Digital technology in PD
Hello my name is michelle who i’m a neurologist from oxford university and i’m going to be talking to you today about digital wearable technology in parkinson’s to date around 23 billion us dollars have been wasted on failed drug trials in the quest to develop a disease modifying therapy that might slow down or even stop the progression of parkinson’s i’ve
Summarized some of the recent studies here in this slide reasons for this are multiple but could be summarized as three main factors firstly individuals with parkinson’s recruited at the time of their diagnosis with emergence of motor symptoms are likely to have been recruited too late in their disease journey at diagnosis most individuals will have a history
Going back at least 20 years of non-motor symptoms including rem or rapid eye movement sleep behavior disorder where participants enact their dreams or depression constipation and other features early diagnosis of parkinson’s can also be challenging particularly when patients have not yet received treatments parkinson’s can mimic other disorders however over time
With increased clinician review the diagnosis generally becomes clearer and more accurate the second reason is that parkinson’s is a heterogeneous condition meaning that it has great diversity both in how individuals present with their parkinson’s at baseline and how they subsequently progress in terms of motor function and cognitive function an illustration of
This is a study we led in oxford which uh looked at the oxford discovery cohort and combined with the tracking discovery parkinson’s cohort from the uk these two cohorts are one of the largest incident cohorts and what we were able to do was use an unbiased data-driven approach combining all of the motor non-motor and cognitive variables to define four baseline
Clusters how does this matter to digitals well the important thing is that when you look at how the baseline clusters then went on to progress in the next three years we see that very similar mirrored both in the tracking and the discovery cohorts but the difference between the fastest uh most progressive cluster shown here is cluster one and the slowest uh cluster
Cluster four was equivalent to three motor updrs points per year this is a clinically meaningful pd trial endpoint currently used however these differences shown are not related to the effects of a disease modifying treatment but simply due to diversity or heterogeneity baseline if we fail to account for this we may have false positive or false negative trial
Results lastly current measures may be inaccurate or insensitive to detecting change parkinson’s is a slowly progressive condition the mds updrs3 motor school is a human rating scale that is uh extremely popular as an outcome measure for clinical trials yet it relies on human raters who are subject to both intra and inter rate of variability rating people with
Parkinson’s who may themselves be prone to fluctuations during the day the graph on the left shows what is typically plotted as the mean score for updrs in the purple dotted line this is shown here as equivalent to 2.5 motor points per year however in the spaghetti junction of different color plots you also see individuals here with parkinson’s who fluctuated the
Most and indeed some individuals like the participant in green changes by 40 points over four visits it is difficult to know whether this degree of change is true change or related simply to the inaccuracy of the measurement task so digital wearables are united in their hardware meaning that they all possess within them an imu or inertial measurement unit which
Is then incorporated into an accelerometer gyroscope or magnetometer single imus cost around 100 pounds but multi-sensor kits that can be bought on the commercial market can cost over 10 000 pounds these sensors are divided in how they can be used is it better to do simple occasional measurements in clinic or to get more frequent measures at home data can also
Be collected in an active way where the participants asked to do the test at certain times or in a passive way where continuous gate for example is measured the analytical techniques vary widely and are typically machine learning algorithm approaches while the data the raw data the imu collects is similar the output metrics can be collected in a different way due
To protocols and the subsequent application calculation can also vary lastly the level of clinical phenotypic detail is often less uh in patients uh who have had lots of digital recordings as opposed to well phenotype cohorts where typically numbers of digital uh readings might be quite low our aim is to apply digital wearables to empower researchers to improve
Their patient selection and outcome measures for patients undergoing clinical trials we want to empower clinicians so that they can use digitals to inform disease management to do remote telemedicine approaches and to maximize available resources and importantly we want to work with people with parkinson’s to ensure that we can give them digital tools to monitor
Their parkinson’s and to facilitate symptom self-reporting so that they are not a passive recipient of their publisher’s care but are actively managing their motive and non-major symptoms i’ve summarized some of the challenges here of digital wearables both the device and analysis based patient based factors including maintaining privacy accounting for diversity
And exposure to previous apps smartphones and computer and email use and compliance or adherence without which and without the patient and caregiver engagement digitals will not work it is important to also recognize that digitals need to include patient relevant outcomes and that they may also provide metrics on hidden symptoms but up to now may not be measured
With standard questionnaires fda and ema really have not produced clear guidelines on what digital endpoint criteria should be used in their validation and this is something for those of us trying to develop these tools that can make life quite difficult at times in opdc we’ve developed a 10-minute smartphone app that evaluates seven aspects of motor function
Shown here including speech balance gait uh manual dexterity reaction task rest and postural tremor we can also look for a symmetry of signs by measuring tremor and finger tapping bilaterally on both sides as well as measuring axial symptoms with speech balance and gait the app importantly is nested within the discovery cohort study funded by parkinson’s uk this
Longitudinal cohort study recruited nearly a thousand people with early parkinson’s in a thames valley population base of two million people shown here in red it also recruited nearly 300 people with sleep study diagnosed rbd in sheffield cambridge and oxford and lastly we had 320 agent gender matched control participants all patient groups took part in 18-month
Assessments incorporating motor non-motor assessments with imaging biosampling and additional digital metrics our first study looked at how well the single smartphone test could separate people with pro-draenor parkinson’s and early parkinson’s from age and gender match controls and using mla approaches we achieve reasonable sensitivities and specificities with
Overall separation accuracies of between 85 to 92 percent remember this is based on the single smartphone test alone no additional information is given to the computer to make the decision christine lowe then took the smartphone test to see whether it could predict the onset of clinically relevant outcomes for people with parkinson’s this included falls freezing
Impaired balance cognitive impairment the need for help with doing hobbies or general tasks along around the home and she was able by doing smartphone recordings in participants who went from not falling to falling to be able to predict these onsets uh 18 months before occurrence with reasonable accuracies again in the range of between 0.8 to 0.99 ie between 81
To 99 percent one can then do an individual smartphone test and a particular time will give a future probability of their false risk freezing cognitive problems or need for extra help with different tasks and shown here you can see the normal range where these smartphone recordings people with measures in that red range do not need have a low risk of going on to
Develop these problems whereas individual patients falling outside the range have a much higher risk of these future complications christine lowe has then spent many uh of the last two to three years developing an overall measure of motor severity using the smartphone app that we call a composite motor score she’s taken a number of existing clinical measures shown
Here to develop the composite score she’s then looked to compare it with the mdsu pdos part three and she’s shown that the composite motor score has lower variability relative to the mean a greater score range particularly in people with rbd and controls where it shows less floor effect and lastly that it has greater mobility to track longitudinal change over time
She’s also been able to show that the 10 minute smartphone test can predict this composite motor score so the participants don’t have to go through time-consuming investigations next steps for us in the app are to use it as an exploratory outcome measure in the exenatide phase 3 trial and to compare the performance of the app with the mds updater s3 rating scale
And this will provide for the first time a direct comparator i’d like to end by thanking all the participants in the study as well as the researchers whose brilliance and perseverance have made the study possible christine lowe’s siddharth aurora and nevin cory thank you
Transcribed from video
Clinical Science I – Michele Hu By World Parkinson Coalition
