Presentation at the Artid workshop, U. Bristol, March 2024, on digital biomarkers for improved clinical trials and monitoring of complex diseases, including neurological & movement disorders.
2. The obligatory background &
disclaimer
Who am I?
● Once an immunologist
● Then a molecular evolutionist, modeller, bioinformatician,
epi-informatician, analyst, data scientist, “computer guy” ...
● Now solve clinical & trial problems for big pharma using data,
stats & AIML
Nothing in this presentation represents projects or policy at GSK
There are no conflicts of interest
3. What is a digital biomarker?
The broad (and oft ignored) definition
Anything measured and recorded
digitally, even:
● Blood pressure
● Height
● Heart rate
● Online survey
● Phone apps ...
The narrow (and stereotypical) definition
Digital recording of complex patient
behaviour and physiology, followed by
processing to produce quantifiable
metrics or endpoints, perhaps via a
device or sensor carried by the patient:
● Using a wearable to track a
patient’s heart rate over time
● Videoing a patient walking and
modelling it to produce a quality
of locomotion
● Sound analysis of voice or a
cough, to detect obstruction
4. Example: Parkinson's Disease 1
Deng, K., Li, Y., Zhang, H. et al.
Heterogeneous digital biomarker
integration out-performs patient
self-reports in predicting Parkinson’s
disease. Commun Biol 5, 58 (2022).
https://doi.org/10.1038/s42003-022-03
002-x
5. Example: Parkinson's Disease 2
Interpretable Video-Based Tracking and
Quantification of Parkinsonism Clinical Motor
States
Daniel Deng, Jill L. Ostrem, Vy Nguyen, Daniel
D. Cummins, Julia Sun, Anupam Pathak, Simon
Little, Reza Abbasi-Asl
medRxiv 2023.11.04.23298083; doi:
https://doi.org/10.1101/2023.11.04.23298083
6. Why use digital biomarkers?
● Fidelity: avoids transcription errors
● Memory: not reliant upon patient recall
● Consistency: across sites, investigators, geographies
● Authenticity: measure in a real-world context, away from
clinical visits
● Patient-centricity
● Temporality: measures across time, perhaps to assess
“average” behaviour or look for rare events
● Safety: monitoring patients for AEs
● Simplicity: distil complex behaviour and physiology to
something simpler and measurable
7. Why not use digital biomarkers
● Does it actually work for the patients
○ Some patients like going to the clinic
○ Also additional medical support
● Hardware & infrastructure & support
● Privacy
● Standardization
● Abstract, intangible measures
● Machine learning magic
8. Validity
Analytical
● Does it take good
measurements?
● Does it measure something
robustly and reproducibly?
● Can it cope with patchy or
imperfect data?
● Do small changes in signal or
source only result in
Clinical
● Does this measure mean
something?
● Does a change in the digital
measure relate to a change in
disease state?
● A form of experimenters
regress?
9. Summary
● Digital biomarkers are any persistent digital
measurement of patients
● Such biomarkers can be used reduce complex behaviours
(e.g. movement) to simpler and more quantifiable
endpoints
● But to do this we need hardware solutions and good data
analytics