Pfizer’s Digital Health Approach to Generate Better Endpoints
Pfizer’s Dr Xuemei Cai, Senior Medical Director, Digital Medicine, discusses their approach to leveraging digital health tech to solve drug development challenges, including a digital endpoint for gait. Pfizer’s Senior Data Scientist, Lukas Adamowicz, sheds light on the unique data algorithm package used to validate gait.
What is Pfizer’s approach to digital health technologies for clinical trial usage?
Xuemei Cai: We built a first-in-industry clinical research unit dedicated to evaluating digital health technologies prior to use in an interventional clinical trial. This allows us to rapidly evaluate devices, algorithms, software, data handling/security, site experience, user experience, and more. We have conducted analytical validation studies as well as built and validated tools, such as SciKit Digital Health, from data obtained in the lab.
We can obtain highly annotated and high-quality raw data in a comprehensive database across different ages and populations. We have further used this age-matched, gender-matched data to look at construct validity (known–groups) when assessing against a clinical trial population with the same endpoint.
In order to always make sure that the endpoint being developed is meaningful to patients, we talk to patients directly. Increasingly we have been partnering pre-competitively across the industry to share and understand what is meaningful to patients, as we tackle the challenges in the current drug development environment.
Your team validated a digital endpoint for gait. What was it about the existing gait measurements that necessitated improvement?
XC: It is difficult to elicit the full range of gait variability and gait complexity that humans normally encounter day to day. One existing gait measurement is the Short Physical Performance Battery (SPPB), which contains only a very brief gait assessment over a few meters. Another is the six-minute walk test. It may not reflect naturalistic gait patterns due to the Hawthorne effect, where people modify behavior in response to an awareness of being observed, and the layout of the straight track.
The ability to accurately measure and analyze gait for longer durations in the environment where function matters most has grown exponentially in recent times.
"Increasingly we have been partnering pre-competitively across the industry to share and understand what is meaningful to patients, as we tackle the challenges in the current drug development environment."
How are you continuing to innovate on this?
XC: We have been experimenting with replicating the complexity of gait by introducing tasks that require both shorter and longer bouts of walking in our studies for the validation of algorithms for gait assessment. We have also introduced non-walking active tasks to train our algorithms to distinguish between walking and other non-walking active tasks seen in the real world.
What is the importance of measuring gait?
XC: Gait is sometimes referred to as the “sixth vital sign” because it is an important marker of physical function and health status. Gait metrics can be used to track and predict clinical outcome measures across many populations and pathological conditions impacting cardiopulmonary, neurological, and musculoskeletal systems.
In Duchenne Muscular Dystrophy, for example, patients and families cited the meaningfulness of losing top walking speed as a major health concern. The importance of gait helped point scientists and regulators towards using 95th percentile stride velocity to assess efficacy for new interventional therapies.
How did you communicate the value of this digital endpoint?
XC: Data speaks volumes. We publish our results extensively and have engaged in multiple pre-competitive consortia such as MOBLISE-D with other companies and device manufacturers who are working on showing the value of at-home monitoring.
Furthermore, we also conducted qualitative interviews with patients and participants to assess the meaningfulness of these measures, and published on the comfort and usability of these devices in a range of patient populations. As the body of literature and scientific consensus grows, I think we will continue to see greater adoption in clinical trials and in the future in clinical practice.
"The ability to accurately measure and analyze gait for longer durations in the environment where function matters most has grown exponentially in recent times."
What is so distinct about the algorithm package used to validate gait as a digital endpoint?
Lukas Adamowicz: SciKit Digital Health (SKDH) is an open-source package containing multiple digital health technology algorithms collected under one common framework to provide a full end-to-end digital health processing pipeline for wearable inertial sensor data.
There are very few other open-source packages that cover such a breadth of areas under a single umbrella. This collection of methods enables SKDH to provide digital health measures from various algorithms in the same framework. It aims to provide a common framework to allow all of those different functionalities to work together.
SKDH did several specific things to simplify and promote reproducibility. Can you share?
LA: Our package simplifies the construction of new data processing pipelines and promotes reproducibility by following a “convention over configuration” approach, standardizing most settings on physiologically reasonable defaults in healthy adult populations or those with mild impairment.
There are often 5-15 parameters that affect how the underlying algorithm (e.g. gait, activity) computes the digital health measures. These can range from the ability to change thresholds, to larger sub-algorithm choices such as the gait model to use for estimating step length. Additionally, users have the ability to save pipelines to a YAML file, which saves information about the parameters, and order of processing steps.
How could this be deployed into future clinical trials?
LA: The ability to save and load multiple different types of pipelines for different inertial sensor processing tasks (e.g. wrist and activity processing vs lumbar and gait processing) enables very efficient at-scale deployment, where SKDH can be deployed in a single Docker image that loads a specified pipeline per file-type and runs the required processing.
Since SKDH encompasses end-to-end pipelines, any type of processing required for deployment in clinical trials is run under a single framework, instead of requiring multiple frameworks for different studies or even analysis types within the same study.
"When there is a disease condition for which there is no good established measure to assess efficacy of a treatment, there are higher stakes and the adoption and usage of a digital endpoint may be high."
What digital endpoint adoption do you see that signals we’re moving in the right direction?
XC: The 95th percentile stride velocity was qualified as a primary endpoint by the EMA. Trials investigating delandistrogene moxeparvovec (Elevidys) for DMD used 95th percentile stride velocity as a secondary endpoint; it may be taken with the other supporting evidence of clinical efficacy when it comes to the full approval of this therapy.
Additionally, regulators have signaled their interest and are sharing their scientific advice in the development of these qualification programs to meet unmet needs for many patients. Once qualifications are accepted, this will further accelerate practical adoption and usage.
Where are you seeing digital endpoints making the most impact?
XC: When there is a disease condition for which there is no good established measure to assess efficacy of a treatment, there are higher stakes and the adoption and usage of a digital endpoint may be high.
When there is a disease condition which fluctuates (for instance movement disorders such as Parkinson's Disease) a digital measure with the flexibility to obtain data that captures that variability is also highly sought after.
We have also seen recently that a digital endpoint, such as moderate to vigorous physical activity (MVPA), was more sensitive to change in patients with pulmonary arterial hypertension compared to the six-minute walk distance. The REBUILD trial was able to reduce its sample size by over 50% using MVPA as a primary endpoint as assessed at home compared to the in-clinic six-minute walk test. This shortened development timelines as the study was able to enroll faster.
Is there more cross-industry collaboration to accelerate digital biomarker adoption?
XC: An emerging area of interest is combining big datasets that are obtained in clinical studies from multiple companies and leveraging those outcomes to improve regulatory decisions. We have engaged with the Foundations for the National Institutes of Health and the Innovative Health Initiative programs, bringing public and private partnerships to drive the development of medicines using these digital biomarkers, and conducting analytical and clinical validation studies in different patient populations.
In This Article
