The translational cluster (WP2, 9, 10) has concluded research determining the acceptability and potential barriers for RMT in people with MS, epilepsy and MDD. A systematic review summarising previous findings on barriers and facilitators to RMT across a breadth of health conditions has been published. Focus groups in the three disorders and three countries (UK, Italy and Spain) have been completed and analysed by WP2, with results now published. Detailed care pathway mapping and surveys have been completed by WP9 identifying the likely points at which RMT will have greatest clinical utility. WP10 has mapped out the current regulatory climate for the application of RMT in healthcare. We have presented our programme at the European Medicines Agency (EMA) and at a seminar at the Foundation of National Institutes of Health where members of the US Food and Drug Administration and EMA were present.
The technical work packages (WP7, 8) have delivered RADAR-base, a platform for the storage and analysis of data from RMT, and developing an analytic framework for future studies. The platform has been successfully deployed first in an in-hospital study of WP4 (epilepsy, study 1). Since November 2017 it has been deployed in the first ambulatory community study of depression and subsequently in MS and most recently epilepsy. All clinical studies have now completed recruitment including a total of 623 patients in our depression study, 537 in MS and 243 in epilepsy. RADAR-CNS has accumulated more patient years exposure of sensor data than virtually any similar study internationally.
The main purpose of our clinical studies was firstly to demonstrate the acceptability of these technologies in real world settings, including recruitment, retention, data completion and patient experience. We have published extensively on this topic demonstrating that with the infrastructure we provided within RADAR-CNS consortium it is possible to gain far more complete and long-term data than previous research would have suggested was possible. The greatest challenge we identified is in data completeness as there are multiple steps required to ensure data completeness, including (a) that the smartphone/wearable is working (charged, functioning properly), (b) that participants are interacting with the technology appropriately (e.g. wearing the wearable, completing questionnaires etc) (c) that there are no technical issues beyond the control of the study team (e.g. changes to operating systems of permissions allowed by Android for data extraction from smartphones). These challenges are integral to conducting distributed studies, and therefore it is necessary to develop frameworks for the adequate description of incomplete data and to ensure that data missingness does not lead to systematic biases. We have explored these issues, for example showing the depression status does not strongly predict data incompleteness.
The second purpose of the clinical studies was to determine whether signals from smartphones or wearables could be used to gain an better indication of disease status. Here we have shown through multiple publications in WP4 that wearables can provide adequate and acceptable indications of seizure activity. In WP5 upcoming publications will demonstrate that the signals from wearables are able to detect “silent progression” in multiple sclerosis – i.e. gradual worsening in functioning which is of such insidious onset that patients and clinicians are unable to detect changes easily. This has major implications for clinical practice, offering the potential for us to intervene earlier in those who require it. In depression we have demonstrated that numerous data streams including GPS, Bluetooth handshakes and sleep, are associated with depressive symptomatology.
Finally, we have preliminary evidence that clinical state (e.g. depressive relapse) can be predicted by data signals collected from smartphones. Using the participant’s own data when they are in a state of health, we were able to train models to relapse. Because they are individualised, these models take account of the wide inter-individual variation in parameters such as GPS and Bluetooth handshake data, detecting departures from the individuals’ normal daily patterns, which appear to predate relapse. Such signals from smartphones could be used in a highly economical manner to trigger actions by the healthcare provider (e.g. further assessments, preventive lifestyle advice) which would meet our goal of moving healthcare from a model of “diagnose and treat” to one of “predict and prevent”. These models will be validated and refined over the next 3 months with submissions to leading journals in the latter half of 2022.
