Skip to main content

Sensing and predictive treatment of frailty and associated co-morbidities using advanced personalized patient models and advanced interventions

Periodic Reporting for period 2 - FrailSafe (Sensing and predictive treatment of frailty and associated co-morbidities using advanced personalized patient models and advanced interventions)

Reporting period: 2017-07-01 to 2019-04-30

Frailty is a syndrome of decreased functional reserve and resistance to stressors, leading to an increased risk for adverse health outcomes. However, frailty seems preventable, and may be delayed, or reversed. FrailSafe (FS) aims to better understand frailty and its relation to co-morbidities; to identify quantitative and qualitative measures of frailty through advanced mining approaches on multiparametric data from the physical, cognitive, social, behavioral and functional domains and use them to predict short and long-term outcome and risk of frailty transition; to create frailty prevention evidence-based recommendations; and to achieve all with a safe, unobtrusive and acceptable system following all ethics directives, while reducing health care costs.
FS system includes devices that continuously monitor physiological and behavioral parameters; the FS devices include bio-impedance scales, blood pressure monitors dynamometers, mobilographs, beacons, mobile phones, and a smart vest (EC certified), together with custom-made docking stations for automatic recharging, down/up-loading data.

Overall, 565 older adults from three clinical centers (Greece, France, Cyprus) were enrolled. Comprehensive clinical assessments were carried out and FS devises were used in multi-point sessions, as per protocol. Furthermore, natural language analysis of written text and data from social media were used to detect subtle frailty related cognitive, language and behaviour changes. The FS Games Framework including augmented reality games, used for both frailty assessment and intervention, was developed and tested for usability, safety, and ergonomics. Dynamic adaptability was integrated. Blood samples were collected to determine the telomeres’ length. In a sub-population, the inflammation profile, the autonomic nervous system, and the mid-thigh muscle surface and volume were studied. Phone follow-ups recorded adverse events, including, falls, fractures, hospitalization, institutionalization, and death.

Data were stored in a HBase database (hosted at the Amazon Web Services cloud) . New methods for the offline and online management and analysis of multimodal and advanced technology data were introduced, and a virtual patient model (VPM) was developed. A GPS tracker application for distance measurement and step counting was implemented. The openEHR, a reference model for building the VPM via archetypes was adapted to address issues of human-computer interaction. A Decision Support System (DSS) engine was incorporated in the VPM, to monitor users' vital signs and create alerts for health professionals. A prediction engine of clinical state was developed, by taking into account all physiological measurements using multiple instance learning (MIL) and tensor decomposition techniques for seamless extraction of features and in-depth analysis of the time series data. The activity classification models based on the FS (devices) variables have been enhanced by accounting for inconsistencies. A linguistic analysis tools was also developed to support necessary user actions. Finally, automated personalized guidelines were created targeting frailty and were incorporated in the DSS to be automatically available to users, caregivers and health professionals.

The FrailSafe system, including data from both clinical assessments and devices, was used
to predict frailty related adverse events (hard outcomes) and worsening in different domains (proxy outcomes). A clinical index, a technical FS device index and a combined clinical and FS index were derived and incorporated in the DSS, providing add-on value to the traditional methods to predict outcome. Furthermore, through cluster profiling a qualitative method was identified to visualize frailty phenotype.The FS platform allowed online data analysis, to identify instability, tendency to fall, and loss of orientation. System usability, utility, friendliness and acceptability in real-life settings (users’ house environment) have been thoroughly analysed. The final version of the FS system has been easily accepted by most of the participants after a short adaptation period. FS system (mostly the games) seems to exert a rehabilitation effect. The future impact to health, social and financial domain was also promising.

The project’s website, the extensive use of social media, workshops, conferences, webinars, fairs and exhibitions, alongside with scientific papers and other written and oral contributions were all directed to increase scientific community and general public awareness. EU regulations and directives were followed and a data management plan was submitted including descriptions of the publicly shared data and links to the public data repositories. FS business models and relevant exploitation strategy were also developed including elaborations of the FS competitive advantages, business models to be commercialized and financial issues and plan for the short-term exploitation rollout. Additionally, the Intellectual Property Rights (IPR) Working Group was defined and a Joint Ownership Agreement was signed. Multiple disseminations actions were carried out and the project’s final event managed to bring the FS solutions closer to its potential multi-stakeholders.
The FrailSafe project achieved the following:
1) Novel platform for continuous multidomain monitoring and assessment; 2)New frailty metrics that can better assess frailty; 3) Automated personalized interventions; 4) FS Integrated system adopted technologies and architectural patterns proper to cloud environments; 5) Integration of FS with the International Patient Summary of HL7; 6) New wearable system for medium- to long-term monitoring; 7) New serious dynamic games for assessment and intervention; 8) Multi-Instance Learning (MIL) techniques and deep convolutional neural networks for creating a clinical state prediction engine; 9) Novel Tensor Decomposition techniques combined with Support Vector Machines to extract relevant features and build predictive models of frailty; 10) Non-linear dimensionality reduction techniques combined with unsupervised clustering to create clinical profiles based on FS variables and associate them with standard phenotypes; 11) A novel Decision Support System (DSS) engine designed and developed as part of the risk assessment platform to generate alerts; 12) A human motion identification module that classifies basic activities of daily living, together with a fall detection module; 13) A novel technique for indoor localization using Bluetooth beacons; 14) A novel technique for unobtrusive indoor activity recognition based on WiFi Channel State Information signals; 15) A novel graphical user interface (DSS UI), providing visualizations of the collected data; 16) A novel buffer to dynamically distribute the desired storage space according to an importance-based distribution function, allowing higher quality complexity in more important regions; 17) Τext mining techniques using only a small number of annotated examples, applying active learning; 18) Evidence-based and domain-specific pamphlets were created to tackle all frailty domains; 19) A FS system which is widely acceptable, usable and appealing to multiple stakeholders.

The benefits of FS are illustrated in the figure below.