AI-assisted Continuous Stratification in Neurorehabilitation of Stroke Using Personalized Digital Twins

Overview

The goal of this clinical trial is to learn if a rehabilitation application on a smartphone, an app, can be used by adults who have had a stroke. The main questions it aims to answer are: Are people who have had a stroke able to use the app? Is the app useful for people who have had a stroke? Will the app adapt to the needs of the person recovering from a stroke? Researchers will compare the app to the usual rehabilitation a person receives after a stroke to see if the app can be used as part of a person's rehabilitation. Participants will: Use the app every day for 6 weeks Have an assessment with a rehabilitation research doctor before starting using the app and after completing using the app Keep a diary of the exercises that they do using the app

Outline of research This project is designed to assess the feasibility of the STRATIF-AI rehabilitation application (app), alongside its accessibility, usefulness and personalisation to patients. The purpose of the app is to support patients in their recovery from stroke, however, since it is at an early stage of development, the main aims for this study are testing the feasibility of having such an app involved in a patient's recovery and if patients find it useful. State-of-the-art stratification technology today is based on machine learning (ML) algorithms, trained on large cohort data. This has two main limitations: a) such ML models cannot use all the variety of different data that are generated about a patient, b) stratification is thus only done intermittently, implying outdated and suboptimal care decisions. To remedy this, a new concept and technology - continuous stratification, using the STRATIF-AI platform was developed. In continuous stratification, all data generated about a patient are cumulatively stored in a Personal Data Vault, controlled by the patient. This personal data continuously updates the digital twin platform. The unique potential with the platform comes from the hybrid architecture, combining mechanistic, multiscale, and multi-organ models with ML and bioinformatics. This technology to allows simulation of patient-specific responses to changes in treatment, and identify changes on intracellular, organ, and whole-body levels, ranging from seconds to years for patients following a stroke. Semantic harmonisation is combined with federated learning to securely re-train the various sub-models when new data become available in one of the cohort databases. This project will explore the use of digital twins in the rehabilitation phase of recovery following stroke and will be carried out in the rehabilitation service at Chapel Allerton Hospital in Leeds. It is one of six simultaneous studies involving eight hospitals across Europe with the aim to refine and validate the models and demonstrate how digital twins can follow patients across different apps, covering all phases of stroke: from prevention to acute treatment and rehabilitation. The scalable platform for continuous stratification forms the foundation for a new interconnected and patient-centric healthcare system. This project aims to evaluate the feasibility of utilising a novel app to develop a continuous stratification method, using all available data for real-time patient assessment by consolidating patient data into an evolving digital twin. It is hypothesized that the use of this technology compared to standard post-stroke rehabilitation care will improve patient engagement with post-stroke rehabilitation care and create opportunities for future work in the use of digital twins in healthcare. Background and Rationale Motivation to engage in a rehabilitation programme and concordance with the rehabilitation programme are two major limitations to people's recovery following stroke and which can lead to suboptimal outcomes. A related, and central, problem in all phases of stroke care is that data lie in silos, patients are not provided with their data, and stratification of treatment is made only intermittently. For these reasons, the STRATIF-AI platform, which is based on digital twins - a digital copy of a patient - which facilitates continuous stratification, within and across all phases of stroke care was created. Different studies run by partner hospitals across the consortium will focus on the earlier stages of stroke care such as preventative measures and the acute monitoring of patients. In these studies, data will be collected to allow for the continuous models to be developed. In this clinical study, how patients and medical staff experience the usage of the new platform in rehabilitation will be studied. This is a pilot study, which will examine proof-of-principle, and collect initial feasibility and acceptability data. Worldwide, stroke is the fourth most common cause of disability for adults with 11.9 million people experiencing strokes in 2021, with the latest statistics demonstrating an increase in stroke incidence from 1990-2021. The global economic impact of stroke is over US$890 billion per year, a figure that is expected to double by 2050. Aside from the economic implications, the effect of stroke on quality of life of the affected individual is substantial, with over two thirds of stroke survivors in the UK leaving hospital with a disability that requires ongoing care needs. Whilst more work is needed in the field of stroke prevention and acute management, increasingly the importance of the rehabilitation phase following stroke is recognised internationally by the World Health Organization who acknowledge the "profound, unmet need" for rehabilitation globally, as well as nationally in the most recent UK Guidance for Stroke Rehabilitation, which outlines the crucial role of the rehabilitation phase of stroke, and the increased quantity of rehabilitation required to maximise recovery. Originally developed for industry, and a relatively new concept in healthcare, a digital twin can be defined as a virtual counterpart of a physical object, system, or process that mirrors its real-world equivalent in both function and appearance. The ongoing progression of digital twin technology has led to the evolution of advanced models that continuously adapt over time to be able to simulate complex behaviours. Digital twins differ from traditional digital modelling, as they require two-way data exchange between the physical entity and the digital entity, thus enabling the physical entity to receive feedback from its digital counterpart which facilitates continuous stratification of data. Given the move towards a personalised approach to treatment in healthcare, the potential for digital twins is being increasingly explored. However, the complexities of human emotion and behaviour, along with the variability of factors such as individual preference, motivation, fatigue and general health require sophisticated digital twins for use in healthcare. A promising area for the use of digital twins is in the field of stroke rehabilitation, where they could be used to support the period of rehabilitation that is often required by stroke survivors by modelling individual recovery patterns and optimizing therapies. By integrating real-time patient data, such as movement patterns, communication and cognitive ability, digital twins can simulate rehabilitation outcomes to tailor treatments to suit individual patients, adjusting therapy in real time, and offer decision support and motivational feedback, potentially increasing patient engagement and enhancing recovery. The concept of digital twins has already been used in human movement analysis through commercially available platforms such as OpenSim, which has been validated for musculoskeletal modelling, demonstrating the feasibility of digital twins in healthcare. The STRATIF-AI project is a Horizon Europe funded project that aims to channel digital twins for use in rehabilitation following stroke. Traditional patient stratification in rehabilitation relies on periodic goal setting carried out by the multidisciplinary team, agreeing rehabilitation goals every two to three weeks with patients and their family members or carers to direct the activities of the rehabilitation programme. However, this approach can miss valuable data between episodes of goal setting, meaning that the most recent goals do not accurately and completely reflect a patient's current status. This loss of data between goal setting episodes can be limiting in current healthcare practices, because treatments remain unchanged until the next formal interaction between the patient and healthcare professionals, even if a patient's condition has evolved. For this reason, STRATIF-AI proposes to develop a continuous stratification method to utilise all available data for real-time patient assessment by consolidating these data into an evolving digital twin, which will then offer real-time feedback to the patient and healthcare team. Data traditionally used in stratification in stroke care such as the patient's diagnosis, risk factors for stroke, past medical history, baseline observations, and scan and blood test results, will be complemented by interim information from wearable technology (off-the-shelf smart watch), patient diaries, and the formalised rehabilitation assessment tool the Functional Independence Measure and Functional Assessment Measure (FIM+FAM). These data will be used to update an individualised digital twin for the patient using hybrid technology, which integrates mechanistic models for the main functions of the human body with machine learning and bioinformatics models. The STRATIF-AI project will implement continuous stratification across the rehabilitation phase of stroke for eligible patients across Leeds Teaching Hospitals NHS Trust, and the results of the project will complement ongoing studies into the use of the platform during the prevention and acute treatment phases of stroke carried out in other centres.