Digital Patient: Personalized and Translational Data Management through the MyHealthAvatar EU Project

Abstract

The advancements in healthcare practice have brought to the fore the need for flexible access to health-related information and created an ever-growing demand for the design and the development of data management infrastructures for translational and personalized medicine. In this paper, we present the data management solution implemented for the MyHealthAvatar EU research project, a project that attempts to create a digital representation of a patient’s health status. The platform is capable of aggregating several knowledge sources relevant for the provision of individualized personal services. To this end, state of the art technologies are exploited, such as ontologies to model all available information, semantic integration to enable data and query translation and a variety of linking services to allow connecting to external sources. All original information is stored in a NoSQL database for reasons of efficiency and fault tolerance. Then it is semantically uplifted through a semantic warehouse which enables efficient access to it. All different technologies are combined to create a novel web-based platform allowing seamless user interaction through APIs that support personalized, granular and secure access to the relevant information.

Full text

 Abstract— The advancements in healthcare practice have
brought to the fore the need for flexible access to health-related
information and created an ever-growing demand for the
design and the development of data management
infrastructures for translational and personalized medicine. In
this paper, we present the data management solution
implemented for the MyHealthAvatar EU research project, a
project that attempts to create a digital representation of a
patient’s health status. The platform is capable of aggregating
several knowledge sources relevant for the provision of
individualized personal services. To this end, state of the art
technologies are exploited, such as ontologies to model all
available information, semantic integration to enable data and
query translation and a variety of linking services to allow
connecting to external sources. All original information is
stored in a NoSQL database for reasons of efficiency and fault
tolerance. Then it is semantically uplifted through a semantic
warehouse which enables efficient access to it. All different
technologies are combined to create a novel web-based
platform allowing seamless user interaction through APIs that
support personalized, granular and secure access to the
relevant information.
I. INTRODUCTION
A recent report by the eHealth Task Force entitled
“Redesigning health in Europe for 2020” [1] focuses on how
to achieve a vision of affordable, less intrusive and more
personalized care, ultimately, increasing the quality of life as
well as lowering mortality. Such a vision depends on the
application of ICT and the use of data and requires a radical
redesign of health to meet these challenges. A main driver for
change is currently taking place under the term “liberate the
data”. The secondary use of care data for research, quality
assurance and patient safety is still rarely supported and the
main barriers to this are the lack of interoperability, common
standards and terminologies [2]. Large amounts of data
currently sit in silos within health and social care systems. If
these data are integrated and used effectively they could
transform the way that care is provided.
The MyHealthAvatar (MHA) EU project [3] is an attempt
for the digital representation of patient health status. The goal
*This work was partially support by the iManageCancer (H2020-
643529) and the MyHealthAvatar (FP7-600929) EU projects.
Haridimos Kondylakis, Emmanouil G. Spanakis, Stelios Sfakianakis,
Vangelis Sakkalis, Kostas Marias are with Institute of Computer Science
(ICS), Foundation for Research and Technology - Hellas (FORTH), (e-mail:
{kondylak, spanakis, ssfak, sakkalis, kmarias} @ics.forth.gr).
Manolis Tsiknakis is with the Institute of Computer Science (ICS),
Foundation for Research and Technology - Hellas (FORTH) and the
Technical University of Crete (e-mail: tsiknaki@ics.forth.gr).
Xia.Zhao, Hong Qing Yu and Feng Dong are with the Department of
Computer Science and Technology, University of Bedfordshire, Luton, UK
(e-mail: {xia.zhao, hongqing.yu, feng.dong}@beds.ac.uk).
is to create a “digital avatar”, i.e. a graphical
representation/manifestation of the user, acting as a mediator
between the end-users and health related data collections. It is
designed as a lifetime companion for individual citizens that
will facilitate the collection, the access and the sustainability
of health status information over the long-term. Among
others, key questions that should be answered in this context
is how to develop optimal frameworks for large-scale data-
sharing, how to exploit and curate data from various
Electronic and Patient Health Records, assembling them into
ontological descriptions relevant to the practice of systems
medicine and how to manage the problems of large scale
medical data.
In this paper, we attempt to provide answers to the
aforementioned questions by presenting a novel data
management infrastructure. This infrastructure is capable of
combining large-scale and multidimensional data that are
semantically enriched and intrgrated to be further used for a
variety of diverse use-cases. More specifically our
contributions are the following:
 A modular ontology named MHA Semantic Core Ontology
capable of modeling all health-related information.
 A scalable NoSQL Data Repository for storing all original
information received from external sources.
 A novel Data Translation Module that uses mappings to
semantically uplift and translate the data to be stored in a
central Semantic Warehouse. These data can come either
from the NoSQL data repository or from other external
sources that are linked through mappings.
 A variety of Linking Services to external sources to enable
interaction with them. Although these sources do not
support direct linking through mappings, they provide
standard interfaces for exporting data.
 A wide range of programmatic interfaces (APIs) allowing
the granular and secure access to all relevant information
in the platform.
The remaining of this paper is structured as follows: In
Section 2 we present shortly the use-cases that the platform
should cover and the different types of data that should be
used. Then in Section 3 we demonstrate the building blocks
of our data management solution. Section 4 reviews other
related projects with similar goals and Section 5 summarizes
and presents an outlook for further work.
II. USE-CASE REQUIREMENTS
In MHA two general categories of scenarios are
investigated: a) system use cases, describing the
Digital Patient: Personalized and Translational Data Management
through the MyHealthAvatar EU Project*
Haridimos Kondylakis, Emmanouil G. Spanakis, Stelios Sfakianakis, Vangelis Sakkalis, Manolis
Tsiknakis, Kostas Marias, Xia Zhao, Hong Qing Yu, Feng Dong

functionalities of the MHA system from the perspectives of
both clinicians and citizens/patients and b) clinical use cases
describing how to use the data from the MHA system in real
clinical scenarios. After performing an extended requirement
analysis, four distinct and diverse clinical use cases were
selected for further implementation, demonstration and
evaluation: Diabetes, Nephroblastoma Simulation Model and
Clinical Trial, Personalized Congestive Heart Failure (CHF)
risk analysis and Osteoarthritis. From the above we present
important functional requirement of the last two use-cases:
Personalized CHF risk analysis:
 Assist individualized self-monitoring of patient’s own
health-status through a “CHF Real-time patient
monitoring” and a “CHF Risk Assessment” service.
 Provide risk analysis for personal risk monitoring for
developing a cardiovascular related episode in the
future.
 Provide comorbidities and drug interaction
information in both the treating physicians, but also
the patient him/herself regarding negative drug
interactions.
 Create a monitoring tool for Personalized CHF risk
assessment using medical sensors together with
mobile application and MHA’s schematics layer.
 Link, through MHA, with external clinical
information systems to acquire specific EHR patient
related data.
 Incorporate verified risk assessment models for
CHF.
 Create individualized mobile apps for easy access to
the service and MHA platform.
Osteoarthritis:
 Visual analytics should be used to display aggregated
lifestyle data aiming to easy interpretation by both
citizens (patients and healthy) and medical
professionals.
 Data collection methods to easily upload health data.
 Personal Diary managing patients/citizens’ health
status and behaviors, including diet, movement,
environment, mood, smoking, symptoms etc.
 Guided interventions for patients/citizens.
 Provide the means to be able to also incorporate
genomic predisposition evaluation for estimating the
risk of developing osteoarthritis.
To support all aforementioned requirements an advanced
data management infrastructure is required to enable real-
time analysis of big data and the interconnection of all
heterogeneous available information
III. CONCEPTUAL AND TECHNICAL ARCHITECTURE
The conceptual architecture of the data management
platform is shown in Figure 1. In the bottom layer, external
sources are pushing data to the original data repository by
using a variety of linking services. In addition, there are
sources that allow access to the available information (such
as the Linked Life Data
1
or the DrugBank
2
) directly from the
semantic integration module. The data are semantically
linked and integrated using the aforementioned module and
stored as triples at the Semantic Data Warehouse to be
served. On top of these repositories various APIs allow the
granular and secure access to the available data either directly
from the original data repository or from the semantically
integrated data warehouse. In the following sections we
present in detail each one of the aforementioned components.
Figure 1. The architecture of the Data Management Approach
A. MHA Semantic Core Ontology
The MHA Semantic Core Ontology [4] is used as the
virtual schema of all data stored within MHA. It is able to
semantically describe the different types of data required and
processed by the platform.
Figure 2. The modules of MHA Semantic Core Ontology
3
1
http://linkedlifedata.com/
2
http://www.drugbank.ca/
3
ACGT: ACGT Master Ontology, BFO: Basic Formal Ontology,
CHEBI: Chemical Entities of Biological Interest, CIDOC-CRM: CIDOC
Conceptual Reference Model, CTO: Clinical Trial Ontology, DO: Human
Disease Ontology, DTO: Disease Treatment Ontology, FHHO: Family
Health History Ontology, FMA: Foundation Model of Anatomy, FOAF:
Friend of a Friend Ontology, GALEN: Galen Ontology, GO: Gene
Ontology, GRO: Gene Regulation Ontology, IAO: Information Artifact
Ontology, ICD: International Classification of Diseases, ICO: Informed
Consent Ontology, LOINC: Logical Observation Identifier Names and
Codes, MESH: Medical Subject Headings, NCI-T: NCI theraurus, NIFSTD:
Neuroscience Information Framework Standardized ontology, NNEW: New
Weather Ontology, OBI: Ontology for Biomedical Investigation, OCRE:
Ontology for Clinical Research, OMRSE: Ontology of Medically Related

The development of the MHA Semantic Core Ontology
was based on the following principles: a) Reuse: Exploit
already established high quality ontologies; b) Granularity: A
single ontological resource is not adequate to model the
multi-faceted ecosystem of eHealth so multiple ontologies
should be used; c) Modularity: Create a framework where
different ontologies would be able to integrate many modules
through mappings and equivalences between ontology terms.
In our case, after an initial evaluation, 34 sub-ontologies
were selected and integrated through an extension of the
Translational Medicine Ontology [5] (eTMO). The result is
shown in Figure 2. The integration is achieved by introducing
terms from these sub-ontologies to the eTMO ontology and
via relations of equivalence (using owl:equivalentClass) and
subsumption (rdfs:subClassof) from eTMO to the various
ontology modules. These relations (~300) were manually
identified and verified using the NCBO BioPortal
4
.
B. NoSQL Data Repository
The lifelong patients’ data to be stored is complex, with
hundreds of attributes per patient record that will continually
evolve as new types of calculations and analysis/assessment
results are added to the record over time. Apache Cassandra
5
is used to store these large-scale original data. Cassandra is
an open-source, peer-to-peer and key value based store,
where data are stored in key spaces. Cassandra has also built-
in support for the Hadoop implementation of MapReduce [6],
considered currently state of the art for real-time data analysis
and has advanced replication functions. As such, Cassandra is
used to store all original data and to provide input to
applications that require big-data real-time analysis. In the
MHA platform the Cassandra repository is an instantiation of
a “data lake
6
” concept that stores the raw data supplied by the
different information sources.
C. Semantic Integration & Data Warehouse
Although Cassandra is an excellent choice for storing and
processing large amounts of data, the restrictions imposed on
querying (e.g. lack of joins) prohibit the interconnection and
the real integration of the data. However, the integrated
information of the whole or parts of the patient profile is
required for the provision of specific health care services. To
achieve the semantic integration of the available data we use
an extension of the exelixis [7] [8] platform, a novel data
integration engine that has two main functionalities: a) It
achieves query answering by accepting SPARQL queries that
are rewritten to the data sources; b) it allows the
transformation of data from original models to RDF/S data
according to the MHA Ontology Suite. The platform allows
the integration of a variety of data sources such as relational
and NoSQL databases, XML, RDF/S and CSV documents,
web services etc. Using the aforementioned platform we
select which of the available data should be semantically
linked and integrated by establishing the appropriate
Social Entities, PATO: Phenotypic Quality Ontology, PLACE: Place
Ontology, PRO: Protein Ontology, RO: Relation Ontology, SBO: Systems
Biology Ontology, SNOMED-CT: SNOMED clinical terms, SO: Sequence
Ontology, SYMP: Symptom Ontology, TIME: Time Ontology, UMLS:
Unified Modeling Language System.
4
http://bioportal.bioontology.org/
5
http://cassandra.apache.org/
6
http://martinfowler.com/bliki/DataLake.html
mappings. Then these data are queried, transformed into
triples and loaded to the Semantic Warehouse where they are
available for further reasoning and querying. A benefit of the
approach is that we can recreate from scratch the resulting
triples at any time. However for reasons of efficiency the
exelixis transforms periodically only the newly inserted
information by checking the timestamps of the data.
As already described, in order to select the information
that is integrated, the proper mappings are established
between parts of the source schemata and the MHA Semantic
Core Ontology. However, the definition of those mappings is
a time-consuming, labor-intensive and error-prone activity.
To assist human in this difficult task, we created an
innovative mapping workflow that manages the core
processes needed to create, maintain and manage mapping
relationships between different data sources over the long
term, with high level of quality control. This novel workflow
is named X3ML [4] and is composed of two main steps,
shown in Figure 3: a) Schema Matching: The domain experts
define a matching between the individual schemata and the
ontology with the help of a graphical tool which is
documented in a schema matching definition file. This file is
human and machine readable and is the ultimate
communication mean on the semantic correctness of the
mapping; b) Mapping definition: In this step the actual
mappings are generated based on the input of the previous
step. In this step only the IT experts are involved and domain
experts have no interest or knowledge about it.
Figure 3. X3ML Mapping workflow
D. Linking Services to External Sources
Besides sources allowing the direct integration through
mappings, the data management infrastructure supports the
incorporation of parts of the patient’s clinical and social
history that are already stored and managed by third party
systems.
Figure 4. Linking MHA with external systems through well-defined
interfaces
For this reason the proper mechanisms are in place for
retrieving relevant user information from these external data
sources. This “linking” mechanism is based on well-known

and established standard interfaces since they allow the
building of generic ports and interfaces and the reuse of
existing code bases. Figure 4 shows some notable examples
for the realization of these links to external resources:
Clinical data are retrieved from Hospital Information
Systems (HIS) through the Clinical Document Architecture
(CDA
7
) guidelines and set of specifications, clinical trial
specific patient data are acquired using the Operational Data
Model (ODM) of the Clinical Data Interchange Standards
Consortium (CDISC
8
), whereas cross-border healthcare
provisioning is supported by the adoption of epSOS
9
Patient
Summary (again, based on CDA) interfaces. All these
interfaces enable pushing data that are stored within the
NoSQL data repository.
E. APIs for Third-Party Access
The APIs focus on providing functions to potential
applications for accessing and managing data stored in the
original data repository and the semantic data warehouse.
The APIs are implemented in RESTful style and JSON is
used as communication format. In addition oAuth
10
provides
secure delegated access to the available resources.
IV. RELATED WORK
Projects with similar goals for collecting, storing and
accessing eHealth data were the eHealthMonitor
11
[9] and the
INTEGRATE
12
projects whereas currently running projects
on the area include the p-Medicine
13
and the EURECA
14
projects. In all those projects, the need for integrating
disparate data sources has led to the adoption of multiple
ontological resources as well. However, it is the first time
that a separation is achieved between the original data
collected from a variety of sources and a semantic repository
to support the data needed integration and interconnection.
Besides European projects, there also exists a set of
initiatives and personal health record systems concerned with
the management of patient data. Some of them are Microsoft
HealthVault, PatientsLikeMe, Indivo-X, Tolven, Dossia etc.
(see [10] for a comparison). However, opposed to the work
presented here they act as a static knowledge spaces –
providing only storage and role based access to the
knowledge resources.
To the best of our knowledge the proposed data
management solution is the only one allowing the separation
of the original and the semantically enhanced information,
combining Ontologies, Semantics and NoSQL databases
allowing a wide range of methods for pushing and retrieving
information. The added value of our approach is that real-
time analysis can directly be performed on the original data
whereas semantic queries on integrated data can be
efficiently answered using the Semantic Data Warehouse
allowing a clear separation of concerns between the
Cassandra and the Semantic Repository.
7
http://www.hl7.org/Special/committees/structure/index.cfm
8
http://www.cdisc.org/
9
http://www.epsos.eu/
10
http://oauth.net/
11
http://ehealthmonitor.eu/
12
http://www.fp7-integrate.eu/
13
http://www.p-medicine.eu/
14
http://eurecaproject.eu/
V. DISCUSSION & CONCLUSION
Our architecture adopts a variation of the command-query
responsibility segregation principle
15
where one uses a
different model to update information than the model one is
using to read. Although the mainstream approach people use
for interacting with an information system is to treat it as a
create, read, update and delete data-store, as the needs
become more sophisticated state of the art approaches
steadily move away from that model. In our case we rely on
NoSQL technologies to store the original data due to their
ability to handle enormous data sets and the “schema-less”
nature, which makes, to a large extent, the import of new
information to be frictionless. But their limitations in the
flexibility of query mechanisms are a real barrier for any
application that has not predetermined access use cases. The
Semantic Warehouse component in the MHA platform fills
these gaps by effectively providing a semantically enriched
and search optimized index to the unstructured contents of
the Cassandra repository. Therefore, our approach tries to
offer best of both worlds: efficient persistence and
availability of heterogeneous data, and semantic integration
and searching of the “essence” of the ingested information.
A key next step is to evaluate the whole platform in a
real-world context in the four clinical scenarios in three
European countries (United Kingdom, Greece and Germany).
Preliminary evaluation performed, provided initial evidences
about the added value and the usability of our approach
which will be extensively reported in a follow-up paper.
Without a doubt data managements is an important area for
healthcare that will only become more critical as healthcare
delivery continues to grapple with current challenges.
REFERENCES
[1] eHealth Task Force Report, “Redesigning health in Europe for 2020”,
2012.
[2] E.G. Spanakis, P. Lelis, F. Chiarugi, C. Chronaki, “R&D challenges in
developing an ambient intelligence eHealth platform”, EMBEC, pp.
1727-1983, 2006.
[3] E.G. Spanakis, D. Kafetzopoulos, P. Yang, K. Marias, Z. Deng, M.
Tsiknakis, V. Sakkalis, F. Dong, "MyHealthAvatar: Personalized and
empowerment health services through Internet of Things
technologies", Mobihealth, pp. 331-334, 2014.
[4] MyHealthAvatar Consortium: D4.2 Extension of the Semantic Core
Ontology, February 2015.
[5] J.S. Luciano, S. Joanne S et al., “The Translational Medicine
Ontology and Knowledge Base: Driving Personalized Medicine by
Bridging the Gap between Bench and Bedside.” Journal of
Biomedical Semantics 2.Suppl 2 (2011): S1. 2015.
[6] J. Dean, S. Ghemawat, “Map Reduce: simplified data processing on
large clusters”, Communications of the ACM, vol. 51, no. 1, pp. 107–
113, 2008.
[7] H. Kondylakis, D. Plexousakis, “Exelixis: Evolving Ontology-Based
Data Integration System”, ACM SIGMOD, pp. 1283-1286, 2011.
[8] H. Kondylakis, D. Plexousakis, “Ontology Evolution without Tears”,
Journal of Web Semantics, 19, pp. 42-58, 2013.
[9] H. Kondylakis, D. Plexousakis, V. Hrgovcic, R. Woitsch, M. Premm,
M. Schuele, “Agents, Models and Semantic Integration in support of
Personal eHealth Knowledge Spaces”, WISE, pp. 496-511, 2014.
[10] I. Genitsaridi, H. Kondylakis, L. Koumakis, K. Marias, M. Tsiknakis,
“Towards Intelligent Personal Health Record Systems: Review,
Criteria and Extensions”, Procedia Computer Science, vol. 21, pp.
327-334, 2013.
15
http://en.wikipedia.org/wiki/Command%E2%80%93query_separation