About the webinar
This session brings together leading experts from the US, UK, and Netherlands to explore where the field is heading. All four talks point toward the same underlying shift in PAH care — from reactive, late-stage diagnosis and generalized treatment toward earlier detection, continuous monitoring, and individualized therapy, powered by AI and better data. For patients, that translates into the possibility of catching the disease sooner and getting treatment tailored to them specifically, rather than a standard protocol.
Download the slides
Integrating real-world and digital data for clinical decision-making
New Imaging Approaches for Screening and Diagnosis of PH
Transcript
Nils Fisher:
Hello everyone. Thank you for joining us for today’s session where AI and Science Converge. My name is Nils Fisher. I am the Associate Director for Patient Engagement at the Partners for Patients Foundation, and I will be moderating today’s webinar on behalf of the Alliance for Pulmonary Hypertension. Today, we will be exploring how new approaches to data, biomarkers, and imaging can inform diagnosis and more personalised care. And we have four exciting presentations and we’ll finish with a live Q&A session. And we’ll close in one hour from now. Please go ahead and send your questions in the chat at any time. We’ll take as many as we can after the presentations, but please keep in mind that we’d like it if you could keep these questions general and avoid including personal medical details as our speakers cannot advise on individual cases or treatment decisions. A warm thank you to all our speakers for sharing their time and expertise, and let us begin with our first speaker.
Today, our first speaker is Professor Joanna R. Preston, director in the Division of Pulmonary and Critical Care at Lahey Clinic in Burlington, Massachusetts. She will discuss integrating real world and digital data for clinical decision making. Professor Preston, welcome and over to you.
Joanna R. Preston:
Thank you so much. It’s such a pleasure to join this esteemed group and lovely to speak with the audience. So my task is talking and having an overview of integrating real world and digital data for clinical decision making. And I have to say it has been a daunting task for me as an, I should say seasoned clinical physician in taking care of patients and being used to doing what we usually do in clinical setting in the care of PAH patients. These are my disclosures. So where are we today and how can we accelerate the diagnosis and treatment for our PAH patients? We start off with non-specific symptoms. We do routine testing for other reasons. The routine testing is not read for PH, so physicians do not have that in the background. That leads to late referral to late diagnosis and treatment. So it takes still in this day and age, roughly two years from symptom onset to expert referral centre for our PAH patients.
So how can we detect earlier? How can we confirm faster? And how can we manage better our patients by integrating artificial intelligence data and software systems to improve our care delivery? So let me talk a little bit about the diagnosis of PAH and the AI emergence in the PAH diagnosis, and I will highlight three aspects. The first one is the University of Maryland EHR analysis tool that recently came and was published, and it was a extraction framework that was tested across over 10,000 patients against the right heart catheterization. And the AI algorithm was spot on in diagnosing pulmonary hypertension compared to the traditional workup. What was interesting was it discovered a small group of patients who did not get the pH diagnosis because they did not have all the data for the diagnosis clinically, but it derived the data from the other parameters to make the pH diagnosis.
So it predicted the main diagnostic parameters and uncovered patients with PH who were not diagnosed in the clinical chart. The second study that I thought was very interesting came from the Mayo Clinic in 2024, and it used an AI-driven called convolutional network, neural network that was developed to look at ECGs. And patients were classified based on this algorithm as pH likely versus pH unlikely, which were the controls. And they were validated for the diagnosis based on the right heart cath on echo and the PHEDA, so this algorithm diagnosed PH patients correctly at the time of diagnosis, but more importantly, it identified patients six to 18 months prior to the corrected diagnosis, showing that there is a potential for an AI-driven software looking at ECGs to diagnose pulmonary hypertension earlier. And then lastly, there is a last FDA clearance of the Tempus ECG PH software, which is actually the second ECG PH software approved by the FDA.
And it analyses the ECG in a very complex fashion, looking at the waveforms and parameters that at the first glance the clinician may miss and it accurately identifies patients with a mean PA pressure of over 20. Now, it doesn’t separate group one versus other group of PAH patients. However, it can alert the physician that they have a patient with one or the other form of PAH, which I thought it was very interesting for the early diagnosis. Now, if you look at the PH care, once you have a diagnosis of PAH, we have a very fragmented care that we do not evaluate patients between clinic visits, and that leaves to a reactive management where the interventions occur after each clinic visit when patients may have symptoms months before that. So that means that we have data blind spots of possible progression of the disease before we see patient in clinic.
And also we have to take into account the subjective blindness that comes from the patient reporting in the clinic where the patient says, “Oh, I think I feel the same.” But if you really look at the data before the visit, you can spot small subclinical signs that the disease is getting worse.
Excuse me one second. Okay. So how can we follow our patients in between clinics? We can use continuous biosensors by different digital gadgets to measure the physical activity, to measure autonomic stress like heart rate variability, to measure oxygen saturation, which is mostly very important also at night, and fluid tracking with smart scales to keep track of the patient weight. This is one of the wearable functional assessment tools, My Heart Counts, which was a very large endeavour driven by the Stanford University, which looked at different patients with different disorders and controls. And it validated lately in idiopathic PAH, it validated the continuous passive smartphone and wearable metrics, and it mapped longitudinal functional decline over months, and it clearly differentiated pH profiles from age-match healthy cohorts. If you look on the right-hand side screen, you can see one example of the steps per minute that healthy controls in the green line perform over a week versus yellow blue idiopathic PAH patients and yellow disease control patients.
What was interesting for me also to see how the healthy controls drop their activity over the weekend, but it also showed seasonal variability and a lot of other parameters, and it correlated those six-minute walk distance in clinic with at home as well as walk distances with the NT-proBNP so we can measure the activity of our patients at home and find signs that their disease starts not being under control before they complain. The other clinical trial that I thought was interesting was from the University of Rochester where patients were measured with a wearable functional device at home and then seen in clinic. And you can see in the graphs, the six-minute walk distance at home correlated fairly well with the one in clinic, but also identified a new AI learned parameter, which is called cardiac effort, which is total heartbeats expended divided by distance travelled and seem to have a good correlation between at home and in clinic measures.
So more to come with smart devices to measure at home. But we know that patients with pulmonary hypertension and all the measures that we do at home and in the risk scores miss one very important component, and I’m sure Harmon will talk about it. It’s the right ventricular performance that lags, and we don’t have it. It’s somewhat difficult to standardise and measure in the echo lab in a clinical setting because it’s laborious and it has a lot of interreader and variability. So there has been development of an AI automated software that reads the right ventricular size and function that seems to be very reproducible and scalable, and maybe we can use it in the future in our clinic. I’m highlighting three very small single centre trials and reports that showed that the right ventricular fractional area change, right atrial area do independently predict mortality with this learned algorithm.
Right ventricular EF has a very good assessment with a very low mean error margin and a newly emerging measurement which is called RV myocardial radiomix, which goes beyond the eye and measures the right ventricular myocardial function, almost very similar, if you would, with an RV biopsy or short of an RV biopsy to understand the path of biology and limitations of the right ventricle in PAH. Of course, there are limitations to these studies. So the AI engine is there. It helps us move forward to a continuous risk triaging and assessment. There’s a very large data that we can accumulate, but we have to implement predictive analytics and machine learning to give us clinicians the important measures that pertain to our patients and also to implement in the next step within the risk assessment in a more continuous and dynamic fashion. So there are challenges and solutions, hopefully solutions coming.
The barriers is that the data overload are immense, so we have to streamline it. There’s differences between different commercial brands. There are system constraints and not the last is the reimbursement structures. So we need to implement intelligent triaging and have good filters so the clinicians have the data, but not too much to make clinical decisions. We have to normalise against different brands. And then lastly, we have to sort out who will pay for all this work and delivery and implementation in our clinic. So with that, I will thank you for your attention and look forward to discussions.
Nils Fisher:
Great. Thank you, Professor Preston. Right on time. Our next speaker is Professor Harm John Bogard, professor of medicine pulmonologist, Amsterdam UMC, who you may remember from past webinars with the Alliance. He will discuss biomarkers for early detection and disease monitoring. Over to you, Professor Bogard.
Professor Harm John Bogaard:
Thank you very much. It’s a great pleasure to be here and to share some ideas about biomarkers in pulmonary hypertension. We will talk about early detection and disease monitoring, and for many of those purposes or for those purposes, many of the biomarkers are quite similar. So what is a biomarker? It is a clue that tells you what’s going on inside your body. So it could be something that tells you about what is happening in the pulmonary circulation, but it could definitely also tell you about the amount of stress the heart is experiencing in a situation of pulmonary hypertension. And ideally, of course, a very simple blood test would tell you everything, would give you an early diagnosis and would also tell you exactly what kind of treatment you need and whether your treatment is working well. Unfortunately, we are not there yet, even though we use biomarkers and we know quite a bit about it these days.
So let’s start with disease monitoring because that’s where we really use biomarkers the most. And it is very clear by the recent update of the European Respiratory Guidelines for the treatment of pulmonary hypertension that biomarkers play a very big role in the monitoring of patients and in making treatment decisions. So this is the updated treatment guideline with, of course, the most notable difference, the use of sotatercept in the treatment of pulmonary hypertension patients, not right away at diagnosis, but at follow-up, patients are reassessed and it is determined whether they are at a very low risk, which would tell you that the initial treatment strategy was sufficient, which is usually an antacerine receptor antagonist with a PDE5 inhibitor. But if any of the biomarkers show that patient is not in low risk at follow-up, it is considered to escalate treatment and sotatercept is playing a very important role.
So the assessment of risk status is extremely important in determining how a patient should be treated, not just at follow-up, but also at baseline. And biomarkers are exactly used for that. This is the table that physicians use to determine what treatment should be used at baseline. And all these parameters that you can see here are in one way or another, a biomarker. So biomarkers in the blood, like NT-proBNP, but also all the other parameters as derived from imaging, from echo or MRI or from an exercise test, all these variables that are being measured are basically biomarkers. And when we look in a more strict sense of biomarkers in blood, we know that BNP or NT-proBNP is a very well established biomarker in the monitoring of patients or also in the initial assessment of patients with pulmonary hypertension. And NT-proBNP also plays a very important role at follow-up with very clear cutoffs that determine whether a patient should be considered in low risk or in any of the non-low risk categories.
And that really guides subsequent treatment decisions. Much research has been done whether there are additional biomarkers in the blood or from imaging, but still when we look at the guidelines, none of these newer biomarkers are really ready for prime time because it’s also quite difficult to validate these biomarkers and to really be sure whether they add to what we already know. Right heart catheterization is of course also a very important measurement, a very important assessment, and different biomarkers from a heart catheterization can be looked at. The pressure, the mean pulmonary arterial pressure or the flow, that’s the cardiac output, or the pressure in the right atrium, the right atrial pressure. These are important biomarkers, but so far they do not play a very big role in the risk assessment and follow up. There is a lot of discussion about this, whether a repeated right heart catheterization is needed, and the updated guidelines give some guidance to that.
And what they basically say that if you determine the risk category with the standard biomarkers that I just showed, so the non-invasive biomarkers, the walk distance, the anti-proBNP, the functional class, if you’re not in a low risk category, then basically a repeated right heart catheterization is recommended to look at these catheterization biomarkers and to determine whether escalation of treatment is necessary. So we still need those invasive measurements to really decide subsequent treatment decisions.
So biomarkers have a clear role in disease monitoring. And what about early detection? Joanna also already talked about it, that the importance of early detection, and of course it would be very attractive if we can use very sophisticated AI algorithms to using all kinds of biomarkers, maybe biomarkers that are not assessed for a specific reason, but still are available in a patient record and whether that could lead to an early detection of pulmonary hypertension. But still we have to look very hard for new biomarkers on top of NT-proBNP to really tell us something. And of course we would really want to do this because the biomarkers that we looked at, exercise capacity, NT-proBNP, functional class, they’re basically all reflecting the function of the right heart. And the general idea in pulmonary hypertension that is, of course, the disease starts with pulmonary vascular remodelling, and it takes a very long time and a very large amount of the pulmonary vascular bed to be affected before right heart function is really depressed.
So it’s easy to assume that probably there’s already a lot of pulmonary vascular remodelling before any of the biomarkers that reflect RV function start to show a signal. And it would be very preferable if we would know of biomarkers that really detect pulmonary vascular remodelling before the first signs of right heart failure. And this is really lacking so far. So what we and others have been trying to do is to look at patient groups that have an increased risk to develop pulmonary hypertension and to follow them over time and to see whether these patients that eventually develop RV failure, whether they have before the development of RV failure, already biomarkers that would indicate the fact that their pulmonary circulation is being remodelled. So what we have been looking at is the carriers of gene variants that put these subjects at risk to develop pulmonary arterial hypertension.
So we’re talking in this study about carriers of a BMPR2 mutation. Of course, we know that they have about a one in three chance to develop pulmonary hypertension during their lifetime. And by following unaffected carriers of such mutations over time, we hope to find biomarkers of early pulmonary vascular disease before the real development of pulmonary hypertension. So these subjects are followed over time, they come for a yearly visit, blood is taken and other biomarkers are assessed. And those patients that ultimately develop pulmonary hypertension serve as a means to identify new biomarkers of early pulmonary vascular disease. At the same time, this study is also giving us an opportunity to describe those carriers of BMPR2 variants in a better way, because of course we assume that they are normal until they develop… You can still hear me, right? There was a muting button showing up, but I assume that I can still continue.
Nils Fisher:
Yes.
Professor Harm John Bogaard:
Yes. Okay. So we also hope that this study will give us a better description of those carriers and maybe show whether they have a real phenotype even if they do not have pulmonary hypertension. So what do we do with those carriers? We have a cohort of carriers and we compare them to their family members that do not carry the gene variant. We see them yearly. At the first round, we offer a right heart catheterization to really be sure that they’re free from pulmonary hypertension. We take blood samples, we do exercise tests, we do cardiac MRI, and we do echoes, and we do this on a yearly basis. And of course look at those subjects that develop pulmonary hypertension. We look back in time to see whether there were any abnormalities that we were not aware of at the time that they were still completely healthy.
So this has given us some insights into this cohort overall. Overall, even though these subjects do not have pulmonary hypertension, their pulmonary artery pressures are a little bit higher than subjects who do not carry the gene variant, even though it’s still within normal limits. And we also see that they have, even when they do not have pulmonary hypertension, that they have some smaller hearts, basically. They have a smaller volume of both the right heart and the left heart. So far, we have not identified any new biomarkers. We looked at PET imaging, so this is a molecular tracer that could potentially look at early changes in the pulmonary vasculature, but we could not identify signal with PET imaging in those carriers. We’ve also looked at their platelets because we had some indications that maybe some RNA expression of some genes was altered in patients with early pulmonary hypertension, but so far we could not confirm that.
But we continue to look and we very slowly over time, we identify subjects that develop pulmonary hypertension and we get some clues whether there are any new biomarkers that could be used. For example, the flow signal in the MRI, it’s not a very easy tool for screening, but we do see some early changes in those subjects that ultimately develop pulmonary hypertension that could teach us something. Very remarkably in this cohort, however, was the fact that even those subjects that on right heart catheterization were determined to have pulmonary hypertension, very mild pulmonary hypertension, but still pressures in a range of 40 or 31 millimetres of mercury, that our conventional biomarkers such as functional class or NT-proBNP did not show any abnormalities at the initial diagnosis of pulmonary hypertension, showing that even the very well established biomarkers may not be sufficient for early detection. So there’s still much to do.
We need to have bigger cohorts. We’re collaborating with many groups in Europe and also beyond to make these cohorts bigger and to hopefully get clues where the really early biomarkers of pulmonary vascular disease can be identified. All right, I think my time is up, right?
Nils Fisher:
Yeah, time is up, but we can have a quick take home message.
Professor Harm John Bogaard:
Yes. So we did identify subtle differences in cardiac morphology and in those unaffected carriers, but so far it’s a very big challenge to identify biomarkers of early pulmonary vascular disease and collaboration is necessary for this. Thank you.
Nils Fisher:
Okay. Thank you so much, Professor Bogard. And now we’ll turn to Professor Martin Wilkins, if you’re able to share your screen.
Professor Martin Wilkins:
Very good. Which screen have I managed to share? Have I managed to share any screen?
Nils Fisher:
I think it’s the other one, or you could drag your presentation over.
Professor Martin Wilkins:
Let’s try that.
Nils Fisher:
There we go.
Professor Martin Wilkins:
Now I need to, excuse me, while I just move things around. Does that work at all?
Nils Fisher:
No. You can hit swap displays there. Ah, there we go. Perfect.
Professor Martin Wilkins:
Thank you. So yes, I’m Martin Wilkins. I’m professor of clinical pharmacology from Imperial College. And what I’d like to chat to you about today is the path to personalised medicine. We’re working towards tailoring treatment to individuals. I’m conscious that some of my talk will overlap with what Professors Preston and Bogart have already covered, but hopefully the narrative will condense to a common goal. So where to start? Well, we’re all different. I hope you were sitting down. If that came as a shock to you, we all like to think of ourselves as different, but that doesn’t stop us from trying to cluster people together according to common characteristics. And that’s where problems can start. And one of my favourite stories illustrating this comes from the 1940s when the American Air Force were trying to customise or optimise the cockpit for their new fighter aircraft. They called him one of their own, a guy called Gilbert Daniels.
He had some anthropology training and decided to make some measurements of as many American servicemen as he could. You can see he was quite thorough. No only did he rely on arm length and leg length, but he also took measurements of the head and distance between the eyes. And he ended up with 132 measurements from over 4,000 pilots. And then he started analysing his data and he decided that he would take the average of 10 measurements that he thought pertained or were most useful to the design of a cockpit. And he took the average of these 10 measurements. He took a very generous average, a middle 30% range. And then he asked how many of his 4,000 pilots, how many of their measurements or how many of the pilots had measurements within that average range? And the answer was zero. And so he lightened, if you like, his task a little bit.
He asked the same question of three dimensions and less than 4% of airmen matched all the average values for the three criteria. So his conclusion was that the idea of an average person is a pitfall. It’s impossible to find the average man in the Air Force. And you might wonder, “Well, what’s this got to do with medicine?” Well, the thing that pertains to medicine is that when we are prescribing, when we’re developing drugs, we rely very much on average values. So in a typical clinical trial, we might take a population of patients with the disease of interest. We will divide them into at least two groups, one of which receives the placebo and one the active treatment. And then after a period of time, we look at the average response of those two groups. And if there’s clear water between those two responses, then we declare the study a success.
The drug becomes one of interest and may well get a licence for use in that disease population. And the next time we see a patient and we’re wondering what treatment we might prescribe, we will look at these average results and expect or hope that that patient will fall into that average group. So we have this assumption that the patient population, the trial population is sufficiently similar to the patient so that trial’s result is relevant to that patient. But we all know that patients vary in their response to drugs. Some have good responses, some partial responses, some no responses, some end up with adverse events. And yet we vary that to some extent when we’re making our prescription for each patient. We know very well if we go back to the average data and we expand that box which characterise the average, we will see a distribution of results of patients around that mean.
That’s true for the placebo. It’s also true for the average drug. And if you take our patient sitting in front of you, we don’t know which part of that distribution that patient comes from. We don’t know which of those dots represent the patient and their likely experience. And so we have this problem where in clinical trials we have an average treatment effect which hides heterogeneity. It represents a population property, but it doesn’t tell us so much about the response of individual patients. And this brings us to personalised medicine where we want to move away from this current approach of one treatment fits all to a more personalised approach where we’re trying to choose the right drug in the right dose for the right patient at the right time. So how can we make progress towards this? Well, the first thing we can do is look more carefully at the responders and the non-responders in any clinical trial population.
We can take that distribution of responses. We can draw a threshold, in this case, 0.2 or 20%, a 20% improvement representing something that we would value as a clinical response. And then we could look at the patients that exceeded that response and compare them with the patients that failed to match that improvement. There is an assumption here in that those responses are consistent. In other words, if we were to repeat the study, we would get the same patients occupying the same space in this distribution, but we’ll ignore that just for the moment. And we’ll look at the two populations as they emerge from this division and we can look at the characteristics of these two populations and try and describe them with greater precision. So we could look at responders and say, well, maybe they are a younger group. Maybe they have a different image pattern to the non-responders.
Maybe their right heart function differs, maybe it’s better preserved. And as we’ve heard from Harm, maybe we can look at biomarkers too. The number of descriptions that we can apply to this population is still quite limited when it comes to trying to tailor treatment.
So we can turn to things we can measure in blood. And again, we’ve heard from Harm a little bit about this. There are many things we can measure in blood these days. We can look at gene profiles, we can look at proteins, we can look at metabolites. These are scalable, they’re quantifiable. They offer a greater vocabulary, if you like, with which to describe individuals and enable us to tailor treatment. There aren’t many examples to date, but here’s one that I thought we might spend a little bit of time on. We know that genes influence response to drugs. We know genes can influence protein levels. And by measuring genes and measuring proteins, we might be able to look at how levels, high levels of expression, if you like, how abundance of these proteins might correlate with response to treatment. And I’ve pulled a recent poster from the European Respiratory Society here, in which data from a clinical trial, which overall wasn’t successful, could be reanalyzed according to patient’s genotype and appear to pull out a response.
In other words, in future studies, maybe one might select patients with a particular genotype in order to anticipate a response to treatment. What about prescribing at the right time? Well, we know patients are on a journey. As we’ve heard from Iona, we often have snapshots of patients rather than a full video of their journey. So how can we address the question of when to prescribe the drug as well as what dose to prescribe? And here we come back to her presentation around the use of biosensors and the potential for continuous monitoring. We know about wearables. Those are readily marketed and many patients have wearables. Increasingly, I think there’s an interest in implantables, biosensors you can implant in the pulmonary artery to measure pressure, sorry, or in the chest wall to measure heart rate. And we have done some studies with implanted sensors and wearables ourselves.
And you can see on the right-hand side of this slide here, changes in pulmonary vascular resistance and heart rate, which vary with time and in response to treatment. And we can use these data to inform the changes in dose and the addition of new drugs. The big question then is how do we integrate all this knowledge, the clinical data, the things we can measure in blood, the variables into a narrative which relates to individuals, to individual patients? And here I’d just like to introduce the concept of the digital twin. So you might have come across a digital twin because engineers have used it in many walks of life. It’s used to monitor aircraft. So every time you take a flight, when the plane lands, it updates a computer model of itself with information about its journey. Here we’re talking about a computer model of patients.
So this is a mathematical model, if you like, which is fed data, data from hospitals, data from wearables, data from blood tests, even data that one might collect and transmit from home. And those data are compared with the computer model’s experience of a population of patients with pulmonary hypertension and comorbidities. It exports its output to a dashboard, which is visible to the patient and to the doctor who can then inform and advise the patient. And it continuously updates itself, so it’s a dynamic model. It’s an interesting approach. It’s still in its prototype form. I’ve mentioned it’s already available in other walks of life. And I don’t know how many of you follow Formula One, but every time a Formula One car races around the racetrack, the driver is updating or asking for updates from the pit wall as the car is racing around the track about tyre pressures, about fuel load, about distribution of brakes.
So it’s getting continuous information, continuous updates and continuous tweaks. I don’t know how many of you would like to think of yourself as a Formula One car, but I think this is the direction of travel for how we use all the data that are becoming available to informing and personalising care in the future. So just to summarise then, I think we all recognise patients vary in their response to drugs. Clustering patients into groups may mask the benefits from those drugs. Personalised medicine involves tailoring treatment to individual patients, and this requires an in-depth knowledge of each patient and for that information to be updated regularly. This presents a challenge to personalised medicine, but I think advances in technology offer a promising way to address this challenge and provide personalised medicines in the future. Thank you very much.
Nils Fisher:
Thank you so much, Professor Wilkins. That was great. Our final presentation is from Professor Sudarshan Rajagopal, associate professor of medicine and co-director at the Duke Pulmonary Vascular Disease Centre. He will be discussing new imaging approaches for screening and diagnosis of pulmonary hypertension. And I don’t know if he’s managed to join, and if not, I actually have a video of his presentation he kindly sent to us ahead of time. So I’ll give him just a minute to unmute if he’s on the call. It doesn’t appear like he’s on the call. So let me go ahead and bring up the video. Okay. Hopefully everyone can see my screen here.
Professor Sudarshan Rajagopal:
I’d like to thank the organisers for the opportunity to speak to all of you today. I’ll be talking about new imaging approaches for screening and diagnosis of pulmonary hypertension. Here are my disclosures. So when we talk about pulmonary hypertension, it falls into this larger category of pulmonary vascular disease. These are diseases that are associated with remodelling, abnormal remodelling of the pulmonary blood vessels. So instead of being nice, big open pipes, they get remodelled and they become small straws, and that makes it very hard for the right side of the heart to pump blood effectively through them. So you can see here you’ve got a pulmonary artery. It should be nice and open. Instead, it’s blocking with all of this tissue that’s grown here. This can also affect other pulmonary blood vessels like the capillaries and pulmonary veins. Currently, when we see patients in clinic, we have a number of tests that we perform.
These include BNP and NT-proBNP. These are blood tests for the stress of the right heart. They’re also sensitive to volume status. So if someone has a lot of extra swelling, they’ll tell us about it. It’s not going to tell anything about pulmonary vascular disease directly. The six-minute walk test we do frequently to assess exercise capacity, many things that impact the six-minute walk test. While straightforward to do clinically, if someone just isn’t good at walking or they’re injured, they’re not going to do a great walk. Now, when we look at imaging of the heart, we perform this to look at the right side of the and we can do this with cardiac magnetic resonance or echocardiography. Again, this is an indirect approach to looking at pulmonary vascular. It’s telling us how the right side of the heart is dealing with that increased.
And also we have the gold standard for monitoring pH, which is regular catheterization. We get a pulmonary vascular resistance and a mean PA pressure that’s invasive. So rather indirect assessment of PVD, it’s something about resistance and pressure. It’s telling us where these abnormal is occurring. So there are a number of new imaging approaches that are being developed to study the pulmonary vasculature in more detail. I’m going to tell you about a few of these today, and I’m going to use a number of slides from my colleagues who I worked with the World Symposium Pulmonary Hypertension a couple of years ago that. These are slides from Martin Ramichiarden. This is a technology using non-contrast CT scans to look at the small blood vessels of the lungs. As you can see here, usually we think about using a contrasted CT scan to look at blood vessels, but this doesn’t even require IV contrast.
And you can see that you can see the small blood vessels here in yellow on edges. And these are the vessels that can be impacted by pulmonary hypertension, so-called VV5 vessels that are smaller than five millimetres squared in sectional area. And a lot of this work has been led by Nick Rahagi and underlies the technology of.
You can do more than just quantify them. Many, you can also characterise them by looking at their tortuosity and 3D fractal imaging, other sort of fragmenting it from the larger blood vessels. This technology is quite robust now and we can separate out arteries from veins. The studies have been performed that show that patients who lose these smaller blood vessels have higher mortality than folks who do not. And this is an example in patient with PAH compared to a healthy control patient here. And there have also been studies that show that if these small blood vessels are sensitive to interventions such as inhaled nitric oxide in the investigational drug cell. Now, positron emission tomography is a way to actually image molecular targets, and that’s by attaching a positron to them through an F18 label or a C11 label. And then so this in the left-hand side here, we can look at glucose uptake in granites that are treated with monoclonal.
So they develop pulmonary hypertension. You can see increased signal in their lungs compared to the controls that pulmonary FDG uptake, and they also get enlargement of their right ventricle. Similar studies can be formed with acetate and with acetate you actually decrease the signal of PAH compared to controls with this border effect. And then you can look at molecular targets like FDLT, which is decreased in patients with hereditary or idiopathic age, much more so in hereditary patients compared to controls. But right now we’re really limited by a lack of knowledge of what the ideal molecular targets are if we’re going to develop these sorts of probes, but a number of groups have studied. I’m going to tell you a lot about some work I’ve been doing with my colleague, Dr. Baskin using hyperpolarizing. This MRI scan of the chest, and usually the lungs just look black.
There’s not much signal there, and that’s because a typical MRI looks at proton signal and that’s from water. And ideally you shouldn’t have water in your lungs, but you can get around this by using a contrast agent that’s inhaled. In this case, hyperpolarizenon, and this patient inhales hyperpolarizenon, you can actually see the lungs and you can see with these marked abnormalities here due to fully from COPD. The really cool thing about xenon is it actually gives different signals depending on where it is in the lungs. So it’ll give different signals in the airspace, the membrane tissue, and in red blood cells where it actually finds the haemoglobin. So with a single breath hole and a two-minute scan, you can image gas exchange in airspace membranes and red blood cell signals, and you can quantitatively map this, compare this to controls. And so this patient here, they’ve got pretty normal airspace signal, and you can see their membrane signal is quite high here associated with very abnormal RPC signals.
And this is classic for a patient with idiopathic pulmonary fibrosis where they have a lot of scarring lung tissue there that takes up the xenon while they are not doing a good job of getting xenon into their blood. And indeed we can see these types of patterns where a healthy patient’s all green, a COPD patient primarily has airspace disease, an IPF patient primarily has membrane disease with some RPC defects. While patients with PAH and left heart failure primarily have red blood cells, they’re pretty normal membrane airspace signal. But if we look at these changes in the blood cell signal over time, we can see how markedly abnormal it is in PAH compared to other disease states, and that’s because they’re not able to get enough blood into the pulmonary capillaries with each pump of the heart. And this is because of the high resistance of those vessels.
And indeed these signals that we observe are related to pulmonary vascular resistance last year. Now what’s interesting too is that these signals can also report on response to therapy. There’s a patient treated with inhaled treprostinil or short-acting inhaler used for treating PAH and Phenotrial A. You can see before drugs, there’s a lot of red signal in their RBC, so abnormal signal. This improves with treatment with the inhaled treprostinil, but then three hours afterwards you can see it gets back to being just as bad as it was before as the drug ished out.
In the last couple of minutes, I’m going to tell you about the role that AI can play in helping us with PAH diagnosis using these novel imaging approaches. So this is work from David Kiley’s group in the Aspire registry in Sheffield where they’ve used AI to automatically segment CT scans of the pulmonary arteries and inpatients with PAH. And they can assess both the right part and its enlargement here, the remodelling of the left ventricle, which gets squished by the right part. So it’s become very e-centric. You see LVD centricity here. They can assess PA volume here in white. That’s the pulmonary artery, very enlarged. And from this, they came up with a CT diagnostic model for PAH to help. And again, a lot of CTPA studies are performed and now you can use some of that data to help diagnose patients with PAH. Now, there are imaging approaches to look at the heart, classically echo and cardiac MRI.
Cardiac MRI gives you much nicer pictures as you can see here of the heart, which I’m highlighting here. There’s a lot of stuff in the abdomen. Echo is a little more prone to variation from sonographers and image quality, but CMR can give you really great image quality that can be quantified. And David’s group has done that as well in the ASPIRE registry where they developed AI approaches to improve and automate biventricular segmentation. You can see how they can tweak their AI to get great separation between the right ventricle here and the left ventricle here. They’ve done this in many patients and many MRI scans, and they’ve shown that some of these features, including novel features, that the AI can pick up itself, so-called feature analysis that are associated with both diagnosis of PH and predicting out mortality. And so these types of techniques are now being much more widely applied to our imaging approaches and can get around that problem where you have a single viewer who reads your MRI scan and says maybe it looks pretty normal.
Now we can actually use AI to get a much more objective standard. And this approach can also be applied to echo, even though echo has worse image quality than CMR, these AI approaches are so robust that we can pull out useful data. This is using AI to diagnose patients with PH in patients at Duke and in Kremlin receptor in France, work with Francis Bauer. And you can see that with severe pH, we have excellent diagnostic behaviour here with the AURC, less so with mild and with good validation in the French cohort. This is all in collaboration with us recently presented at ESC. So with that, I’d like to thank you for your attention and the WSBH Task Force six members from whom I literally stole slides, my collaborators and colleagues at Duke and S2A out in France and at Stanford. I’m happy to take any questions.
Thank you.
Nils Fisher:
Hi everyone. That was great, and thanks to Professor Reddle, and thank you to all the speakers. If you don’t mind turning your cameras back on, it seems like most of you have. It’d be great to do a little bit of a Q&A session. I’m just looking through the chat now and seeing if anyone has any questions. And if you have any, please go ahead and add them to the chat if you haven’t already done so.
Joanna R. Preston:
Well, gentlemen, let me start off with, we’ve talked a lot about very advanced technologies. They’re all expensive and we know pulmonary hypertension occurs worldwide and the opportunities to make the diagnosis to implement a lot of diagnostic tests is very variable. So I think as we are evolving and developing top-notch, high quality and very sophisticated measures, I think we also have to keep in mind if we can develop tools that can be applied worldwide that don’t require as much sophistication, that can identify patients in the parts of the world that do not have these very advanced technologies.
Nils Fisher:
I think that’s a great question. Thanks so much for leading with that. Anyone want to jump in?
Professor Martin Wilkins:
It’s a challenge for sure. Maybe Harm’s got the answer, but I think one is hoping for improvements in technology that are more easily deployed. A digital stethoscope might be cheaper than an echo machine in the long run, who knows? But yeah, the ones looking for innovation in this space.
Joanna R. Preston:
Right. And I don’t think that should stop us from developing new algorithms and new pathways, but we also have to keep in mind how can we develop tools that can be implemented worldwide at some point at least.
Professor Harm John Bogaard:
Yeah, I absolutely agree. But for a rare disease, it’s going to be very difficult to develop these tools, I think. I just recently heard about a study in COPD patients who of course can have exacerbations and somebody developed a tool just using a cell phone and recording the voice of a patient, and the voice was already giving away early signs of an exacerbation and just having a patient speak into the phone to the physician was enough to make a diagnosis. And of course we would love to have such things for pulmonary hypertension, but I guess it’s going to be very difficult to develop such tools. But I agree with you. I really like to talk about PET scans and all kinds of fancy imaging tools, but that’s never going to be widely available across the globe. MRI is not going to be available for all patients.
So we can use these tools for discovery and maybe to have a way of validating other tools, but we will ultimately have, we need to have a simple blood test or something like a voice or something really easy. Maybe heart rate is going to tell a lot already. Analysing heart rate can tell you a lot about cardiac function.
Joanna R. Preston:
Heart rate recovery that new cardiac performance. But the EKG itself, I think the AI tools that screen the EKG and pick up RV strain and signal it to the physician, I think it’s a good start.
Nils Fisher:
Yeah. And that would make it easier, I think, to also include in clinical trials if there are more simple tests that are widely accepted, that might make things easier. I just wanted to, as we’re really close to time, there were a couple of questions in the chat, and one of them was about how women have been underrepresented in clinical trials. I don’t know if Professor Wilken, you can speak to that a little bit.
Professor Martin Wilkins:
Indeed, that has certainly been the tradition, particularly in early phase clinical trials, what we call phase one, where we’re uncertain about the toxicity of the drug. I think that’s less so now. I think it’s more inclusive in the early phases. And of course, when you get to phase two and you are doing studies in the patient population and females are more overrepresented in pulmonary arterial hypertension, then certainly in phase two, you’ll see a greater representation of women in studies.
Nils Fisher:
No, thanks for that. I think that is it. We’re getting towards the end of the questions there, and I think maybe there was a question about the evidence of working mechanisms of new molecules in clinical trials. I think I’m not completely understanding this question, so maybe…
Professor Martin Wilkins:
I think sometimes drugs are introduced where we know partially how they work, and that tends to be more common than having the full picture. And I would say sotatercept is a prime example of that. We have an understanding of how it binds to key proteins, but how it’s working I think is still going to be a subject of research for a little bit of time yet.
Joanna R. Preston:
Yeah. Also, prostacyclins, they were introduced in humans way before we had a clear understanding of their cellular mechanism, but there were clear signals that in humans it works. So sometimes we go ahead for better or for worse. Some Sometimes we friend ourselves. I think the digital twin in a concept in clinical trials is fascinating and it may not be prime time for a rare disease, but I think we should keep our mind open to maybe trying to apply to our future clinical trials in PAH.
Nils Fisher:
Absolutely. And with that, I think we’re right at time. So thanks again to all the speakers. This has been a really interesting session. And please visit the PH knowledge sharing platform at ph-ksp.com for webinar resources and news from the Alliance. And we look forward to welcoming you to a future webinar. And thank you so much and goodbye.
Professor Harm John Bogaard:
Have a good night. Yeah, bye-bye.