk to have nutrition access to central portions of the disk and further the degenerative cascade the challenges. Just because this is the most commonly identifiable feature, does that mean it's the best thing for us to target? And we're going to talk a little bit about that and why I think that some of our targeted approaches related to to the inter vertebral disc have not move the needle in terms of care. And I say this in all. Um um uh admitting that we we spent our laboratory spent probably a decade trying to identify novel treatments for distant generation. And I'll tell you why? I think that that that path is still somewhat challenged. So let's talk a little bit about the biology of the disk. So what do we want to do if we're thinking about addressing disc degeneration from a biologic approach? Well, what we would need to do first and foremost is intervening prior to end stage an atomic disease. Once that disk is collapsed, it is unlikely that any biologic approach is going to restore the full size of that disk. And then we also need to uh to make sure we're addressing the underlying pathology, not just the pain which is the presenting symptom when patients come in. So what do we need to do? We need to shift this balance from one of cata bolic activity to one more favorable for synthetic activity to try to rebuild the structure of that disk and do so before the horse is out of the barn if you will. So, gene therapy has been an area that that many investigators have explored our our group included and really gene therapy, the idea behind gene therapy is to restore that advantageous anabolic cata bolic balance but with a sustained effect rather than a growth factor injection or an a single intervention. The idea with gene therapy is to try to do this in a way that because this degeneration is a chronic chronic degenerative process, you need a sustained effect. So a couple of different approaches that have been thought about for the disc, One is in vivo where you actually can inject a viral vector loaded with your gene of interest into the inter vertebral disc. Um And by doing so get the host cells to produce your anabolic or anti cattle bolic gene of interest. Another approach is the ex vivo approach where you can transducer cells in culture with your viral vector loaded with your gene of interest within the goal of restoring of reintroducing those cells back into the inter vertebral disc. Obviously there's risk benefits of both approaches. Um Both in terms of the efficiency of the effect as well as the number of procedures, numbers of interventions etcetera. But both have been examined. So here's some data from animal studies where we um looked at a gene therapy approach for inter vertebral disc generation and what we did here was um So these are just mid sagittal M. R. I. Images um in this case of a rabbit model of this generation where you can see the nice bright white inter vertebral discs on these T. Two weighted M. R. I. Images. Um And you can see a nice disc height. When we do an annular puncture to the inter vertebral disc. What we see is a slow sustained degenerative cascade that we've characterized biomechanically and histological E. And from a gene expression perspective to very much mimic the course of disc degeneration in humans. So you can see that loss of that T. Two weighted signal loss of disc height in these punctured disks. If in that same disk generation model we inject the disks with um add no associated virus loaded with bmp. To an anabolic molecule that helps to facilitate cartilage and apple is um what we see is relative preservation of that T. Two weighted disk signal as well as the inevitable disc height. Similarly, if we take our ad no associated virus and loaded with temp. One which is tissue inhibitor of metallic proteus is an anti cata bolic factor. We also see relative preservation of this kite as well as T. Two weighted signal indicating preservation of the proteins like Yoon. Um the main molecule that imbibes water and inter vertebral disc. So this looks great. Um We look at the biomechanics of those disks because not only do they have to look good on imaging but they have to function well. And if we just look at creep curves which is what you see here um After applying axle load to that disk, what you what I hope you can appreciate here is that there are some differences whether we load the disks with an anabolic molecule or an anti cata bolic molecule and in fact those loaded with BMP two are stiffer than those loaded with temp one. Nevertheless we do see some restoration of disk biomechanics as well. Um which suggests this is a promising intervention. We also can look at what's happening from a turnover perspective by looking at released fragments of collagen to So this is a marker ctx two which is the cto peptide of collagen to that is released during remodeling and we can pick this up in the periphery in the in the bloodstream and so we can measure this over time even without sacrificing the animals and what we can see is normally over time and animals that undergo a puncture, you see continued release of college and two fragments into the periphery, which is what you can see in the red lines there. Um But both interventions, both with the antique alley bolic molecule and anabolic molecules show blunting Of that release of collagen to um demonstrating that from a biological perspective, we're also having a beneficial effect. So you know, these things look great. So this these work were published back in 2012, almost 10 years ago now. And so the question is why has this not been translated um broadly into human trials? And the punchline is safety related issues. And so unlike other areas where gene therapy has made some progress in the inter vertebral disc space um there are significant safety concerns related with gene therapy approaches. And what we've seen in other examples is that even even if we think we have a perfect placement of the gene therapy vector, if you have super therapeutic doses or you have in uh placement of the injection in the in the intramural space in particular, you can see fibrosis inflammation um which can lead to paralysis or para seizures and the animals and and significant morbidity. So it's not without risk. And um a significant concern in a chronic process like this degeneration. So what about stem cells? Stem cells have had a lot of attention placed on them. Why not think about stem cells to restore the inter vertebral disc. Um And we did some work in this space as well. Looking using that same animal model of disc degeneration and this. Uh these cells were these were in uh stem cells isolated from their umbilical derived stem cells. And so they were placed in a hydrogen carrier or just in buffer. And what you can see here is um some real again relative preservation when in the disks that were injected with the umbilical drive stem cells in the hydrogel carrier um and again promising from an animal model perspective, why has this not been translated? Well, very similarly, there's a lot of safety concerns if you have um stem cell leakage stem cells differentiate depending on their environment of course. And while in the inevitable disk they differentiate nicely into contra acidic like cells which can be advantageous for the disk when they leak out um into the epidural space for example, they can differentiate into bone. And bone formation is not a good thing in and around the neural foramen or the central canal and can be quite problematic. And and we've demonstrated that um in the animal studies as well. So in thinking about improving that, can we do something to keep the cells in place. And this was a series of experiments that were done by a small army of orthopedic residents who worked in the lab with us for a couple of years and looked at a novel delivery mechanism by which we could use a fiber optic um like to go in and prelim arise hydrogel carrier in situ within the inter vertebral disc in hopes of keeping those stem cells in place. And again looked very promising. Um You could see again relative preservation of disc height, dis signal area, even biomechanics of that disk. Nevertheless, despite this we did see inflammatory changes um and differentiation in the bone with significant osteo fight formation in that epidural space which is obviously not advantageous. So um despite best efforts, there's still a lot of um safety concerns related to these biologics. And so when we rethink you know, we think about that, Why are these things not making it to prime time? This is largely the reason. Um And and kind of the take home here is you know, there's been a lot of thought about natural um um and and autologous approaches in the inter vertebral disc, the inter vertebral disc is a very unique environment. It is relatively immune protected. It is relatively a vascular and so it's a very unique space. Um And so just because we're thinking about using autologous stem cells or others doesn't necessarily mean it's safe. This was a study that we did years ago that really hit that home to me um looking at glucose to me. And so this came up just really in in clinical work patients saying ah you know um should we be taking, should we think about glucosamine for back pain and and the cartilage degeneration associated with back pain. So this was a study done by lou Dean Jacobs who was orthopedics resident that spent some time with us and what she did was literally fed the animals that underwent the inner innovative real stab, this degeneration model oral glucosamine Um and followed them over 20 week period. And what she saw was quite surprising and and um we actually didn't, we thought maybe the lab numbers were flip flopped or something. And that is the animals that were um fed glucosamine actually had worse degeneration um than those that were not. So um you know assuming that just because it's it's a natural or a supplement or um you know something autologous that it would not cause harm is is clearly not the case in the inter vertebral disc giving it its unique environment. Um And and um and biology. So so what about of course the disc does not live in isolation. It lives under constant mechanical load. And so how do we have to think about the interplay between the biomechanics of the spine and the biology of the spine. And this is a very important consideration when we're thinking about trying to alter the biology. So I like to think about because I'm a fizzy a trist. I like to think about mechanic biology as perhaps the safest biologic and if we could harness the beneficial effects of mechanical loading to actually facilitate a beneficial biologic response that could be very advantageous if we were able to understand that. So if you think about how the disk experiences mechanical load it transmits that load through the extra cellular matrix and cell adhesion molecules um to result in signaling cascade changes that we can then measure in the laboratory. And so we've done this in a number of different ways. We've looked at tensile stress. We've looked at compressive loading on inter vertebral disc cells to see what is happening um from a biologic mechanisms standpoint in response to these loads that would normally be experienced by the inter vertebral disc. So here's some examples of those data just to help you understand that the the effects occur at the level of gene expression. So M. RNA expression. So these are data that show relative gene expression in universal disk cells isolated from the nucleus pulp closest most center portion of the disk and what we can see is depending on the magnitude of compressive load. You can see anti inflammatory effects at the gene expression level. So this first panel here shows you decrease in molecules like china's or MMP three or cox two in response to mechanical loading. Um And in fact suggesting that some modest levels of mechanical loading can actually be anti inflammatory. What's interesting is the time effect. So if you take that same load that's anti inflammatory but you extend it out for a prolonged period of time in this case 24 hours. Which is what's shown in the second panel here you now start to see things being pro inflammatory. So you know maybe at the most basic level. This helps us to understand why sustained compressive load sitting at your desk all day for example um could have a negative effect on the inter vertebral disc itself on the on the tissue itself. Similarly we looked at tensile strain and we looked at this in the annual ist. Fibrosis cells that primarily express primarily experienced tensile loading and what we saw was very similar that under conditions of inflammation which is what's shown in the white bar. If we applied tensile strain to the inter vertebral disc cells we saw a decrease in the amount of gene expression of inflammatory mediators or matrix Metella produce is responsible for metabolism um of the disk. And so this again demonstrated some anti inflammatory effect of controlled levels of mechanical stretch. But again, duration seemed to be the most important variable. You could fatigue that effect. So if you take that beneficial tensile load but you do it for a sustained period of time you start to see a reversal of that of that anti inflammatory effect. What's even more complicated is if we look at cells from the degenerated disc. So if we look at fanatically degenerative cells and how they experience mechanical load it's quite different. So unlike normal cells, Fiona typically normal cells degenerative cells have an enhanced response to inflammatory stimuli, loss of that beneficial response to anti inflammatory mechanical load and enhancement of their response to traumatic levels of stress. Excuse me. So, what this really tells us is we need to be thinking about how we can hi trait if you will, different types of mechanical loading based on where that person is biologically or structurally with their disk, which gets to the idea of needing to personalize this. And we're going to talk a bit about how we may be able to start to think about personalizing this based on people's individual biology and biomechanics. So could we combine these, Could we think about combining regenerative approaches that I showed you previously with mechanical approaches. And I like to think the idea of regenerative rehabilitation, which is a term that has really garnered a lot of interest. Um and Fabio Ambrosio here at University of Pittsburgh, I'm so sorry, has done a lot of work in this space. This idea that that that synergy between the biology and the biomechanics can actually have an even further enhanced mechanical effect so that there are favorable levels of loading, unfavorable levels of loading and combining those with biologic therapies may give us the best effect. But in order to be able to titrate that in order to be able to understand in the human, not in the animal model or in the cell culture disk dish, how we can identify that and titrate, that is one of the key consideration. So how would we do that? Well we need some type of peripheral marker. We can't biopsy the disk. We don't yet have sophisticated enough imaging capabilities to do molecular imaging on a regular basis in the disk. So we need some sort of peripheral marker to do this to be able to translate this. Um And this is some work that was done by a very talented graduate student and postdoc in our lab who created a bioreactor for a spine segment, bone, disc, bone could keep it bio active and then apply different levels of mechanical loading, actual compression, repetitive flexion, extension, torsion and measure what happens to the biology, importantly, what he found is in addition to what happens to the tissue biology. He could also find released biomarkers into the surrounding media and in this case that shows you an example of where he measured a marker called CS 8 46 which is a marker, a pretty black young turnover. Um That could that change in response to loading of that inter vertebral disc and and bone that that functional spine unit if you will. And so this took us in the direction of thinking, well perhaps we can consider finding molecular biomarkers that could help us in in targeting treatments. Um Not only is it a needle in a haystack, it's a bit more like a needle in a needle stack. It's really difficult to sift these things out and identify where the biomarkers are coming from in order to allow us to use these as potential targets. Nevertheless um it is showing some signal and I'll show you some evidence of that. So this is a study that we did um looking at um older adults with axl low back pain and peripheral circulating biomarkers. So we drew blood from these patients before and after um physical activity where they were doing some walking, some repetitive um actual torsion, some sit to stand activities. And we looked at the change in those molecular circulating biomarkers to see. Was there any difference that seem to associate with their experience? And we found many biomarkers that did. I'll give you an example of a few that are particularly interesting one is Rantisi's which is a systemic marker of inflammation and what we found and these pie charts to show you the r squared value. So the relative predictive contribution to the observation. So grantees had a significant association with patients gait speed. It also had a significant association with their performance on a short performance, physical battery. Um And so this was interesting to us to see that a marker of inflammation was changing in response to physical activity and actually correlated with their performance in these functional tasks. We also looked at europe peptide yy, which was which is a marker um that a circulating neurotransmitter that has been associated with resilience in patients with chronic pain and what we saw interestingly enough was that the change in N. P. Y. Seemed to be associated with these patients response uh emotional response if you will. Things like the effective score on the McGill, which asked questions like is your pain punishing or cruel fear avoidance behavior. So their fear of movement as well as depression scores. And so you know, there was some differentiation between these two biomarkers as to these patients um experiences of pain and what seemed to be primary drivers for them in terms of their overall pain syndrome. We also noticed that these two different biomarkers again associating with different aspects of their pain at baseline. So the N. P. Y. A baseline was associated with again the McGill effective the depression and the fear avoidance. While as rantisi's uh seem to associate more with functional capacity and while you might say well these are really modest um and unimpressive r squared values which I would agree with. Um This is a small cohort of subjects. Um And I would compare I when we compare that to the association with what is quote unquote, our gold standard for diagnosis and low back pain, which is the M. R. I. Where we have a little more a little more enthusiasm about those modest associations because the association in these patients with their M. R. I. Scores, whether we look at that based on a computerized index that is calculated based on T. Two weighted signal and the size of the disk or whether we look at that based on um the radiographic read the radiologists read um of those in a scoring um system. We see absolutely no correlation with their pain, this shows their pain. But we also look for correlations of their M. R. I. Findings with their function and and other markers and so essentially um no correlations. So it suggests that there's something to look at there and we and I'll tell you a little more about what we're doing now to further hash that out. But of course when? Yeah. Question since he invited us to ask questions in the middle. I will ask um now. But so that those biomarkers that you were is this in chronic, it's very interesting that you have a stronger association with the mental health markers than other things. So I'm wondering if this in chronic patients and in the countries of when are you as making the essay of this molecular biomarkers? Are they doing an exercise on and then you with your blood or are they just a baseline? And it's such a big distance between, I don't know. Yes, it's a great question. So these are all This cohort was all older adults, 60 in order with chronic axial back pain. Um And what we did was we just brought them into the lab. So it wasn't a certain period of time after the onset of their symptoms. They had to be defined as chronic. Um but we measured, we assayed them when they first arrived to the lab and then we did a series of physical activity maneuvers and assayed them after that. And we were looking at that change and then we're also looking at the association between some of these metrics and their baseline value. Now your question is important because baseline is going to be different than everyone. Um it's going to depend on their diet, their medications, their medical comorbidities, all of these things. And we do see quite a bit of variability in just those baseline levels that are completely unrelated, likely on completely unrelated to their chronic back pain. And so sometimes looking at this um, delta or change in response to physical activity can be more helpful if you will than looking at the baseline. But if you think about it from a pragmatic standpoint, what we really need is something to give us a trigger on what we should be doing. Um So we need something that's gonna give us information about where that patient is at that moment, um which is why we're starting to become more interested in genotyping and I'll talk a bit about that. Um But yes, the the, which is why I alluded to it as a, as a needle in a needle stack because baseline levels are affected by so many different things. Um and with these circulating markers, most definitely. Yeah, so to that point, what here, here's an example of where we looked at this um at with at the genotype level. So to to in hopes of trying to address the issue of when to essay how to assess a um other um contributors. So because we were interested in neural peptide Why. Um and um for lots of reasons. One um it's actually surprisingly produced by the inter vertebral disc, even though it's a neurotransmitter that was somewhat unexpected to us. We did look at this and the disk itself can be a source of neuro peptide y, which is in and of itself interesting. But we were interested in this marker and wondered if we could see some genetic variations. So we looked at um known single nucleotide polymorphism in the N. P. Y. Gene. Um and we looked at this in a population of patients with spinal stenosis and symptomatic spinal stenosis. Um and we subdivided these patients as responders or non responders to the intervention. So this was a this was a piggy back to a randomized trial um that was looking at different treatments for spinal stenosis. And what we saw was there was a difference based on the genotype in the responders and non responders and who carried the snip and who did not. What was even more interesting is when we looked at the three different intervention groups within the clinical trial. So there were three interventions they were randomized to either usual medical care which which essentially included medications and epidural steroid injections and um or group exercise? Just a group exercise program or standardized physical therapy. And what was interesting is in the patients that underwent the usual medical care, there was no difference in their response rate based on genotype, but in both the group exercise and the physical therapy, there was a difference between responders and non responders based on genotype. Um, so, you know, again, this is just a potential signal um, in many, many signals that could be explaining this, but one that has garnered a lot of interest. So what can we take this one step further? It's one thing to to look at, whoops. My slides are stuck. There we go. Um It's one thing to look at this in a cohort, but it's but it's quite another to think about it, potentially guiding treatment. So we did a study looking and we asked, well, let's let's think about something episodic because it's hard to say respond or non responder in chronic low back pain because it's this chronic disease that you have to look over. Um a long period of time and a lot of patients, but something episodic like response to an epidural steroid injection. Could we look at this in that cohort and see if there are biomarkers that might help us figure out who should and should not undergo an injection. We did this in actual low back pain and you might say, well, why are you doing epidurals in actual low back pain anyhow. Um The the evidence isn't that great for it. But what we see over and over in the literature and over and over in clinical practice is there are a lot of injections that are going on for actual low back pain. There seemed to be a cohort that do respond to it. We just don't know who they are a priori, which means when you look at this over, over the entire cohort, you see a pretty poor treatment effect. So is there a cohort within there that actually has the potential to be treated? And can we identify them a priority to then? Only inject that group. And so we we recruited patients that were adults that had low back pain worse than any other pain in their body to try to limit that contribution from other areas. And we're having a spinal injection performed as part of their routine care. We purposely did not influence whether or not they would be receiving an injection. We approached them for consent only after they consented to undergo Um an injection. So we had a relatively small cohort 48 subjects that underwent a lumbar epidural lumbar epidural steroid injection. And we define responders as 50% or greater improvement in their numeric pain rating scale. So a pretty high bar. Usually we think about clinically significant improvement is much smaller than that. But we kept a pretty high bar. Um and we ended up with 17 responders and 31 non responders um based on those criteria. And importantly we also looked at those patients disability index and the those that we classified as responders also had a clinically meaningful change in their ancestry disability index as well. So what we saw is that there were genetic biomarkers that again were different. That N. P. Y. Marker that we were using previously was different in responders and non responders as was the vasopressin receptor which has been looked at in chronic pain populations as well. But even more impressive than that was the difference in cata cola mean method transferees um obviously involved in the energetic system and what we saw was pretty specific that um that non responders or those carrying the alien will um in the snip or all non responders. So it was it was a fairly dramatic even though this is a relatively small cohort to hang your hat on, it was um suggestive that maybe you know this might be a direction that we could go to use this as a predictive tool. So how can we put this all together? And so one of the things we're thinking about now and this is really just to get you thinking about how could we use these. So what we very much don't think is that there's going to be a magic biomarker. There's gonna be one marker that is going to be suggestive or predictive of patient's response to treatment, particularly in something as complex as low back pain. But can we start to combine large data sets in ways that give us better predictive capabilities. And I'll give you just a bit of a teaser on that. So this was that uh that same cohort of patients that underwent epidurals. And if we look at just their clinical data, um we looked at their profile based on their medical comorbidities, medications, etcetera. And created these phenotype sense of patients who were responders and non responders just based on their clinical data. And that's what's shown here. The red being non responders, the blue being responders. We did the same thing with the genetic data, the single nucleotide polymorphism data. And again, we saw some that were clearly defined as responders, some that were clearly defined as non responders, but a lot of overlap of those peaks. Right? So a lot of uncertainty still in that analysis. But if we took the clinical data and the single nuclear single nucleotide polymorphism data and we put them together and did that same fanatic typical analysis, we saw much better separation of those cohorts, so much better ability to predict who is going to be a responder and a non responder. And this is really the direction we would like to take this next. Um and I'll tell you a bit about how we're doing that now, but just um Suffice it to say that, you know, because this is a syndrome and because it is multifactorial, putting all of those factors into the analysis is likely going to lead us down the road of a better predictive capability. But one thing we can't ignore is the behavioral component of this, which is so important and so substantial in chronic low back pain. So I'll touch just briefly on these items because this is something that we need to incorporate into our clinical decision making. So there are a lot of barriers that patients have to their recovery. So outside of the structure of the spine, outside of the inflammation they're experiencing, there are a lot of things that are impacting limiting um and complicating patient's recovery, things like their mobility itself, pain that prevents them from reactivating and becoming active. Fear of movement, anxiety and depression, which are so, so common in this patient population and create can create a vicious cycle and lots of coma, coma, coma, sorry, medical comorbidities that commonly coexist. So what we've done um clinically is we've brought like many others have we brought these practitioners together in a single clinics in order to try to um tease out these different components and remove some of the barriers that patients have to recovery. And this just gives you a sense of how we have thought about that here in Pittsburgh are what we could, what we have have titled, our muscular skeletal health program where patients are managed by a nurse care coordinator. We collect patient reported outcomes. The most of these patients are quarterbacked by a fizz ia trist. They're always seen by a physical therapist and their initial assessment and then we connect them with some of these other resources. We have a dietician, we have a pain psychologist, rheumatology, sleep medicine, anesthesiology. All working together to try to coordinate and and um provide the comprehensive care for these patients. And so what what has this looked like from an outcomes perspective despite the fact that we get really the most complex um patient population. So the vast majority of our patients are classified as medium high risk based on start back tool which is a common assessment for risk for response. Um they've done very well and so the biggest, probably the biggest win is almost 70% of them are actually converting into get, going to physical therapy and reactivating and becoming more active. And a quarter to a third of them have seen significant improvements in their functional scores. Their disability scores with relatively low utilization. So this is why these programs work because you're actually removing some of the the um unnecessary utilization and replacing it with some utilization that's actually going to move the needle on on removing barriers for long term recovery. Um this works nicely in our environment because we are an integrated delivery and finance system which means we own our own insurance company. So about 40% of the patients that we serve are self insured, which means the money we dump into these programs is realized as a benefit to the insurance side of the house um and can be reinvested in the program. And so that helps to keep things um from uh keep things to be sustainable. Um And so how can we then think about further um individualizing the care that we provide for these patients? Obviously not everyone needs a dietitian. Obviously not everyone needs a spinal injection. Not everyone needs a psychologist. How do we um Hit that sweet spot to make sure we're giving the right treatment for the right patient. And so we really need to refine our decision making capabilities. This is some, this gives you an example of one of the dashboards that we've created to do just that. And this is again, this example is looking at epidural steroid injection. Again we kind of use that as our as one of our ah test cases because it's episodic and it's easy to measure. But here's an example of where we use just the clinical data To create 10 different phenotype sis of patients who are going to respond or not respond. And then we can use this. This is actually a visual on the dashboard that the clinician can sit down and look at this with the patient and say, okay here's your individual likelihood based on our clinical prediction um of your likelihood to respond to an injection which is what's shown in the green or were actually worse than from an injection and do a shared decision making with that patient. Um This is a page. This is an example of a patient who has a modest degree a moderate degree of disability but really no comorbidities. If instead we look at a patient who is taking a muscle relaxant and opioids and nonsteroidal and antidepressant, they have fibromyalgia and I. B. S. You can see how dramatically that changes their likelihood to respond to that injection. And again this can be used in a shared decision making way. And then finally if you take a patient just like that and you layer on anxiety and depression you then see a much more a much higher odds of not responding to that injection. And so we've used this and we've done some quality improvement initiatives looking at what are the outcomes. Um and our our patients choosing um to still move forward with the injection interestingly enough some do um and we followed that over time. But this has been a very useful tool for our clinicians um in in trying to sort this out and and individualize the those treatment decisions. So in the end what we need to do is the right treatment for the right patient at the right time and I'll leave you with this idea of how I think we can get there. And it really relates to big data. So what I've shown you are small signals in small data sets and in order to really get this to the next level. We need lots of data, comprehensive data and large data sets to do this. And um so here's an example of that where we were able to use big data. So we have Um within our system access to over 500,000 adults with low back pain. And so we can mine those data sets and start to develop more robust phenotype in capabilities. And this is an example where we looked at that cohort of subjects. And we were particularly interested in the question of the patient's likelihood to continue taking an opioid for their back pain versus their likelihood of discontinuing and opioid for their back pain as a surrogate marker for things going in the right direction if you will. And what we were able to create is all these distinct phenotype. And so all these different colors represent a different phenotype for a patient and their own likelihood to either remain on that opioid or come off of that opioid. And again we can map an individual patient onto this based on their individual characteristics. But what we can also do is mind the phenotype itself. And so look for the inner connectivity of the different variables within that phenotype to give us information about what else we should be looking for. If we just keep looking for treating the anxiety and the depression or we keep looking for treating the inflammation or we keep going after the same variables we're missing. We're likely missing some important pieces. And so this is an example of how we can do that where we take one of those phenotype and we say what are the characteristics that make up that phenotype and what are the associations within each variable? And we see some interesting nodes that we can then start to develop some hypotheses around. So in this case we saw in one of our cohorts that are very, you know, interconnected node was whether or not they saw their primary care physician. So what do we need to do with intervention at the primary care level to start to have an impact on these patients trajectories. And so it helps us to look at some things that we might not be thinking about doing. So that's exactly what we're doing with our um our newly funded center. So we were fortunate enough to receive a very significant grant from the NIH to fund a Mechanistic Research Center and low back pain. This is part of the backpack initiative which falls under the NIH heal initiative. Um and what we are doing as part of um as as uh this funded center is we are recruiting 1000 subjects with chronic low back pain and we are doing deep phenotype ng. So pretty much everything I showed you today, we're going to look at in all these patients and then some, we're going to bio sample what we are bio sampling these patients. We're collecting saliva, serum and spinal tissues for anyone that moves forward into going into surgery. We're doing proteomics, genomics, transcript atomics. We're doing deep phenotype ng of them from a biomechanical perspective where we're looking at them in the lab based on biplane or fleur ah graffiti. Where we can look at segmental biomechanics. We're doing sending them home with wearable motion sensors for their spine as well as accelerometers to look at general activity and look what's happening in the field. We're doing assessments of them uh from a physical function perspective in the laboratory and then we're collecting a very significant number of patient reported outcomes, Social determinants of health and lifestyle factors that we think that if we start to look at this and then combine all of those things. I showed you that brief example where we just combine clinical metrics and and single nucleotide polymorphisms and got better predictive capabilities. Our hope is from a machine learning approach. If we take all of these literally tens to hundreds of thousands of data points and start to put them together to define better phenotype. So that that will lead us to the next step of not only directing care more appropriately but also figuring out how do we develop novel approaches to care in this patient population. So this is a really exciting um project and maybe if Pablo invites me back in a few years I can I can tell you what we've learned. Um hopefully we've well you will move the needle on this. But point being it takes a it definitely takes a village to do this. And so we have people from bioinformatics, from physical therapy, from P. M. And R. From psychology, chiropractic engineers, molecular biologists all working together to try to put everything about this syndrome together and hopefully better define where we're going. So it's very daunting. I am always painfully glass half full and so I think that we will be able to eventually um figure this out in a more um targeted way so we can improve treatments. Um but you know with that it just it takes interdisciplinary teams. I'm speaking to the choir here, you guys know how to do that um very well already, but it also makes for a very fun work environment to bring people from very distinct backgrounds together around this common problem. And I'll just end by mentioning one of the things that we're doing um is um starting to put clinicians and researchers together. So this is an exciting new initiative and that construction site pictures is now a little data. We actually have four floors built on our new vision and rehabilitation tower which is going to pair researchers and clinicians Located in the same space bouncing off one another. Um so that we can try to continue to move things forward more quickly. And it's interesting because we are combining ophthalmology and rehabilitation. You might think, well that seems a bit distinct, but if you think about visual impairments, cognitive impairments and mobility impairments and how people negotiate their environment, there's so much synergy and overlap that we're finding um that it's been a very, very exciting project. So I'll end there by just thanking you for the time and your attention today and certainly more than excited about hearing your thoughts or suggestions or questions if we have a few minutes wen thank you so much for this fantastic talk. I really appreciate the a chronological perspective of where we are today on on this story. So this is great. Uh I let people ask questions um maybe while they're commuting, I'm going to ask maybe the first but the second now. But so this Hill project, the backpack project that this sounds awesome. Um It is in the context of standard of care medicine. So patients are going through whatever the physicians or other providers are suggesting to the patients to do how you take that into consideration. This also, you know, put this as a variable in machine learning and hopefully something comes out. Yeah, it's a great question. So this is an observational cohort. We are not in any way directing treatment, We are not in any way influencing treatment and what we are hoping to to um Acquire is patients who go through a lot of different treatments, right? And that is what we're seeing so far. So we're about 220 subjects into our 1000 cohort. And what we're seeing is with the exception of we're actually seeing very few surgical patients, which is interesting and we'll have to figure that out. But lots of different treatments. And we have a very um detailed um questionnaire that and we pulled this from the electronic medical record to that that outlines their treatments, the timing of their treatments um when those treatments are in comparison to our assessments as well. And yeah, my hope is that, you know, dumping this into that machine learning approach will help us say, okay in the cohort that underwent physical therapy in the cohort that underwent an injection in the cohort that went, you know had medication management. How do we sort that out? But we are part of a larger consortium. So there are three mechanistic research centers, one at UCSF and one at michigan. And we're also going to all dump our data in together. Um And and look at that what we are current. The consortium is currently planning a randomized clinical trial That will start in 2022. That will also be doing all this deep phenotype ng but now we're going to direct the treatments. And so we're gonna have a forearm study. It's going to be a smart design where there will be some crossover. Um and and hope to be able to help um tease some of this out that way as well. But this group at this stage because we don't know the phenotype yet is purely an observational cohort. Yes. Great thanks. Eric Yeah. Gwen yeah, great, great talk. Um I was really curious. I was interested that um you you mentioned this sort of tool that lives in the r system that you know sort of has all these different data points that you know sort of give this odds ratio about whether the patient's going to um you know, be successful for treatment, things like that. Could you just talk a little bit about I mean it seems like you guys are collecting a lot of data. You mentioned, you know, depression, you know, information this and that and and how are you guys, how's that? How's that? How are you guys you know, what spends any you know, advice suggestions in terms of in terms of systematic collection of that so that you guys can kind of do that in in clinical carry you sort of you mentioned, you know, this is maybe a quality improvement project. But can you speak to that a little bit because I think sometimes, you know, having you know, providers collecting a lot of data can be very challenging. And so I just was wondering how you guys have been successful with that. Yeah, it's a great question and that example that I showed was very little pr oh based data it was more pulling data out of the you know diagnoses, medication prescriptions, those sorts of things rather than promos. Because you're right the the um kind of systematic collection of patient reported outcomes is challenging in its consistency and it's challenging in its implementation on a broad scale. So our clinical analytics while we have the promos and we put them into we were we use Epic here and so we have them in discrete fields and Epic when we're collecting them of course they're not being collected all the time. We're moving toward promise measures that we will have much more robust collection because they'll be collected in the primary care setting and the specialties office, all of those sorts of things. Not these kind of one off episodic which is what we've had to deal with in the past. Um but we have focused those explorations more on the the standard fields in the E. H. R. Than we have on the P. R. O. S. Um With the exception of for that example we used us West Street disability index which in our system had been collected on on subjects whether or not they were in P. M. And R. S. Office or PTS office or surgeon's office etcetera. So we were able to use that. Um but we've tried to mine the data that um is standard lee access accessible and not rely as much on those things that we rely on the patient to collect for the for the reasons that you outline. Um And because we have big numbers were able to do some of that um Now with our cohort ng with our mechanistic our research projects were able to we collect like 45 minutes, you know, of questionnaires where patients fill out almost, you know, it takes almost an hour to pull out their questions. We can't do that in the clinic. Obviously we need to we need to um pick and choose and so that's why we're going to focus more on some promised measures that hopefully will be a little more streamlined. So All right, thanks so much. May I ask a question please, Jeff? Thanks. I'm this is not going to be a so much general interest but I wonder about having so many different data points going in. And is there something that you do to reduce the likelihood that you're gonna find type one type two errors? Just on account of there being so much data? Yeah, it's a great question. So, you know, are we gonna send ourselves on a wild goose chase for the next decade because something pops up by chance that we decided to to chase down. It's a great question. Um and one I wrestle with with all of our bio informatics colleagues on our machine learning folks and Ai folks. Um and that's why we're going to do both. So we're going to take multiple approaches from a machine learning perspective but we're also going to take standard statistical approaches in testing some of those hypotheses within our cohort. And what we were actually just talking about earlier today was just on with our NIH program officer. And we're talking about using each other's large data sets within backpack to validate some of those. So we'll develop these and what I think is going to come out of our data set. Our hypotheses that need to be validated on other existing data sets and that's what we've exactly discussed doing is swapping big data sets. So we'll have this massive database of you know, a couple 100,000 data points will develop our clinical predictive ideas are hypotheses essentially. And then hopefully we'll swap with UCSF and say let's test them on this other cohort of subjects and see if there if we're able to validate it. But I think so my short answer is we're gonna need to take multiple approaches so that we don't go down a rabbit hole and then we're clearly gonna need to validate this on other datasets and then obviously down the road will need to definitively test this prospective lee um to see if it's actually meaningful in patients. Um or or have we just you know, found the red herring. So it's it's a bit more beautiful answer. Thank you. Gwen. Hi Barbara. Hi uh this isn't scientific. I just wanted to say, I noted a nice touch where on many of the references you included? A photo of the young author. Yeah, Thank you. So, we we really, really, it's it's one of for me, the most rewarding part of all this work is seeing some of our scientists um, in training uh, start to accomplish this. So, thanks for noticing that. Yeah, we like great. Well, I think we are a few minutes after the hour when. Thank you so much for joining us today and really sharing this. This is really nice work. So hopefully people get inspired and maybe you want to connect and follow up with you and so you have lots of data to mind. So yeah. Yeah, Absolutely. Great. Thank you again. And thank you. Goodbye Jody. Like this. Created by