Learning Human Health and Diseases from 24-hour Wrist Movement
AuthorsYong Wang, Dylan McGagh, Katya Broomberg, Zizheng Zhang, Jonathan Carter, Junayed Naushad, Laura Brocklebank, Yang Sun, George Nicholson, Dianjianyi Sun, Canqing Yu, Jun Lv, Maxim Barnard, Hubert Lam, Andrew Steptoe, David W. Eyre, Liming Li, Zhengming Chen, Naomi Wray, Spiros Denaxas, Gary S. Collins, Huaidong Du, Aiden Doherty, Hang Yuan
Resources
Sensori turns a day of ordinary wrist movement into a reusable health representation that can improve prediction of many diseases without requiring clinical visits or retraining.
Key results
Participants across four population-based cohorts used for pretraining and evaluation.
Person-days of free-living wrist-accelerometer recordings.
Parameter count of the selected model variant.
Eligible conditions with significant AUROC improvement after adding Sensori.
Median improvement over clinical covariates.
Eligible conditions with significant six-year Uno’s C-index improvement.
What the paper found
Sensori is a self-supervised foundation model that learns health representations directly from 24-hour raw tri-axial wrist acceleration, rather than relying on predefined features such as steps or sleep duration. Its wav2vec 2.0-inspired architecture uses convolutional encoding, five-minute pooling, a transformer, masked reconstruction, and day-level contrastive learning to convert 10-Hz recordings into 768-dimensional daily embeddings. Training and evaluation covered 122,640 participants and 683,617 person-days across UK Biobank, China Kadoorie Biobank, ELSA, and NHANES, with external testing on PAMAP2, RealWorld, WISDM, and CAPTURE-24. The selected model used 40M parameters and generalized across populations without retraining; its activity embeddings achieved mean Cohen’s kappa of 0.852 on PAMAP2 and 0.824 on RealWorld. When added to clinical covariates, Sensori significantly improved prevalent-disease classification for 52 of 102 conditions, with a median AUROC gain of 0.060, and improved six-year incident-risk prediction for 26 of 87 conditions, with a median Uno’s C-index gain of 0.064. The largest benefits involved Parkinson’s disease, multiple sclerosis, essential tremor, depression, and other neurological or psychiatric disorders. The approach could complement products such as Fitbit Coach and ChatGPT Health by supplying richer raw-sensor representations to wearable health systems.
Original abstract
Much of human health and function unfolds beyond the clinic, through the movements of everyday life. Wrist-worn accelerometers capture these movements continuously, yet their rich signals are often reduced to a small set of predefined behavioural summary measures. Here, we present Sensori, a self-supervised foundation model that learns general-purpose health representations directly from 24 hours of raw tri-axial wrist movement. We developed and evaluated the model across four population-based cohorts from the United Kingdom, China and the United States, comprising 122,640 participants contributing 683,617 person-days of free-living recordings. Sensori condensed each day of movement into a representation that captured diverse movement behaviours, demographic characteristics, health axes and physical function. Evaluation in independent cohorts showed that these representations generalised across populations and measurement settings without retraining. When added to common clinical covariates, Sensori significantly improved prevalent disease classification for 52 of 102 eligible conditions (median delta AUROC, 0.060; range, 0.012-0.242) and incident disease risk prediction for 26 of 87 eligible conditions (median delta Uno's C-index, 0.064; range, 0.025-0.172), with the largest gains for neurological and psychiatric disorders. These findings establish 24-hour wrist movement as a rich and scalable source of health information, with the potential to support passive health monitoring and disease prediction at population scale.
Read the original paperMore in Foundation Models
Browse all 47 papers →How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text
Jenna Russell, Ben Glickenhaus, Katherine Thai, John Wieting, Mohit Iyyer, Max Spero, Bradley Emi
AI-generated web text can help language models at first, but beyond a tipping point it degrades performance on human writing, making data filtering and separate evaluation increasingly important.
TabFM: A Zero-Shot Foundation Model for Tabular Data
Weihao Kong, Erez Louidor Ilan, Shuxin Nie, Taman Narayan, Rajat Sen, Yichen Zhou, Deqing Fu, Samet Oymak, Abhimanyu Das
TabFM is a large synthetic-data-trained model that aims to make accurate tabular predictions instantly, without retraining for each new dataset.
When Do Biological Reasoning Models Use Their Biological Inputs?
Ada Fang, Nikitha Thoduguli, Lukas Fesser, Hanlin Zhang, Sham M. Kakade, Marinka Zitnik
The study finds that many biological reasoning systems appear to succeed without meaningfully using the biological inputs they were designed to reason over.