A machine learning pipeline named ActiTect identified isolated REM sleep behavior disorder (iRBD) from wrist actigraphy recordings alone, reaching areas under the curve (AUROC) of 0.84 to 0.95 across four cohorts in Germany, Britain and Denmark, according to a study published in npj Digital Medicine on June 13, 2026. The tool, built by a team including David Bertram and senior author Katarzyna Bozek, was designed to work without the overnight, technician-scored polysomnography currently required to diagnose iRBD, a sleep disorder in which people physically act out dreams. iRBD often precedes Parkinson's disease, dementia with Lewy bodies or multiple system atrophy by years, and the paper describes overnight sleep-lab testing as a bottleneck that limits how many people can be screened.
The researchers developed ActiTect on wrist recordings from 78 people, 55 with iRBD and 23 without, enrolled in the German CogTrAiL-RBD trial, then tested it on a separate blinded set of 31 people at the same site, where it scored an AUROC of 0.86. Two outside cohorts served as external validation: 113 actigraphy recordings from 103 people in the Oxford Discovery cohort at the Oxford Parkinson's Disease Centre, where it scored an AUROC of 0.84, and 57 recordings from 31 people in the PACE cohort at Denmark's Lundbeck Foundation Parkinson's Disease Research Center, where it scored an AUROC of 0.94. Repeated cross-validation that held out one whole dataset at a time produced AUROC values of 0.84 to 0.89 throughout, which the authors describe as evidence the model's performance held up across sites and devices.