A1 · Overview
A research group turning consumer wearable signals into clinically tractable phenotypes.
2 active · 1 in preparation
Tam³ is a UK-based research group working at the intersection of consumer wearables, clinical
records and population-scale health data. Our work centres on a single question: which
sensor-derived features carry information that matters for human health, and how reliably
can they be measured across the device ecosystem people actually wear.
The output is an
engineered feature library that moves from
PPG, ECG and accelerometry sensors to clinically meaningful phenotypes.
CONTACT · research@tam3.co.uk
A2 · Programme
Three studies, one feature stack
Each study contributes a layer that the next builds on. The same engineered features run
end-to-end, from raw device signal to clinically meaningful phenotype.
01 / 03
Wearable–EHR cohort
Linking continuous sensor streams to clinical records
A cohort of approximately 2,500 individuals' wearable streams (heart rate, accelerometry, sleep architecture) and structured electronic health records. The cohort supports concurrent EHR-side work on automated extraction from unstructured clinical notes.
The cohort is powered to test whether composite features engineered from wearable streams, capturing cardiovascular fitness, activity volume, sleep regularity and recovery, carry information about underlying clinical phenotype that is incremental to standard anthropometric and biomarker measures. A central hypothesis is that combining wearable-derived features with structured EHR signal improves the identification of higher-risk phenotypes relative to either modality alone, surfacing high-risk profiles that routine records may not flag in isolation.
02 / 03
Device normalisation
How far are consumer wearables from clinical ground truth?
A paired co-wear protocol comparing six consumer wearable vendors against simultaneously recorded clinical-grade reference monitors. Participants wear devices across resting, active and sleep regimes.
The protocol is designed to quantify how far vendor-supplied summary metrics from consumer devices depart from clinical reference across regimes, and to characterise how any bias varies by device class and by physiological state. Where such bias is present, features built from raw streams will not generalise across the device ecosystem without calibration.
03 / 03
Phenotype overlay
How much wearable data is enough to predict clinical phenotype?
A planned analysis that would apply the feature library from Studies 01 and 02 to research-grade accelerometry and linked clinical records from an established population research resource, subject to data access approval. It asks which engineered features carry independent and additive association with incident clinical phenotypes, beyond standard anthropometric and biomarker measures, and what wear-time and data-quality thresholds those associations require.
Findings become a feature catalogue specifying which metrics to extract, how to normalise them and what wear-time coverage is sufficient, a standard for validating data across the consumer device ecosystem.
A3 · People
Investigators
James Hale
PRINCIPAL INVESTIGATOR
University College London
Wearable biosignals, signal processing for ambulatory sensors, federated learning. Lead methodologist on the device normalisation protocol and the engineered feature library.
James Andrew Butler
CO-INVESTIGATOR
University of Oxford
Machine learning, agent-based modelling and quantitative systems pharmacology, with prior work in the RA-MAP consortium on multimodal characterisation and longitudinal disease trajectories in early rheumatoid arthritis.
↗ Google Scholar
Harry Duckworth
CO-INVESTIGATOR
Imperial College London
Computational modelling of brain biomechanics and cerebral vascular injury, with doctoral work at Imperial on finite-element and smoothed-particle-hydrodynamic simulation of traumatic brain injury. Leads the EHR side of the linked cohort and the consent and governance framework.
↗ Google Scholar
Paul du Long
RESEARCH SOFTWARE ENGINEER
Independent researcher
Time-series machine learning, accelerometry pipelines, reproducible research infrastructure. Maintains the group's feature library, analysis stack and open-source releases.
A4 · Outputs
Selected prior work
Prior work from across the group in biomechanical modelling, immunological phenotyping and systems pharmacology, establishing the multimodal-data, simulation and consortium methods the wearable programme now builds on.
2022
Frontiers in Bioengineering and Biotechnology
Published
2022
Scientific Data
Published
2021
Int. J. Numer. Methods Biomed. Eng.
Published
2019
Journal of Biomechanics
Published
2018
Nature Reviews Rheumatology
Published
2015
CPT: Pharmacometrics & Systems Pharmacology
Published