PRECISE4Q

Create a harmonized data infrastructure in order to normalize and precisely characterise data from heterogeneous and longitudinal sources in different formats, languages and degrees of structure.

The first objective aimed to provide the data infrastructure as a prerequisite for modelling. To this end, a data warehouse that is capable of hosting and providing the data used in the study was created. The data warehouse was built after identification of all potential data source characteristics, user needs and data exchange needs. It adheres to all legal and technological standards allowing state-of-the-art date storage and exchange, a crucial prerequisite for the project. Furthermore, data was harmonized to allow pooling of similar data in heterogeneous data sources. Data harmonization efforts revolved around collecting requirements for data integration and modelling, constructing a thesaurus for the languages of the used data sources and, consequently, a common ontology-based data model. Next to this, natural language processing was used to structure clinical texts. Data harmonization and the development of models represented an iterative process organised in release cycles. To allow early development of models, the first release cycle of harmonized data comprised of structured data, whereas a later release provided the unstructured data that took more time to extract.

Identify health factors, risk factors, resilience factors and life events in stroke affecting well-being (integrated quality-of-life-concept).

This objective was highly cross-disciplinary and had to be performed prior to modelling, but iteratively in parallel to Objective 1. It entailed defining the inputs and outputs for the modelling phase. First, the most likely scenarios and use cases for the clinical phases of stroke were defined. Then, we identified rank and weighed input factors of stroke available in our databases in terms of both risk and resilience. The activity consisted of iterations of dialogue between clinicians and modellers. It also entailed experimentation with the data to identify correlations between factors and patient well-being and the use of techniques such as Latent Class Analysis and clustering to identify if masked subgroups (such as early or nonresponders to certain treatments) were present in the harmonized datasets. Defining the outputs for the models, it was crucial to understand how the model output would directly link to interventions in patient treatment and how the model would fit into the treatment processes within the care setting. Most of our models provided complex outputs. For some models, this added complexity in output was necessary because they explicitly take time scales into account. For other models, this added complexity was required to reflect a broader understanding of the factors contributing to resilience and well-being. A central motive of our study was to use a quality-of-life framework as a proxy for resilience and wellbeing. Our models predicted a set of interrelated factors that reflect the variety and evolving factors contributing to an individual’s quality of life. Thus, in this objective the outputs variables were identified, weighted and integrated in a quality-of-life framework.

Develop a set of predictive models that target the four different stages of a stroke patient’s trajectory (prevention, stroke treatment, stroke rehabilitation, stroke reintegration)

This objective was our core objective as it encompasses the development of the predictive models for each of the four clinical phases of stroke, i.e. prevention, treatment, rehabilitation and reintegration. Our models were based on the harmonized data sources from objective 1 and the input/output definitions from objective 2. A major focus of this objective was hybrid modelling for the first two clinical stages of stroke. Hybrid modelling is the combined use of mechanistic and phenomenological models. Hybrid modelling combines the best of two worlds: i) the simulation-based possibilities of mechanistic models with ii) the predictive versatility of machine-learning models. In the stroke prevention stage, the PRECISE4Q model predicts the evolution of key patient features over in response to different treatments, such as diet, exercise, and medication in a hybrid modelling approach. It predicts the overall risk of stroke for a patient participating in a treatment plan. Using the outputs of this model a healthcare professional is able to develop and compare personalised intervention regimes for their patients to increase the patient’s resilience to stroke. In a hybrid modelling approach the outputs of an existing biophysiological mechanistic imaging-based model of brain blood flow and perfusion is fused with patient-specific clinical data to predict a modified Rankin Scale for the patient 3 months post-stroke. The second model integrates a range of heterogeneous patient-specific data (e.g. lab, OMICs etc.), information extracted from the textual records, and the modified Rankin Scale estimated by the other model. The output of this second model is a complex quality of life (QoL) profile for the patient, including QoL indicators such the probability of the patient returning to work. For the rehabilitation stage PRECISE4Q creates a model that generates a patient specific daily schedule of rehabilitation activities annotated with the expected performance level of the activity along with the expected cognitive and functional profile of the patient at the end of the rehabilitation programme. The output of the model supports doctors in designing personalised rehabilitation plans for each patient. The PRECISE4Q model supporting patient treatment during reintegration phase predicts a long-term (1-year post-stroke) complex QoL profile for a patient. The output of the model informs the design of individualised support plans to long-term resilience and quality of life. 

Gather in prospective clinical studies unique personalised multi-dimensional longitudinal datasets reflecting patient’s trajectory for validation.

This objective aimed to collect high-quality patient data (n=1200) from two prospective studies. The first prospective study targeted stage 2, acute stroke. The other prospective study targeted stage 3 and 4, stroke rehabilitation and reintegration. For both studies, the definition of study parameters were performed according to the defined inputs and outputs in objective 2. This ensured the seamless merging of study data with the project data warehouse and data harmonisation. Additionally, for both studies a patient-reported outcome concept was developed. At the time, the gathering of  data was researcher-centred as data are collected topdown by the performing clinical staff. This is often inefficient and does not consider patient motivation leading to high rates of “lost-to-follow-up”. To facilitate the gathering of high quality data, end-user-based technologies were employed to obtain patient-centered data. To this end an electronic case report form (eCRF) was developed allowing the fine-meshed recording of patient-reported outcome by research staff and by the patients themselves. The result was an electronic patient recorded outcome (ePRO).

Validate the predictive models using data from prospective studies and to integrate the final models in a comprehensive platform.

The clinical data from the studies conducted was then used to validate models for stages 2, 3 and 4 that were developed using retrospective registry data in objective 3. Once the final models were established, they were integrated in a one-stop-shop web-based platform, coined the “Digital Stroke Patient Platform”. This platform is an open service allowing researchers to test and apply our models on their own data.

Develop personalised and stagespecific interventions to increase resilience to stroke and increase well-being and reintegration after stroke.

The “Digital Stroke Patient Platform” was used to simulate patient courses. Additionally, the weighting of the input factors from the models was taken into account. In this first round, possible factors that can be the target for interventions were identified. Finally, these factors were assessed for clinical feasibility. The final output of this objective was a set of clinically feasible interventions for each stage of stroke.

Develop a medical decision support system for stroke with user interfaces.

To enable the future validation of suggested interventions a set of dedicated software interfaces was provided. These were based on the “Digital Stroke Patient Platform”. However, since the interventions were tailored for each of the stroke phases, a unique clinical decision support system for each of the four phases was developed. This allowed us to design the interfaces of the clinical decision support systems in a way that fits the unique healthcare setting of each phase. In order for the software to be usable in all clinical settings, the decision support systems have also been made available as stand-alone versions.

Establish the European Modelling Platform for Open Stroke Research (EUROPE-Stroke), an open research platform for data aggregation, integration, analysis and publication to promote science.

This objective planned and designed an Open Innovation organisation or virtual institute for stroke prevention and treatment. The aim was to fully exploit the PRECISE4Q data eco system integration and application scenarios, to support hospitals and physicians in stroke modelling and data-guided treatment and rehabilitation planning through various measures. These include educational tools and programmes, support for in-hospital machine learning process optimisation and a community platform for peer-to-peer exchange accessible to stroke specialists.

Identify and pursue commercial exploitation of scientific results by using generated knowledge and intellectual property.

PRECISE4Q not only offers a cost-effective Digital Stroke Patient Platform, but also provides the methods, social science and economic expertise to formatively assess clinical effectiveness. Furthermore, it offers lifestyle management integration, efficacy, and socio-economic cost-benefit ratio as a quantitative foundation for initial but solid service, development and commercialisation strategies, plans for scaling up, as well as actual diffusion of the innovation capacity. Coupled with clinical assessments, inputs from validation  paired with a deployment scenario based cost-benefit analysis covering a wide range of stakeholders and potential end-users, this objective ensures the scalability and sustainability of PRECISE4Q working under real conditions, in real healthcare and social care settings. The project objectives are fully measurable, safeguarded through a clear innovation management work plan and had been designed in line with targeted exploitation plans, ensuring a realistic and sustainable impact of the project.