![Key dimensions of a Digital Health Framework for enhancing communication among Informal Care (top layer), Health Care (central layer) and Biomedical Research (bottom layer), as first described in ref. [24]. Informal care refers to all factors, traditionally not included in health information systems with an impact on health, as depicted in the figure, including patients' self-tracking data. Moreover, recent studies [19] strongly support the use of population-health information (registry data) to enrich multilevel health risk prediction for clinical purposes, as is extensively explained here.](https://www.researchgate.net/profile/Akos-Tenyi/publication/329593363/figure/fig2/AS:722916652101636@1549368169330/Key-dimensions-of-a-Digital-Health-Framework-for-enhancing-communication-among-Informal_Q320.jpg)
Content uploaded by Ákos Tényi
Author content
All content in this area was uploaded by Ákos Tényi on Feb 05, 2019
Content may be subject to copyright.
Clin Chem Lab Med 2019; 57(3): 317–327
Opinion Paper
Josep Roca*, Akos Tenyi and Isaac Cano
Paradigm changes for diagnosis: using
big data for prediction
https://doi.org/10.1515/cclm-2018-0971
Received September 3, 2018; accepted November 21, 2018;
previously published online December 11, 2018
Abstract: Due to profound changes occurring in bio-
medical knowledge and in health systems worldwide, an
entirely new health and social care scenario is emerging.
Moreover, the enormous technological potential devel-
oped over the last years is increasingly influencing life sci-
ences and driving changes toward personalized medicine
and value-based healthcare. However, the current slow
progression of adoption, limiting the generation of health-
care efficiencies through technological innovation, can be
realistically overcome by fostering convergence between a
systems medicine approach and the principles governing
Integrated Care. Implicit with this strategy is the multidis-
ciplinary active collaboration of all stakeholders involved
in the change, namely: citizens, professionals with differ-
ent profiles, academia, policy makers, industry and pay-
ers. The article describes the key building blocks of an
open and collaborative hub currently being developed in
Catalonia (Spain) aiming at generation, deployment and
evaluation of a personalized medicine program address-
ing highly prevalent chronic conditions that often show
co-occurrence, namely: cardiovascular disorders, chronic
obstructive pulmonary disease, type 2 diabetes mellitus;
metabolic syndrome and associated mental disturbances
(anxiety-depression and altered behavioral patterns lead-
ing to unhealthy life styles).
Keywords: big data; Integrated Care; personalized medi-
cine; systems medicine; value-based healthcare.
Three main forces driving paradigm
change in health care
Since the beginning of the 21th century, medicine has been
experiencing profound qualitative changes that are driven
by the convergence of three main forces that are impos-
ing a paradigm change. Firstly, advances in biomedical
knowledge, with changes in disease understanding and
in prevention and treatment in the post-genomic era, have
shaped the methodologies associated to systems (preci-
sion) medicine [1, 2]. Moreover, functionalities of health
systems worldwide have been challenged by the epidem-
ics of chronic conditions that are shaping an entirely new
health and social care scenario [3, 4]. Last, but not least,
the enormous technological potential developed over the
last years is increasingly influencing life sciences and
driving changes toward personalized medicine and value-
based healthcare [1, 2].
Systems (precision) medicine
Since the early 2000s, two key phenomena are prompt-
ing substantial changes in both biomedical research and
clinical management of chronic patients. Firstly, systems
biology methodologies (i.e. ‘omics’ technologies, use of
computational modeling, etc.), taking a holistic approach
for solving biomedical challenges, are being progressively
embedded into medical practice shaping the practicalities
of systems medicine [1, 2]. This new field promises a novel
approach to disease, shifting its definition from phenotyp-
ical signs and symptoms towards molecular subtypes (i.e.
endotypes) of diseases, which is indispensable for precise
characterization of disease relations and for the evalua-
tion of shared mechanisms [5–9].
Simultaneously, digital health initiatives and weara-
ble devices are facilitating access to an enormous amount
of patient-related information from whole populations to
personal levels, and state-of-the art computational models
and machine learning tools demonstrate a high potential
for health prediction [1, 2]. Given the extremely long and
*Corresponding author: Josep Roca, MD, PhD, Hospital Clínic,
IDIBAPS, Facultat de Medicina, Universitat de Barcelona,
C/Villarroel 170, 08036, Barcelona, Catalunya, Spain; and Centro
de Investigación Biomédica en Red de Enfermedades Respiratorias
(CIBERES), Av. Monforte de Lemos, 3-5. Pabellón 11. Planta 0,
28029, Madrid, Catalunya, Spain, Phone: +34-932275747,
Fax: +34-932275455, E-mail: jroca@clinic.cat
Akos Tenyi and Isaac Cano: Hospital Clínic, IDIBAPS, Facultat de
Medicina, Universitat de Barcelona, Barcelona, Catalunya, Spain;
and Centro de Investigación Biomédica en Red de Enfermedades
Respiratorias (CIBERES), Madrid, Catalunya, Spain
Brought to you by | Kando Muszaki Foiskola
Authenticated
Download Date | 2/5/19 1:00 PM
318 Roca etal.: Paradigm changes for diagnosis: using big data for prediction
expensive bench to clinics cycles of the biomedical sector,
these technologies promise a fast-track approach where
scientific evidence can support clinical care while simul-
taneously collected insights from daily clinical practice
promote new scientific discoveries and optimize health-
care optimization.
The application of systems biology tools to analyze the
underlying molecular mechanisms of specific diseases, as
well as the complexities of disease co-occurrence can con-
tribute to explain patients’ heterogeneities. The proposed
approach should lead to the generation of novel biomedi-
cal knowledge for the enhancement of patient stratifica-
tion while exploring applicable strategies for improved
patient management and the decrease of disease burden.
Integrated Care
Healthcare systems worldwide are undergoing profound
changes to adapt to novel patient-oriented efficient ser-
vices necessary to tackle the increasing epidemic of non-
communicable diseases (NCDs) and their risk factors
caused by unhealthy lifestyles and population aging.
Facing this challenging situation implies a profound
reshaping in the way we approach delivery of care, as
well as cultural changes both at citizen and health pro-
fessional levels. Patient empowerment is an essential part
of the healthy lifestyles interventions and the preven-
tion and management of NCDs. The roles of patients and
healthcare professionals are evolving to place patients at
the center of the management as an active partner of the
healthcare process, both as co-designer of the interven-
tions and manager of her/his condition [10].
The World Health Organization (WHO) [11] defines
Integrated Care as “the organization and manage-
ment of health services so that people get the care they
need, when they need it, in ways that are user friendly,
achieve the desired results and provide value for money
system”. Despite known uncertainties in the characteris-
tics of on-going deployment processes, the new models
of healthcare delivery are prompted toward large-scale
implementation. The term scaling-up is defined by WHO
as “deliberate efforts to increase the impact of health
service innovations successfully tested in pilot or experi-
mental projects so as to benefit more people and to foster
policy and program development on a lasting basis” [12].
It is acknowledged that there is a need for health-value
generation of innovative information and communication
technologies (ICT)-supported healthcare services. The
final aim is to enhance health outcomes in a cost-effective
manner such that it allows containment of overall health
costs, with a more proactive experience for patients and
healthcare professionals.
The challenge imposed by the epidemic of chronic
disorders constitutes one of the main driving forces that
are prompting an extensive European deployment of Inte-
grated Care services supported by ICT through initiatives
like the European Innovation Partnership for Healthy and
Active Aging (EIP-AHA) [13] and the European Institute of
Technology-Health (EIT-Health) [14].
Digital health tools
It is acknowledged the role of ICT enabling enhanced bio-
medical knowledge and facilitating the change in health-
care, as already alluded to above. However, the current
slow progression of adoption, limiting generation of health-
care efficiencies through technological innovation, can be
realistically overcome by fostering convergence between a
systems medicine approach and the principles governing
Integrated Care. Implicit with this strategy is the multidisci-
plinary active collaboration of all stakeholders involved in
the change, namely: citizens, professionals with different
profiles, academia, policy makers, industry and payers.
The emerging health scenario should facilitate adop-
tion of technologies capable of longitudinally measuring
multilevel health parameters at high resolution that will
be coupled with dynamic knowledge repositories and
sophisticated analytics feeding decision support systems
addressed to citizens (patient decision support systems,
PDSS) and/or professionals (clinical decision support
systems, CDSS).
A long journey toward a novel
health scenario
The Catalan Reference Site, endorsed by EIP-AHA, is facil-
itating convergence among: (i) initiatives associated with
the regional Health Plan 2016–2020; (ii) resources from
main healthcare providers; and, (iii) input from different
competitive grants, favoring an ecosystem that ensures
transition from previous pilot experiences [15] to large-
scale deployment and adoption of on-going activities
showing health value generation (Figure 1).
As a successful example of this transition from pilot
experiences showing cost-efficacy to implementation in
real world settings assessing health-value generation,
we would like to introduce general characteristics of the
on-going Nextcare program [16]. Nextcare stands for Inno-
vation in Integrated Care Services for Chronic Patients. It
Brought to you by | Kando Muszaki Foiskola
Authenticated
Download Date | 2/5/19 1:00 PM
Roca etal.: Paradigm changes for diagnosis: using big data for prediction 319
is a RIS3 (Research and Innovation Strategies for Smart
Specializations) program (2016–2019) that constitutes
the umbrella for all the initiatives included in the current
research. It has three main objectives: (i) Regional deploy-
ment of Integrated Care services for chronic patients with
a personalized medicine approach; (ii) Development of a
test-bed, willing for international leadership, for the use
of ICT in novel services that generate value in the health-
care system of Catalonia; and, (iii) Development and mon-
etization of novel products and services with a high level
of transferability to other healthcare systems, contribut-
ing to strengthen Catalan industrial competences.
The project addresses five actions (Figure 2) that
encompass the main challenges encountered during the
deployment of Integrated Care services. The main objective
of Action 1 is Health Risk Assessment and Service Selec-
tion [17] with a special focus on non-pulmonary manifes-
tations and co-morbidities seen in patients with chronic
obstructive pulmonary disease were analyzed as a use
case [18, 19]. Action 2 aims at promoting healthy lifestyles
with a focus on physical activity [20, 21]. Action 3 deploys
community-based management of complex chronic
patients (CCP) [22]. Action 4 addresses regional deploy-
ment of transfer of diagnostic testing to primary care
focusing on forced spirometry as a use case [23]. Finally,
Action 5 promotes interoperability between healthcare,
informal care and biomedical research shaping the so-
called digital health framework (DHF) (Figure 3) [24, 25],
as a technological facilitator that supports collaborative
adaptive case management among health professionals
and with patients.
To this end, coordination between the healthcare
system and the informal care environment is a central
component of the five actions aiming at achieving an effi-
cient management of multimorbidity. The regional shared
electronic health record (HC3) and the personal health
folder (La Meva Salut) are the backbone components of
the current technological support to the services. NEXT-
CARE is transforming the current personal health folder
into a citizens’ self-management tool and developing tech-
nological support to adaptive case management of clinical
Figure 1: In the new health scenario, innovative ICT-supported
health services will deliver cost-effective care for chronic patients
based on systems (precision) medicine approach with a preventive
orientation. It will imply enhanced multilevel health risk predictive
modeling feeding patient decision support systems (PDSS) and
clinical decision support systems (CDSS). Cloud-based computing
architectures for data analytics will be a need for dynamic
assessments while complying with privacy and regulatory issues.
Allin all, the novel scenario requires building-up efficient innovation
ecosystems involving all the relevant stakeholders contributing to
convergence between systems (precision) medicine and Integrated
Care as the strategy required to shape the new health paradigm.
Figure 2: Schematic representation of the five actions, A1 to A5, addressed by NEXTCARE aiming at regional deployment of Integrated Care
services for chronic patients.
Brought to you by | Kando Muszaki Foiskola
Authenticated
Download Date | 2/5/19 1:00 PM
320 Roca etal.: Paradigm changes for diagnosis: using big data for prediction
processes to support collaborative work among the actors.
The articulation of the protocols described in the main
text provides the frame for a comprehensive evaluation of
regional deployment of Integrated Care.
The ultimate aim of Nextcare is to set the basis for
large-scale deployment of innovative ICT-supported
healthcare services with proven high potential for trans-
ferability to different scenarios and geographical areas.
It is hypothesized that the outcomes of the program shall
contribute to adoption of cost-effective services aiming
for high impact on prevention of multi-morbidity, as well
as the progression of chronic patients toward the tip of
the population-health risk stratification pyramid. Conse-
quently, the implicit general hypothesis of the ongoing
actions is that these novel services can effectively facili-
tate decrease in both the health and societal burden of
chronic conditions.
Four challenges
The process of deployment of ICT-enabled services within
the framework of Nextcare has facilitated identification of
four major areas requiring attention in order to achieve a
successful transition toward adoption of systems medi-
cine for chronic patients, within an Integrated Care sce-
nario, as described in the following subheadings.
Enhanced clinical predictive modeling
Health risk assessment and service selection are widely
accepted tools facilitating large-scale adoption of
Integrated Care of chronic patients and contributing to
enhance healthcare outcomes and patient experience
of care while reducing costs and improving the health
of populations. The wealth of data available may allow
to consider several avenues of computational modeling,
including artificial intelligence and deep learning or
machine learning tools, based on knowledge and exper-
tise of the data analytics team. However, their application
may face major limitations when it comes to accessing and
mining health data currently stored in distributed silos of
information.
To overcome these major limitations, there is a clear
need for application of holistic strategies for subject-
specific risk prediction and stratification that incorporate
multi-level determinants of health (e.g. socio-economical,
lifestyle, behavioral, clinical, physiological, cellular and
“omics” information [17] into risk models, which would
substantially increase their predictive accuracy and their
use in clinical decision-making (Figure 4). To this end,
the open data trends of biomedical research should be
similarly extended to clinical practice by solving privacy
and regulatory constraints. Articulation of these strate-
gies with systems-oriented biomedical research would
provide continuous cross-fertilization between research
and patient care [2].
An ideal healthcare setting (Figures 1 and 4) should
facilitate an optimal support to care decisions and deliv-
ery by reducing the complexity of the massive amount
of multi-disciplinary data being produced every day
and should improve efficiency of health outcomes both
in terms of well-being and expenditures. Such a health
system relies on the availability of health-related data and
analytics tools. Enhanced clinical predictive modeling
and personalized diagnostic and treatment tools, such
Figure 3: Key dimensions of a Digital Health Framework for enhancing communication among Informal Care (top layer), Health Care (central
layer) and Biomedical Research (bottom layer), as first described in ref. [24]. Informal care refers to all factors, traditionally not included
in health information systems with an impact on health, as depicted in the figure, including patients’ self-tracking data. Moreover, recent
studies [19] strongly support the use of population-health information (registry data) to enrich multilevel health risk prediction for clinical
purposes, as is extensively explained here.
Brought to you by | Kando Muszaki Foiskola
Authenticated
Download Date | 2/5/19 1:00 PM
Roca etal.: Paradigm changes for diagnosis: using big data for prediction 321
as CDSS and PDSS, can contribute to the acceleration of
transfer of scientific evidence to practice, helping in the
identification of gaps in care and in targeting interven-
tions to the most appropriate sub-populations of patients.
Development of enhanced CPM to feed DSS will
require consideration, and eventual integration, of com-
putational modeling of four different dimensions: (i)
Underlying biological mechanisms; (ii) Current evidence-
based clinical knowledge; (iii) Patients’ self-tracked data,
including sensors, behavioral, environmental and social
information; and, (iv) Population-based health risk assess-
ment data, as already depicted in Figure 3 and described
in the text. CDSS/PDSS should be designed to interoperate
with existing centralized or distributed hospital informa-
tion systems, and ultimately with learning health systems
(i.e. systems in which science, informatics, incentives
and culture are aligned for continuous improvement and
innovation, with best practices seamlessly embedded in
the delivery process and new knowledge captured as an
integral by-product of the delivery experience).
Nextcare is successfully developing a systems medi-
cine approach to better understand patients with chronic
conditions with a focus on chronic obstructive pulmonary
disease (COPD) as a use case. The overall strategy and spe-
cificities of the program, as well as key reported achieve-
ments, have been recently analyzed in detail in [18]. Briefly,
the novel approach has contributed to uncover shared
underlying mechanisms of non-pulmonary phenomena,
such as skeletal muscle dysfunction and common co-
morbid conditions, observed in patients with COPD. The
results pave the way toward enhanced health risk assess-
ment of these patients with relevant implications on clini-
cal management of chronic conditions.
Technology: cloud-based architecture
andservices
Holistic implementation strategies of cloud comput-
ing environments, tackling privacy and regulatory con-
straints, have been identified as main enablers of this
ideal healthcare setting [25, 26]. Currently, the articula-
tion of the main technical building blocks, i.e. multi-level
biomedical data integration, tools for clinical predictive
modeling in the cloud and high-performance computing,
as one integrated system is as yet, an unmet potential.
This is strongly highlighted by the exponential growth
of health-related data that could enhance current health
information systems with functionalities to assist clini-
cal decision-making of health professionals, as well as of
patients for self-management. End-to-end exploitation of
cloud infrastructures for large-scale data analytics has so
far been held back by the lack of a well-integrated set of
reliable and flexible services, and user-friendly interfaces.
This problem is particularly true for medical research and
clinical applications where the additional complexity of
handling personal information requires particular care. In
this context, a cloud-based data analytics platform shall
unlock the full potential of clinical predictive modeling
by solving issues around integration, harmonization and
Figure 4: Scheme for dynamic enhancement of clinical predictive modeling (CPM) feeding clinical DSS (CDSS) and/or patient DSS (PDSS) in
cloud-based environments.
Brought to you by | Kando Muszaki Foiskola
Authenticated
Download Date | 2/5/19 1:00 PM
322 Roca etal.: Paradigm changes for diagnosis: using big data for prediction
privacy of data coming from different sources in an inte-
grated manner, and by providing a general interface for
developing and deploying predictive models.
The added benefit of such a solution includes rapid
local prototyping, cost-effective parameter sweeping and
validation scale-out to high-performance, large-scale
modeling. It should also enable the deployment of the
same services across different infrastructures, institutes,
laboratories or projects according to local policies, while
maintaining interoperability and consistency at the plat-
form layer. Specific implementation of such a platform
should identify and deploy cloud-based services for
private and public uses with a design that is versatile, scal-
able, trusted and abstract enough to support a wide range
of clinical predictive modeling approaches and applica-
tion in broad operational contexts for digital health and
daily clinical practice.
A high-level description of a proposed cloud-based
data analytics platform is displayed in Figure 5 indicat-
ing the four types of data sources that are considered
for multi-disciplinary computational modeling. Compu-
tational modeling in combining these four dimensions
should provide the basis for enhanced clinical predictive
modeling and elaboration of a cloud-based CDSS/PDSS
embedded into clinical processes.
In the implementation process of such a platform,
specific requirements include (i) development of a data
interoperability engine for standardized data exchange
and semantic interoperability among multi-disciplinary
data sources; (ii) deployment of technologies to handle
personal data in compliance with agreed legal, policy and
standardization requirements; and (iii) defining standard
interfaces in compliance with user needs.
Data interoperability engine
A critical aspect of the proposed platform is the develop-
ment of an efficient data interoperability engine compli-
ant with European regulations and FAIR data principles:
(i) Findable: data must be easy to find by both humans
and computer systems; (ii) Accessible: data must be put in
long-term storage in such a way that either the data itself
or its metadata can be accessed easily; (iii) Interoperable:
datasets can be combined by humans as well as computer
systems, in which the use of shared vocabularies and/or
ontologies is of special importance; and, (iv) Re-usable:
data can be used for future research and to be processed
further by computer programs. To this end, several ini-
tiatives are available for consideration, e.g. the ELIXIR
(www.elixir-europe.org) interoperability platform or com-
mercial medical data federation tools such as FedEHR
(www.fedehr.com).
Data security and data protection
Given the need to handle sensitive data, particular atten-
tion is needed to be paid to security and data protection
aspects. Future data architectures should assume that
Figure 5: Conceptual view of the proposed cloud-based data analytics platform.
Properly harmonized multi-disciplinary data sources provide the basis for enhanced clinical predictive models that feed real-time decision
support systems able to guide health professionals in the clinical decision-making process.
Brought to you by | Kando Muszaki Foiskola
Authenticated
Download Date | 2/5/19 1:00 PM
Roca etal.: Paradigm changes for diagnosis: using big data for prediction 323
sensitive data cannot be moved around, but rather code
and models have to be moved to the data. When consid-
ering such architectures three levels of data should be
considered: (i) public data; (ii) anonymized data; (iii)
pseudo-anonymized data. Public data includes public
datasets that can be freely shared and accessed from well-
known, community curated data banks and repositories.
In this context, creation of public “data lakes” shared
across the community and used, for example, to train the
machine learning services and build reference models are
needed. Anonymized data need to be defined in compli-
ance with existing and future legal definitions and need
to be accessed and contributed based on agreed policies
of use within specific projects or communities. Such data,
although anonymized, is richer than completely public
data as it might convey a specific context of use, thus it
is of high interest for research purposes. Finally, pseudo-
anonymized data should be used to generate models for
the CDSS/PDSS that can then be applied to patients in
clinically relevant scenarios. In light of the new European
General Data Protection Regulation (GDPR) new technolo-
gies are also needed to be considered to manage and audit
access to data and to move in the direction of implement-
ing the GDPR norms of data ownership and usage. Data
transactions and “smart” user consents could be stored
and managed using blockchain technologies to enable
the implementation of transparent, end-to-end policies
for data governance.
User-profiled interfaces
One of the main barriers for translational researchers and
clinicians in accessing cloud-based services is currently
the lack of intuitive, easy-to-use user interfaces. The
main objective of the technical implementation should
be to provide an intuitive way for healthcare researchers
and practitioners to exploit the capabilities of distributed
infrastructures, i.e. accessing a wide range of data sources
and services without having to understand any of the
complexity of the system. At least two levels of interfaces
should be provided in a desired platform, which should
be co-designed with the communities of use of reference:
– Data analysis interfaces for translational researchers:
this type of interfaces provide an intuitive, scriptable,
portable access to the platform services. Technologies
like Jupyter Notebooks, CERN SWAN (Service for Web
based ANalysis) and other similar approaches provide
good examples of such environments. The possibil-
ity of supporting different types of scripting engines
including familiar systems such as the R analysis
framework is critical to ease the transition from local
computers to cloud-based services.
– Decision support interfaces for healthcare profession-
als and patients: this type of interface should provide
clean, web-based access to specialized services and
applications, hiding from the end-user (i.e. health-
care professional and patients) the complexity of the
underlying infrastructure service layers.
To comply with the vision of the cloud-based data analyt-
ics platform, methodologies used in biological and clinical
modeling should converge towards standard operations
and tools that can be integrated into general pipelines
and implemented in analysis platforms. These pipelines
shall be ready to analyze data independent of their source
or type, i.e. molecular, clinical or wearable measure-
ments, and integrate them in an operational manner. This
assumes pre-processed and well-formatted data input, as
well as standardized outputs (Figure 6). When considering
subject-specific health risk prediction and stratification
as the desired output of such systems, the framework of
machine learning defines these input and output needs
as well as the identified challenges. In this regard, the
main factors to consider should be the dimensionality of
the data sources, i.e. the number of features that are used
for health prediction, their sample size and differences in
sample sizes when considering the integration of multi-
ple datasets. Registry data, EHR data and wearable tech-
nologies come with the great promise to bring biomedical
research to the Big Data era with population/subpopula-
tion size data, whereas molecular data have great poten-
tial to gain biological insight into disease mechanisms,
however, for these data sources population-wide avail-
ability is still awaited. The integration of such data sources
should enable mining health related patterns from data
with state-of-the art technologies, such as deep learning
that show exciting potential for identifying non-linear pat-
terns from large amounts of raw biomedical data [27–29].
The major potential of this technology is that it promises a
universal approximator for many learning and prediction
tasks that could substitute several processes that are cur-
rently done separately in biomedical and machine learn-
ing fields. A fascinating way of using deep learning could
help to select biologically important features, organize
them into higher abstraction level biological assemblies
(e.g. pathways, disease modules), highlight their role in
the disease and also to predict disease risk using them [27].
A major obstacle, however, is that they are often associated
with the need for large longitudinal sample sizes, which
is a barrier especially in molecular data sources. Further-
more, more research is needed in this field to abolish its
Brought to you by | Kando Muszaki Foiskola
Authenticated
Download Date | 2/5/19 1:00 PM
324 Roca etal.: Paradigm changes for diagnosis: using big data for prediction
current stigmatization as a “black box” approach, which is
often seen as a barrier for clinical application.
Moreover, integration of data from different sources
and with different formats is a major challenge of in-silico
modeling pipelines. Current practice shows that in dif-
ferent fields different models evolve, such as the disease
maps, disease trajectories, mechanistic models, other
multiscale hybrid modeling already combining some of
the previous approaches; or, data-driven approaches
using machine learning. In order not to waste the field
specific knowledge encapsulated in these models, inte-
grative approaches are needed. In this context, the patient
similarity framework shows immense potential to be
applied in the clinical field, mainly due its high abstrac-
tion level leading to broad applicability, its patient-cen-
tered approach and its transparent methodology, which
is especially important for acceptability from the clinical
side [30, 31]. Patient similarity enables the separate com-
parison of patients on different biological organizational
levels, e.g. using molecular profiles (transcriptome,
genome, epigenome, etc.), clinical traits, comorbidities,
and allows retrieving groups of similar patients, or the
most successful treatments based on similar cases, as
well as to predict health risk in an unsupervised manner
[32–35].
Evaluation and adoption of decision support
systems (DSS)
For successful adoption in real world settings, interopera-
bility of the proposed cloud-based data analytics platform
with healthcare information systems is indispensable
(Figure 6). On the one hand, cloud-based services should
be integrated at the site level with the required structured
and unstructured data sources.
On the other hand, information systems departments
of clinical sites should take into account in-place health
information exchange infrastructures, where standard ter-
minology (e.g. SNOMED-CT, SERAM, SEMN, LOINC, etc.),
message encoding (e.g. HL7 2.x/3.x, MLHIM, openEHR,
ISO 13606, etc.), message routing and security (e.g. IPSec,
Audit trail, Node authentication, etc.) are of special impor-
tance. Where available, existing controlled vocabularies
such as the ICD-10 or the FMA human anatomy, standards
for data and metadata format (e.g. ISA-TAB) and content
Figure 6: Proposed interoperability architecture.
The data sourcing layer is responsible for integrating cloud-based services to required structured and unstructured data sources at site-
level. The extract-transform-load (ETL) layer is where data is extracted from homogeneous or heterogeneous data sources and transformed
for storing in the proper format or structure for the purposes of querying and analysis the target data lake or data warehouse. Finally,
the application layer provides access to common platform services for data analytics whereas the pre-processing layer is responsible for
integrating DSS with site-specific clinical workstations and patient gateways (e.g. La Meva Salut in Catalonia, Spain).
Brought to you by | Kando Muszaki Foiskola
Authenticated
Download Date | 2/5/19 1:00 PM
Roca etal.: Paradigm changes for diagnosis: using big data for prediction 325
(e.g. MSI or MIAME) should be used. Where standards are
currently not yet broadly accepted, agreements should be
generated to deploy site-level interoperability middleware
based on an HL7 FHIR standard specification (e.g. HAPI
FHIR, hapifhir.io).
Moreover, successful adoption in real world settings
will likely require specific evaluation designs to assess
effectiveness and health value generation of complex
interventions, ranging from standard randomized con-
trolled trials to different implementation science designs
[36, 37]. The need for looking for complementarities
among different types of study designs has been recently
pointed out [38].
Barriers and facilitators for deployment
and adoption
Governance of a cloud-based system based on FAIR data
principles generates several major challenges in terms of:
(i) data/services administration accessibility; (ii) continu-
ous control of quality assurance programs; (iii) compli-
ance with ethical and regulatory issues; as well as, (iv)
sustainability of the approach over time. These man-
agement and governance challenges are expected to be
overcome by adopting block chain technology that allow
complete traceability of transactions, and most impor-
tantly finely granular enforcement of rules in observance
of regulation at data origins, and of consent design. All
actors of cloud-based systems should be able to verify that
sharing, analysis and other handling and use of the infor-
mation is in accordance of the individual’s intentions and
applicable laws, regulations and processes.
Regulatory aspects
Several ethical and regulatory issues currently fall into
gray areas in terms of regulation (i.e. computational mod-
eling and DSS assessment for medical use). In this regard,
the generation of recommendations on gray areas should
be addressed in future work, with the specific objectives
to be covered:
– To ensure the protection of privacy and compli-
ance with the GDPR (EU) 2016/679, Directive 95/46/
EC and ISO27001 conformance for secure storage
of pseudonymized multi-disciplinary data, which
will improve trust in health research and therefore
will facilitate adoption of innovative digital health
services;
– To advise on applicable legislation and regulations to
which innovative mathematical models, and compo-
nents using them, need to comply before they can be
deployed and used in healthcare settings;
– To ensure the ethical use of innovative models within
patient and professional decision-making.
Service adoption
Adoption and both organizational (e.g. data processing
agreements, liability/responsibility aspects, etc.) and
financial sustainability (e.g. entrepreneurial actions)
of cloud-based services constitutes a major challenge,
wherein service models such as those developed in pro-
jects like MyHealthMyData (ICT-18-2016-732907) could be
considered. The development of a roadmap for large scale
deployment and adoption of cloud services at national
and EU level should be of main interest for future initia-
tives pursuing adoption of novel cloud-based services.
Conclusions
The cloud-based data analytics platform has been pro-
posed to successfully address the implicated potentials of
health risk assessment and stratification and to facilitate
large-scale adoption of Integrated Care of chronic patients
[17, 39], contributing to enhance healthcare outcomes
and patient experience of care while reducing costs and
improving the health of populations. Applying holistic
strategies for subject-specific risk prediction and strati-
fication, that consider multilevel covariates influencing
patient health, would increase the predictive accuracy
and facilitate clinical decision-making based on sound
estimates of individual prognosis.
Acknowledgments: Supported, in part, by NEXTCARE
(COMRDI15-1-0016), PITES-TliSS (PI15/00576) and AGAUR
research groups (2009SGR911 and 2014SGR661).
Author contributions: All the authors have accepted
responsibility for the entire content of this submitted
manuscript and approved submission.
Research funding: None declared.
Employment or leadership: None declared.
Honorarium: None declared.
Competing interests: The funding organization(s) played
no role in the study design; in the collection, analysis, and
interpretation of data; in the writing of the report; or in the
decision to submit the report for publication.
Brought to you by | Kando Muszaki Foiskola
Authenticated
Download Date | 2/5/19 1:00 PM
326 Roca etal.: Paradigm changes for diagnosis: using big data for prediction
References
1. Torkamani A, Andersen KG, Steinhubl SR, Topol EJ. High-defini-
tion medicine. Cell 2017;170:828–43.
2. Maddox TM, Albert NM, Borden WB, Curtis LH, Ferguson TB, Kao
DP, etal. The learning healthcare system and cardiovascular
care: a scientific statement from the american heart association.
Circulation 2017;135:e826–57.
3. Murray CJ, Lopez AD. Measuring the global burden of disease. N
Engl J Med 2013;369:448–57.
4. Blumenthal D, Chernof B, Fulmer T, Lumpkin J, Selberg J. Caring
for high-need, high-cost patients – an urgent priority. N Engl J
Med 2016;375:909–11.
5. Hidalgo CA, Blumm N, Barabási AL, Christakis NA. A Dynamic
network approach for the study of human phenotypes. PLoS
Comput Biol 2009;5:e1000353.
6. Menche J, Sharma A, Kitsak M, Ghiassian SD, Vidal M, Loscalzo
J, etal. Uncovering disease-disease relationships through the
incomplete interactome. Science 2015;347:1257601.
7. Jensen AB, Moseley PL, Oprea TI, Ellesoe SG, Eriksson R,
Schmock H, etal. Temporal disease trajectories condensed from
population-wide registry data covering 6.2million patients. Nat
Commun 2014;5:4022.
8. Goh K-I, Cusick ME, Valle D, Childs B, Vidal M, Barabási
A-L. The human disease network. Proc Natl Acad Sci USA
2007;104:8685–90.
9. Lee D-S, Park J, Kay KA, Christakis NA, Oltvai ZN, Barabási A-L.
The implications of human metabolic network topology for
disease comorbidity. Proc Natl Acad Sci USA 2008;105:9880–5.
10. ACT@Scale project. Advancing Care Coordination and Telehealth
at Scale. 2016-2019. https://www.act-at-scale.eu/.
11. World Health Organization. Integrated health services: what and
why? Switzerland: World Health Organization, 2008.
12. World Health Organization. Practical guidance for scaling up
health service innovations. 2009. http://apps.who.int/iris/bit-
stream/10665/44180/1/9789241598521_eng.pdf.
13. Bousquet J, Farrell J, Crooks G, Hellings P, Bel EH, Bewick M,
etal. Scaling up strategies of the chronic respiratory disease
programme of the European Innovation Partnership on Active
and Healthy Ageing (Action Plan B3: Area 5). Clin Transl Allergy
2016;6:29.
14. EIT-Health. European Institute of Innovation and Technology in
Health. https://www.eithealth.eu/.
15. Hernández C, Alonso A, Garcia-Aymerich J, Grimsmo A,
Vontetsianos T, García Cuyàs F, etal. Integrated care services:
lessons learned from the deployment of the NEXES project. Int J
Integr Care 2015;15:e006.
16. NEXTCARE program Innovation in Integrated Care Services for
Chronic Patients, COMRDI15-1-0016 2016. http://www.next-
carecat.cat.
17. Espieén ID, Vela E, Pauws S, Bescos C, Cano I, Cleries
M, etal. Proposals for enhanced health risk assessment
and stratification in an integrated care scenario. Br Med J
2016;6:e010301.
18. Tenyi A. A Systems Medicine approach to multimorbidity:
towards personalised care for patients with COPD. PhD Thesis,
2018. http://hdl.handle.net/2445/124046.
19. Vela E, Tényi Á, Cano I, Monterde D, Cleries M, Garcia-Altes
A, etal. Population-based analysis of patients with COPD
in Catalonia: a cohort study with implications for clinical
management. BMJ 2018;8:e017283.
20. Barberan-Garcia A, Gimeno-Santos E, Blanco I, Cano I, Martínez-
Pallí G, Burgos F, etal. Protocol for regional implementation of
collaborative self-management services to promote physical
activity. BMC Health Serv Res 2018;18:560.
21. Barberan-Garcia A, Gimeno-Santos E, Blanco I, Cano I,
Martínez-Pallí G, Burgos F, etal. Personalised prehabilitation
in high-risk patients undergoing elective major abdominal
surgery: a randomized blinded controlled trial. Ann Surg
2018;267:50–6.
22. Cano I, Dueñas-Espín I, Hernandez C, de Batlle J, BenaventJ,
Contel JC, etal. Protocol for regional implementation
ofcommunity-based collaborative management of
complex chronic patients. NPJ Prim Care Respir Med
2017;27:44.
23. Vargas C, Burgos F, Cano I, Blanco I, Caminal P, Escarrabill J,
etal. Protocol for regional implementation of collaborative
lung function testing. NPJ Prim care Respir Med 2016;26:
16024.
24. Cano I, Lluch-Ariet M, Gomez-Cabrero D, Maier D, Kalko S,
Cascante M, etal. Biomedical research in a digital health frame-
work. J Transl Med 2014;12:S10.
25. Cano I, Tenyi A, Vela E, Miralles F, Roca J. Perspectives on Big
Data applications of health information. Curr Opin Syst Biol
2017;3:36–42.
26. Directorate-General for Health and Food Safety (European Com-
mission). Study on big data in public health, telemedicine and
healthcare. 2016. https://ec.europa.eu/health/sites/health/
files/ehealth/docs/bigdata_report_en.pdf.
27. Alkawaa FM, Chaudhary K, Garmire LX. Deep learning accurately
predicts estrogen receptor status in breast cancer metabolomics
data. J Proteome Res 2018;17:337–47.
28. Miotto R, Li L, Kidd BA, Dudley JT. Deep patient: an unsupervised
representation to predict the future of patients from the elec-
tronic health records. Sci Rep 2016;6:26094.
29. Rajkomar A, Oren E, Chen K, Dai AM, Hajaj N, Hardt M, etal.
Scalable and accurate deep learning with electronic health
records. NPJ Digit Med 2018;1:18.
30. Brown S-A. Patient similarity: emerging concepts in systems and
precision medicine. Front Physiol 2016;7:561.
31. Gallego B, Walter SR, Day RO, Dunn AG, Sivaraman V, Shah N,
etal. Bringing cohort studies to the bedside: framework for a
‘green button’ to support clinical decision-making. J Comp E
Res 2015;4:191–7.
32. Lee J, Maslove DM, Dubin JA. Personalized mortality prediction
driven by electronic medical data and a patient similarity metric.
PLoS One 2015;10:e0127428.
33. Ng K, Sun J, Hu J, Wang F. Personalized predictive modeling and
risk factor identification using patient similarity. AMIA Jt Sum-
mits Transl Sci 2015;2015:132–6.
34. Panahiazar M, Taslimitehrani V, Pereira NL, Pathak J. Using EHRs
for heart failure therapy recommendation using multidimen-
sional patient similarity analytics. Stud Health Technol Inform
2015;210:369–73.
35. Wang B, Mezlini AM, Demir F, Fiume M, Tu Z, Brudno M, etal.
Similarity network fusion for aggregating data types on a
genomic scale. Nat Methods 2014;11:333–7.
36. Pinnock H, Barwick M, Carpenter CR, Eldridge S, Grandes G,
Griths CJ, etal. Standards for Reporting Implementation
Brought to you by | Kando Muszaki Foiskola
Authenticated
Download Date | 2/5/19 1:00 PM
Roca etal.: Paradigm changes for diagnosis: using big data for prediction 327
Studies (StaRI): explanation and elaboration document. BMJ
2017;7:e013318.
37. Peters DH, Adam T, Alonge O, Agyepong IA, Tran N.
Implementation research: what it is and how to do it. Br Med J
2013;347:2–7.
38. Gershon AS, Jafarzadeh SR, Wilson KC, Walkey AJ. Clinical
knowledge from observational studies. Everything you wanted
to know but were afraid to ask. Am J Respir Clin Care Med
2018;198:859–67.
39. Cano I, Alonso A, Hernandez C, Burgos F, Barberan-Garcia A,
Roldan J, etal. An adaptive case management system to sup-
port integrated care services: lessons learned from the NEXES
project. J Biomed Inform 2015;55:11–22.
Article note: Lecture given by Prof. Josep Roca at the 2nd EFLM
Strategic Conference, 18–19 June 2018 in Mannheim (Germany)
(https://elearning.eflm.eu/course/view.php?id=38).
Brought to you by | Kando Muszaki Foiskola
Authenticated
Download Date | 2/5/19 1:00 PM
