Good morning, everyone. Great to be here in Washington. So I’m Patrick Harrington from London. I’m going to talk about digital data capture, particularly focusing on wearable devices and some other novel approaches to data capture that we’ve been focusing on in the UK.
So this is a really, like the previous talk, this is a really hot topic currently. The image on the right here was from a recent review article in the New England Journal of Medicine, and it shows some of the ways that digital devices are being utilised across healthcare currently...
Good morning, everyone. Great to be here in Washington. So I’m Patrick Harrington from London. I’m going to talk about digital data capture, particularly focusing on wearable devices and some other novel approaches to data capture that we’ve been focusing on in the UK.
So this is a really, like the previous talk, this is a really hot topic currently. The image on the right here was from a recent review article in the New England Journal of Medicine, and it shows some of the ways that digital devices are being utilised across healthcare currently. So we can see that in patients with type 1 diabetes, a wearable device aligned to a continuous glucose monitor and an insulin pump is now becoming a standard of care for these patients. And there are some novel applications as well. So for instance, in patients with epilepsy, wearable devices are now being used to detect seizure activity. So I’m going to discuss some of the ways that we’ve been using this in our MPN patient population.
So we set up something called the MyMPN Voice app a little over two years ago now, and there’s an associated research study alongside this in which all patients with a classical MPN diagnosis in the UK are eligible to enrol. This has a unique study design with virtual recruitment, where patients are not required to attend any particular hospital site. They can enrol virtually. They are issued with a patient information sheet online and then electronically sign a consent form. So all the normal steps are adhered to. And using this approach, we have thus far recruited over five years… in the UK. These patients have a median age, which is very much representative of the MPN patient population, of 58, and we’re pleased that we have several patients in their 80s using the app, and we’re collecting a huge amount of data using this approach, over 3.4 million data points thus far collected. We’re also pleased that we’ve managed to maintain interest and activity in the study for patients who have been enrolled for a longer time period. So of patients who have been on the study for over 12 months, of 264 patients, over three quarters are remaining active on the study, and we have a weekly response of four entries.
So this is how the app looks on a patient’s smartphone. You can see we are primarily monitoring the patient-reported outcomes, so focusing on the MPN-10 total symptom score, as shown there. But we’re also looking at some other PROs, so the EQ-5D-5L is a quality-of-life metric that many people will be familiar with, and this is not MPN-specific, but it’s used across many chronic diseases. And you can see that these trends in symptom burden and quality of life can be monitored over time to assess for any changes, such as disease progression, and assess the impact of therapies.
Another aspect that I suggested previously is the collection of biometric data. And we’re doing this using the device shown here. So this is a Withings ScanWatch 2, which is an FDA-approved wearable device that feeds data directly into the app. And this uses the technique of actigraphy, or measured accelerometry, so looking for small movements multiple times per second to provide information on sleep quality and activity.
And these two images are taken from patients who attend my clinic and who are involved in the study and using the app regularly. So on the left here, you can see a patient who was monitoring the frequency of her headache severity and monitoring the impact of a treatment to control this. And she liked to take the old-fashioned approach of printing out all her data to show me. And on the right, you can see a patient who has recently started a JAK inhibitor therapy and chose to show me electronically the downtrend in their symptom burden after doing so. And what we’re hoping is that this process is being replicated across the country by patients who are enrolled in the study, to allow patients to improve the monitoring of their condition and really take control of this, and share this with their clinician when they attend their appointment to show what’s been happening between the appointments.
Okay, so let’s move on to some of the data from the associated research study, focusing initially on the activity and sleep metrics. So what we have here is the activity and sleep broken down based on MPN subtype. And what we see, as we may expect, is that patients with a diagnosis of myelofibrosis have a significantly reduced step count, reduced exercise of soft intensity, and reduced elevation compared to the other MPN subtypes. And similarly, when we look at sleep quality, these patients also had a reduced total sleep and light sleep duration when compared with ET and PV.
If we move on to the different MPN subtypes broken down based on the treatment arm, what we have here is the PRO data for essential thrombocythemia patients, and we can see that patients taking hydroxycarbamide had the highest symptom score and associated lowest quality of life. And in contrast, patients not taking any treatments or undergoing active surveillance had the lowest symptom score and the highest quality of life. And on the right we can see a spider diagram which shows how the symptom burden is broken down across these treatment arms. And you can see, as was mentioned yesterday, that fatigue is really a key issue for our ET patient population and, in fact, all of our MPN patients in this study. By far the most prevalent symptom.
Moving on to the biometric data in our ET patient population, so there’s quite a lot of data on this slide, but what we’re showing is various different activity metrics on your left, and what we can see essentially is that patients taking hydroxycarbamide had significantly lower activity across multiple variables when compared with active surveillance and interferon-treated patients. And again, this is as we might expect, given hydroxycarbamide is typically reserved for our older patient population. On the right, though, we have some data on the sleep, and this is quite interesting. So what we see in the light blue bar here, this is patients taking interferon, and these patients had a reduced total sleep duration as well as deep and light sleep duration when analysed separately. And this is in keeping with what we see in patients with interferon toxicity, so it’s quite interesting to see that sleep seems to be affected in this wider interferon patient population.
Okay, moving on to PV, what we see here is that patients taking ruxolitinib had by far the highest symptom score and lowest quality of life. We should bear in mind that ruxolitinib is reserved as a second-line therapy in the UK for patients with PV. So when we look at the two first-line therapies, what we see actually is that interferon-treated patients had a higher symptom score than hydroxycarbamide, interestingly. And on the right, we have some data on average heart rate, both during activity and sleep, showing a higher average heart rate in patients taking interferon and active surveillance. And really, this demonstrates the breadth of the data that we can collect using this approach.
Moving on briefly to myelofibrosis, this is our smallest cohort with the largest number of different treatment options. But what we can see here is that patients taking momelotinib had, again, the highest symptom score and associated lowest quality of life, notably lower than ruxolitinib. But again, this reflects the real-world nature of our study. So momelotinib is typically used as a second-line JAK inhibitor and also would typically be used in more advanced disease, where patients have concomitant anemia.
Okay, so this slide shows a correlation matrix looking at correlations between all of our individual variables across the study, so all of the individual activity metrics, sleep metrics, and all of the patient-reported outcome symptoms. So firstly, what we see at the top here is a strongly positive correlation, as we would expect, between all of our activity metrics, but we also see a significant inverse correlation between various symptoms and the activity metrics. So if we look at the highlighted bar there, it should show the MPN-10 total symptom score, and we see a significant inverse correlation between the MPN-10 TSS and various activity metrics, demonstrating that patients who report a high symptom score genuinely do have impaired activity, and this nicely brings together these two forms of data.
Okay, so finally, what are the next steps with regards to this project? So as part of the enrolment process, we ask patients to complete a subject access request form, and this allows us to access relevant aspects of their electronic health records. And so thus far, we’ve received paired data from over 300 patients, so both hospital and primary care records. [We are using machine] learning processes to interrogate and analyse this data, but we’ve received huge amounts of bulk data in PDF format, so it’s difficult to analyse both the structured and unstructured data. And some early insights of what we’re doing here are shown on the right here. So we’re starting to compare the correlation between laboratory values from the full blood count with both the symptoms and the biometric data, and broadly what we’re seeing is there seems to be significantly greater correlation between the biometric data, so the activity, and the blood count values than there is with the symptoms. For instance, at the top here, we have the MPN-10 total symptom score versus hemoglobin, with no correlation at all, whereas below, in our ET patient population, we see that patients with a higher platelet count genuinely do seem to have significantly decreased activity.
So finally, what are the future steps with regards to development of these kinds of platforms for patients across healthcare generally? We should be aiming to design platforms that can link multiple different healthcare providers and directly link to electronic healthcare records. That’s something that we had aspired to when we set up this project, but it was increasingly complex in the UK due to the vast number of different electronic healthcare record systems that are in use, but that would be really beneficial for our patients, actually. We should also aim to develop platforms that can provide direct interaction between patients and their healthcare team. And as we’ve heard in the previous talk, we’re going to continue to see increasing use of machine learning and large language models in this kind of work. And we should always remember to design platforms that cater for a diverse range of socio-economic and educational backgrounds. Thanks.
AI-generated transcript. Please refer to the video as the authoritative source.