3. Results
3.1. CRI definition and categories
3.1.1. CRI definition and scope
In 2009, Embi and Payne proposed a definition of clinical research informatics (CRI)
as “the subdomain of biomedical informatics concerned with the development, application,
and evaluation of theories, methods, and systems to optimize the design and conduct
of clinical research and the analysis, interpretation, and dissemination of the information
generated” [[1],[4]] Defined as the intersection of the field of clinical research and the field of
biomedical informatics, CRI focuses on the development and evaluation of tools and
methods to support researchers in clinical research activities including study design,
patient recruitment, data collection, integration and analysis. This is further illustrated
in the conceptual framework of CRI proposed by Weng and Kahn ([Figure 2]) [[2]]
Figure 2. Clinical Research Informatics (CRI) conceptual framework [[2]].
Subsequently, Johnson proposed a search strategy using two conceptual axes defining
CRI and sharpening the boundaries with closely related fields such as computational
statistics and patient care informatics [[3]]
As noted by these authors and others, CRI approaches and capabilities have the potential
to play a prominent role in supporting widespread advances in medicine, healthcare,
and public health [[5]]
3.1.2. Methodological pillars and categories
Progress by the community of CRI researchers and developers can be depicted as advancing
along major pillars of the informatics sub-discipline such as data integration, information
extraction, data linkage, data quality assessment, phenotyping, data and text mining,
knowledge engineering, and health technology development and assessment (workflow
support, human computer interaction, evaluation) in order to reinforce methods and
tools supporting the different stages. The papers selected by the AMIA year-in-review
and IMIA Yearbook processes have been classified into other sub-categories related
to topics and/or steps of the clinical research lifecycle for which they provide solutions
[[3]] . [Table 2] summarizes the classification of the final list of selected papers identified via
the combined review processes utilized for the current review.
Table 2.
Categorization and sub-categorization of the 205 selected papers.
|
Categories
(Acronym, label, number of papers)
|
Sub categories
|
Corresponding stage of the clinical research study life cycle
|
Sources
|
|
SD-PR-EXE
|
Study design & executionPatient recruitment
|
n=43
|
Study Design (new CT designs)
|
CRI step 1: Analyzing Study Designs, feasibility studies
|
[[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]]
|
|
Patient Recruitment
|
CRI step 2: Getting Participants into Studies
|
|
Study Execution (CRI methods & systems, collaborative workflow systems, usability
and needs)
|
CRI step 3: Executing Studies
|
|
PTF
|
Large scale research platforms
|
n=12
|
Study Execution (large scale collaborative platforms, data networks)
|
[[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]]
|
|
DM
|
Study data management
|
n=59
|
Managing Study Data (focus real-world data/phenotyping)
(n=17)
|
CRI step 4: Managing Study Data
|
[[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]]
|
|
Managing Study Data (focus semantic interoperability, data integration and standardization)
(n=19)
|
[[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
[88]
[89]
[90]
[91]
[92]
[93]
[94]
[95]
[96]]
|
|
Managing Study Data (focus data quality assessment) (n=14)
|
[[97]
[98]
[99]
[100]
[101]
[102]
[103]
[104]
[105]
[106]
[107]
[108]
[109]
[110]]
|
|
Managing Study Data (focus security, confidentiality) (n=9)
|
[[111]
[112]
[113]
[114]
[115]
[116]
[117]
[118]
[119]]
|
|
MKD
|
Data/text mining
|
n=18
|
Using Study Data (data or text mining, data visualization)
|
CRI step 5: Using Study Data
|
[[120]
[121]
[122]
[123]
[124]
[125]
[126]
[127]
[128]
[129]
[130]
[131]
[132]
[133]
[134]
[135]
[136]
[137]]
|
|
COM-PUB
|
Results communication and publication
|
n=5
|
Communicating Study Results (Research results dissemination)
|
CRI step 6: Communicating Study Results
|
[[138]
[139]
[140]
[141]]
|
|
Analyzing Study Publications (knowledge representation, management, or engineering)
|
CRI step 7: Analyzing Study Publications
|
[[142]]
|
|
INFRA
|
Technical infrastructure
|
n=7
|
Architecture (Big Data, Cloud computing)
|
|
[[143]
[144]
[145]
[146]
[147]
[148]
[149]]
|
|
ETH-REG
|
Ethical & legal issues
|
n=11
|
Ethical issues
|
|
[[150]
[151]
[152]
[153]
[154]
[155]
[156]
[157]
[158]
[159]
[160]]
|
|
Legal, regulatory issues
|
|
|
SOC
|
Societal issues
|
n=27
|
Societal issues (focus governance, stakeholder participation, engagement)
|
|
[[161]
[162]
[163]
[164]
[165]
[166]
[167]
[168]
[169]
[170]
[171]
[172]
[173]
[174]
[175]
[176]
[177]
[178]
[179]
[180]
[181]
[182]
[183]
[184]
[185]
[186]
[187]]
|
|
Societal issues (focus patient participation, health literacy, patient consenting)
|
|
|
Societal issues (focus business model)
|
|
|
Societal issues (focus workforce, education and training)
|
|
|
LHS-SURV
|
Learning Health Systems Surveillance
|
n=10
|
Learning Health Systems, disease, drug & device surveillance
|
|
[[188]
[189]
[190]
[191]
[192]
[193]
[194]
[195]
[196]
[197]]
|
|
TRE
|
Trends
|
n=13
|
Trends
|
|
[[1]
[2]
[3]
[4]
[5],[198]
[199]
[200]
[201]
[202]
[203]
[204]
[205]]
|
When the literature is plotted by categories over time (as in [Figure 3]), a different visualization of the evolution of the CRI field is highlighted, with
different categories and topics varying by year.
Figure 3.
a. Number of papers shortlisted through both AMIA and IMIA review processes (n=327)
divided in the ten main categories over time.
b. Number of selected/shortlisted papers in the four sub-categories of data management
over time (n=90).
3.2. Main CRI contributions supporting the life cycle of clinical research (2013-2022)
Based on our combined analysis of CRI articles selected over time by the approaches
described above, this section provides an overview from 10 years of publications demonstrating
progress in the multifaceted aspects of CRI supporting research and innovation in
healthcare and biomedicine. Each of the sections that follow cover major categories
with exemplars highlighted within.
3.2.1. Study design, participant recruitment and study execution
To support the increasing complexities of clinical research over time, a number of
initiatives, including the Human Studies Database [[15]], the database for aggregate analysis of ClinicalTrials.gov (AACT) [[7]], the more recent knowledge base of clinical trial eligibility criteria [[36]], and others [[21],[35]] proposed access to clinical trials repositories allowing significant information
to be leveraged and even re-used across studies. In addition, initiatives emerged
that were aimed at optimizing clinical research deliverables while meeting the requirements
enunciated by the U.S. Food and Drug Administration for electronic systems [[8]]. Major contributions over the last decade also focused on so-called secondary use of health data collected via routine care for optimizing clinical trials [[6],[9],[22],[28]]. Indeed, many projects and initiatives have explored the value of EHR data for
clinical trial design, patient recruitment [[12]
[13]
[14],[17],[19],[25],[26],[30],[31],[33],[34],[37],[43],[180]], or for accelerating data collection [[11],[39]
[40]
[41]
[42]]. Facilitated by the generalization of clinical data warehouses, some initiatives
were carried out on a large scale within European projects such as the EHR4CR [[10],[16],[20],[23],[24],[29]] or TRANSFoRm [[27]] projects. The EHR4CR project contributed to the development of the InSite commercial
solution (formerly Custodix, Belgium, currently part of TriNetX) which reached the
evaluation phase [[32]].
While randomized controlled trials (RCTs) are the gold standard for estimating treatment
effects in medical research, there has also been growing interest in and use using
real-world data (RWD) for use-cases including drug development, outcomes research,
and observational studies. One such use case is the construction of external control
arms for evaluation of efficacy in single-arm trials, particularly in cases where
randomization is either infeasible or unethical [[44],[46],[47]]. In close alignment with a recently released draft framework by the Medical Device
Innovation Consortium on real-world clinical evidence and in vitro diagnostics, publications report the possibility of substantially reducing the length
and cost of diagnostic test evaluation for diseases with low prevalence to support
regulatory decision-making by leveraging subjects from a RWD source [[38]]. The past decade has also seen the explosion of observational research using EHR
data with CRI publications better illuminating our understanding of the specific caveats
and considerations/guidelines for conducting observational studies and comparative
effectiveness with RWD, while limiting biases [[18]].
3.2.2. Data management
Real-world data and Clinical Data warehouses
Clinical data warehouses (CDW), built early on at the level of healthcare facilities
(Enterprise Data Warehouses) enabled evidence generation based on real-world data
to become a reality during the last decade [[63],[64],[67]]. In the United States the CDW projects, supported by the Clinical and Translational
Science Award (CTSA) funding program of the National Institutes of Health (NIH), have
seen increasingly high levels of maturity [[62],[75]]. One successful example is the Informatics for Integrating Biology and the Bedside
(i2b2) solution that has been widely used by academic health centers in the U.S. and
internationally. Notable i2b2 use-cases include discovery research, case-control studies
[[66]], enable cost-effective genome-wide studies, identify important risks from commonly
used medications [[61]], capture and evaluate patient reported outcomes [[48]], access image from clinical archive [[69]], and enable data sharing at scale [[70],[79],[92]]. Hospital-based CDWs and multi-site data-sharing activities have also emerged in
Europe and Asia over the past decade [[65],[71],[72],[76]].
Interoperability and standards
Interoperability is another major pilar of CRI that has evolved. In the domain of
clinical research, the Clinical Data Information Standards Committee (CDISC) has developed
a number of standards for study design, data collection and analysis, and submission
to the regulatory bodies [[91]]. The Operational Data Model (ODM) has been updated to enable the implementation
of case report forms supporting more efficiently the complete clinical study lifecycle
from the design phase to the collection of patient level data [[86]]. The complexity and heterogeneity of healthcare data collection and storage approaches
has remained a motivating challenge, and there has been movement in the CRI domain
toward certain common data standards.
Several research networks focused on cross-institutional data analyzes at scale developed
so-called “common data models” (CDM) that require each participating system to transform
their underlining data model to the relevant CDM. Examples include the OMOP CDM from
the OHDSI consortium [[89],[94],[120]], the FDA Mini Sentinel's CDM, I2B2-SHRINE [[79]], STRIDE or SHARPn [[81],[143]]. New experience has been gained from these projects and the resulting models have
been analyzed and compared [[88]]. While some require transfer of data to a centralized solution, another approach
involves federated data and metadata management. One example is the ISO/IEC 11179
standard, that has been demonstrated to semantically link disparate common data elements
defined by different organizations [[80]]. Minimum data sets based on common data elements have been built for different
disease areas [[56],[83],[93]].
Furthermore, some initiatives have been conducted to bridge standards used in clinical
care with clinical research standards, often developed through parallel, and often
disparate, efforts. Other harmonization efforts have been conducted within the SHARPn
project in mapping Healthcare Clinical Element Models (CEMs) adopted by the Office
of the National Coordinator (ONC) to data elements extracted from the CDISC templates
[[85]]. More recently, relying on the increasing adoption of HL7's Fast Healthcare Interoperability
Resources (FHIR) by the healthcare industry, many studies have reported the interest
of FHIR integration profiles to facilitate the integration of care and research activities
[[39],[92],[95]]. The HL7 Vulcan acceleration program[1] develops FHIR resources needed to execute prioritized use cases of secondary use
of real-world data and especially EHR data. While existing information model standards
(e.g., ISO 13606, openEHR, HL7 FHIR, and CDISC) define the basic semantics of health data,
semantic interoperability are enabled by medical terminologies, nomenclatures, and
ontologies. It is the use of these terminologies through explicitly defined value
sets (terminology binding) that brings unambiguous semantics to health data. Consequently,
it remains challenging to implement the shift from the “syntactic” level to the “semantic”
level of data exchange enabling smart services and applications supporting communication
among clinicians, researchers, and healthcare providers. In this context, Semantic
Web technologies provide solutions for managing and sharing knowledge and for making
the health data FAIR (Findable, Accessible, Interoperable, and Reusable) [[87]]. In contrast to other major foundational models for clinical research informatics
(e.g., BRIDG, CDISC ODM/SDTM, HL7 FHIR, OMOP CDM), the Ontology of Clinical Research (OCRe)
takes a logic-oriented ontological modeling approach and attempt to model, not only
operational and administrative needs, but also study validity, confounding, and bias
needed for assessing study design strength [[15]]. Beyond the design of ontologies in healthcare (e.g., for epilepsy [[78]]), research in Semantic Web technologies also includes activities contributing to
CRI such as developing efficient ontology editors, and integrating ontologies into
healthcare information systems [[82],[84],[90],[96]]
Data quality assessment
Data quality (DQ) is essential to many CRI activities. Identifying the need for common,
standardized approaches to DQ, Kahn et al. were motivated to initiate a community-based effort in collaboration with the Electronic
Data Methods Forum (EDM) to harmonize terminologies in CRI [[99]]. This DQ framework has been extended beyond intrinsic aspects to cover technical,
and contextual categories across the data life cycle enabling the assessment and management
of RWD repositories to ensure fitness for purpose. An ontology for characterizing
DQ for secondary use of EHR data has been proposed [[98]]. An increasing number of studies describe DQ assessment methods or tools measuring
completeness of data items and other DQ dimensions [[100],[107],[109],[110]]. Two recent reviews summarize health DQ issues and assessment practices [[104],[108]]. Some initiatives involve patients to assess the quality of their clinical data
[[103]]. Medical centers [[102]] or large-scale research platforms [[101],[106]] are setting up continuous DQ evaluation and improvement programs.
With the increasing use of observational, non-experimental data for various purposes,
the potential for introducing biases and confounding factors that are often hidden
in the data must be addressed or at least carefully considered, such as when conducting
RWD studies or developing and validating predictive models [[74],[97],[105]]. Recent retractions of articles published in high-profile journals reporting COVID-19
studies based on EHR demonstrate that conforming to best practices in developing robust
research based on RWD is critical to promote and foster rigor, quality, and reliability
of this rapidly growing field [[73],[140]].
Security and data privacy-enhancing techniques
New research opportunities as well as artificial intelligence and machine learning
techniques raise the need for greater data access but also bring new potential risks
to privacy [[113]]. The evaluation of this risk in releasing datasets is a major concern [[111],[112],[114]]. Many studies focus on de-identification methods and privacy-enhancing techniques
[[111],[115],[119]], including homomorphic encryption techniques enabling federated analysis of RWD
while complying with data protection requirements [[117]]. An increasing number of papers propose methods for generating synthetic health
data [[116],[118]].
Large scale research platforms
Many informatics platforms enabling collaborative research using multi-institutional
distributed heterogenous data continued to emerge during the last decade [[49]]. Large-scale data-sharing initiatives are developing at national level. As examples,
in the U.S. the SHARPn project that has proposed a scalable and standards-driven infrastructure
for secondary use of EHR data [[143]]. SHRINE implementations have been used for nationally scalable multi-site disease
studies of autism co-morbidity, juvenile idiopathic arthritis, peripartum cardiomyopathy,
colorectal cancer, diabetes, and others [[79]]. The PCORnet initiative links multiple sub-networks via adherence to a common data
model. Taken together, PCORnet sites include clinical data from a cohort of 80 million
patients and enable rapid access to patient populations for pragmatic clinical trials,
epidemiological research, and patient-centered research on both common presentations
and rare diseases [[53],[58]].
In Europe, the Health Data Hub in France and the Medical Informatics Initiative in
Germany are major national initiatives supporting secondary uses of health data [[72],[77]]. In U.K., we can cite two major initiatives, the Clinical Practice Research Datalink
(CPRD) [[51]] and the ClinicAl disease research using LInked Bespoke studies and Electronic health
Records (CALIBER) programs enabling secondary use of nationwide big data from linked
electronic health records to improve outcomes in respectively primary care and cardiovascular
diseases [[57]].
At a broad international scale, the Observational Health Data Sciences and Informatics
(OHDSI) initiative installed the OMOP common data model as a de facto standard for observational research based on large-scale real-word data and demonstrated
the value of large-scale data sharing in the analysis of drug use [[89]]. Finally, during the Covid19 pandemic, national Covid databases were implemented
(for example in US (e.g. NC3) [[60]], Germany or Spain [[93]]) and international data-sharing platforms were set up (such as 4CE [[149]] or SCOR [[59]] projects). In addition, CRI efforts resulted in practical implementations of biobank
research systems [[50],[54]].
3.2.3. Data/text mining, artificial intelligence/machine learning, knowledge discovery
One of the most striking findings in CRI over the last decade is the exceptionally
rapid development of highly flexible, reusable tools developed to explore, visualize
and analyze complex data sets and enable training of artificial intelligence (AI)
models that are likely to transform medicine. Growing access to both vast data sets
and advanced computational capabilities are enabling the rise of AI technologies and
methods for diagnostic or therapeutic decision-support, and other approaches of relevance
to CRI. Many studies report the use of machine learning in pharmacovigilance or in
precision medicine to predict outcomes such as hospital admissions or in-hospital
mortality for specific populations. A significant research effort has emerged to enable
federated learning of predictive models from federated data sets [[127]]. Particular attention is also paid to methodological and organizational issues
related to studies using EHR data and to propose methods for bias reduction in studies
using EHR data [[128],[131],[133],[140]].
New infrastructures and methods for data integration/fusion, record linkage, data
and text mining have enabled new data discovery opportunities. Significant research
efforts in the field focused on the development and validation of phenotyping methods.
Rule-based phenotype definitions are collaboratively developed and evaluated within
initiatives such as eMERGE [[121]] or PheKB[2] [[122]] in U.S. or the CALIBER platform[3] [[57]], and the phenotype health data gateway[4] in U.K. Many studies exploit the advances in natural language processing (NLP) to
extract clinical information from clinical texts [[120],[126],[135]]. Others have implemented and promoted innovative approaches for high-throughput
phenotyping [[81],[120],[123]
[124]
[125],[129],[130],[132],[134]].
Study results communication and publication
Although of undeniable interest in the generation of evidence, especially in areas
where traditional clinical trials would be unethical or infeasible, RWD should be
reused considering the highest research standards and updated guidelines such as the
methodological reference for conducting observational studies developed by European
Network of Centers for Pharmacoepidemiology and Pharmacovigilance, ENCePP () under
the aegis of the EMA (European Medicines Agency) and the FDA (Food and Drug Administration)
or by research communities [[140],[141]]. For example, the CONSORT-AI (Consolidated Standards of Reporting Trials-Artificial
Intelligence) extension and its companion statement for clinical trial protocols:
SPIRIT-AI (Standard Protocol Items: Recommendations for Interventional Trials-Artificial
Intelligence) from the EQUATOR network provide recommendations (checklist items) promoting
transparency and completeness in reporting clinical trials for AI interventions [[205]]. These recommendations assist editors and peer reviewers, as well as the general
readership, to understand, interpret and critically appraise the quality of clinical
trial design and risk of bias in the reported outcomes. In the radiology domain, DECIDE-AI
is a new reporting guideline for artificial intelligence studies in radiology. Some
other specific concerns of study research communication are e.g., the increase of
self-citation [[139]] or tracking and efficiently displaying research evidences [[138],[142]].
Technical infrastructure
In the current era of massive-scale digitalization of data and computationally-intensive
quantitative data analytics, the CRI community has developed new infrastructure capacities
that can process, analyze and store petabytes of health data [[145],[146],[148]]. The management of massive, unstructured and heterogeneous data is still challenging,
especially for supporting the identification of environmental exposures as determinants
of physio pathological processes and providing a coherent framework for dealing with
multi-scale population data including the phenome, genome, exposome, and their interconnections
in the context of the One Health paradigm [[65],[144],[147]].
Ethical, legal and societal issues (data governance, patient engagement, sustainability)
A major lesson that the coronavirus disease 2019 (COVID-19) pandemic has taught the
scientific CRI community is that healthcare institutions shall continue to strengthen
their expertise in data driven evidence generation. Since new opportunities also bring
new societal challenges, most of the institutions in charge of CRI and/or CDW for
research put in place governance committees of varying forms in charge of setting
up the policies of data requests for internal research projects and/or external data
sharing agreements as well as reviewing these requests [[162]]. These governance bodies need to guarantee that the intensive data-driven research
is conducted in respect of the law and in line with ethical principles (e.g., that data release fits the purpose of the research), especially in the context of
large-scale collaborative research. Ethical and regulatory issues related to the use
of large-scale linked data and new AI technologies are especially challenging and
need to be addressed to respond to the current rapidly changing data-driven agenda.
Researchers have to understand complicated and sometimes contradictory legal requirements
inherent to big data projects and to consider ethical obligations in order to balance
potential of discovery with legal and ethical considerations [[151],[152],[154],[159],[160]]. Ethics review committees may have to be reformed to deal with big data research
and improve their oversight capacity [[156],[158]] and supportive tools (eIRB systems) could be useful in this context. Some publications
focus on lessons learned and recommendations related to ethical, legal and societal
issues in the specific context of the covid-19 pandemics [[157]] or rare diseases. In Europe, the European Institute for Innovation through Health
Data (i~HD) plays a strong role in many national and international collaborative projects
or initiatives involving academic research groups from all over Europe collaborating
with global pharmaceutical companies in supporting the ecosystem in especially addressing
ethical, legal, societal issues [[177]]. Key national initiatives are promoting CRI [[161],[166],[171],[172],[178],[179],[181]]. Significant research efforts in the CRI field also focused on specific societal
issues : patient information and consent [[150],[153],[155],[168],[182]], patient involvement in the design and execution of data-intensive medical research
and clinical trials [[164],[170],[175],[176],[183]], health equity in CRI [[180],[184],[186]], sustainability [[68],[163]], the role of IT vendors [[167],[185]], barriers to data sharing [[165],[173]] and more generally educational issues [[169],[174]].
Learning Health System and large-scale surveillance
CRI contributions have also advanced support for the realization of a learning health
system that enables healthcare systems to objectively and continuously monitor their
practices, pragmatically compare the outcomes of care over time, including patient-reported
outcomes, and simulate and evaluate the impacts of certain strategic decisions on
organizations or the quality of care [[191],[192],[194],[196]]. Major research efforts also focused on data-driven approaches for signal detection
in pharmacovigilance especially in post marketing drug and vaccine safety surveillance
and more generally patient safety [[188]
[189]
[190],[193],[195]].
3.3. Recent (last two years) research trends and perspectives (2022-2023)
3.3.1. Health technology assessment, maturity of CRI services and innovative clinical
trial designs
Despite considerable progress over the past 10 plus years, certain challenges remain
a focus for CRI now and into the future. CRI interventions still appear too often
as anecdotes or quasi-experimental studies, whereas methodologically rigorous studies
(e.g., randomized, controlled studies) of CRI interventions remain rare. As the field matures,
more efforts will be needed to demonstrate the successes and value of CRI interventions.
Moreover, there is a need to enhance the level of formalization of CRI services in
large academic centers. Ensuring high-quality research relies on multiple factors including variations in organizational culture of academic
medical centers, retention of research workforces, efficient multi-departmental organizational
collaborations (e.g., relationship between biomedical research expertise and enterprise
IT), technical skills, operational efficiencies and transparent communication of standard
policies, procedures and key performance indicators [[199]].
Another emerging concern relevant to CRI is the development, validation, and evaluation
of AI-based healthcare technology. In the context of medical AI, that survived several “winters” since the 1970's,
there is a need to better understand how new AI technologies will challenge current
strategies for regulating and validating AI devices for medicine and research. Recent
reviews focusing on the development of AI in medicine state that AI could transform
clinician workflows, patient care, administrative tasks, and augment medical knowledge
and decision making [[203],[204]]. They also underline unsolved challenges, from issues about training ML systems
to unclear accountability making AI's implementation difficult and largely ununderstood
by healthcare professionals still struggling to implement AI in their daily practice.
Guidelines for clinical trial protocols for interventions involving AI are available
and some papers propose AI implementation frameworks and approaches for AI development,
implementation, use, monitoring and governance [[131],[133],[198]].
As the regulatory environment becomes progressively receptive to utilizing real-world
evidence in clinical trials, many publications focus on various innovative hybrid clinical trial settings. Of particular interest is the use of RWD to accelerate and perhaps even approximate
certain traditional research approaches. Among innovative trial designs, platform
trials including multiple trial arms (conducted simultaneously and/or sequentially)
on different treatments under a single master protocol. The capabilities of medical
centers to perform EHR-based protocol feasibility assessment, clinical site selection,
and patient pre-screening are keys for successful platform trials [[45]]. An increasing number of initiatives, such as the TAES (TriAl Eligibility Surveillance)
system, leverage existing real-world data, using artificial intelligence and standard
data models, to enhance identification of patients potentially eligible for clinical
trials and generate timely notifications to physicians and simultaneously decrease
the burden on research teams of manual EHR review [[43]]. Target trial emulation can be used for eligibility determination [[200]] and also to assess agreement between clinical trials and analogue real-world evidence
studies. Concordance in results vary depending on the agreement metric. Emulation
differences, chance, and residual confounding can contribute to divergence in results
[[201]].
3.3.2. Translational research, better integration of care and research and Learning
Health Systems
There is a need to promote organizational and technical interoperability between care
and research domains towards the implementation of “clinical trials as care option”. There are an increasing number of initiatives aiming at avoiding double data entry
through the integration of electronic case report forms (eCRF) systems with electronic
health records (EHR) – the “eSource” or “EHR2EDC” use case. While technical capabilities are important, organizational priorities, structure,
and the site's support of clinical research functions are equally important considerations.
According to a recent study assessing site readiness for eSource, only 21% of sites
were using Fast Healthcare Interoperability Resources (FHIR) standards to exchange
patient data with other institutions. Respondents generally gave lower readiness for
change ratings to organizations that did not have a separate research information
technology group and where researchers practiced in hospitals not operated by their
medical schools [[41]]. Although a recent study demonstrated that secure, consistent, and automated end-to-end
data transmission from the treating physician to the regulatory authority was feasible,
the industry-wide implementation of EHR2EDC still requires policy decisions that set
the framework for the use of research data based on routine care data [[42]]. To overcome the difficulty of integration with EHR systems due to information
security and data protection restrictions, a novel architecture for a federated electronic
data capture system (fEDC) has been proposed [[202]].
Despite recent efforts, especially those led by HL7 organization, for example the
HL7 FHIR Vulcan accelerator, formal representation of multimodal and multi-level data supporting data interoperability
across clinical research and care domains is still challenging. Additional efforts
are required from standard development organizations (CDISC, HL7 and DICOM) to better
align the standards used in care and research domain and from the vendors of clinical
and biomedical research applications to implement interoperability by design in the
development of their solutions within a “single source” approach. Common data models
solve many challenges of standardizing electronic health records (EHR) data but are
unable to semantically integrate all of the resources [[96]]. Moreover, semantic interoperability enabling secondary use of clinical data requires
a rigorous governance process to ensure internationally the quality of the data standardization
process consistently across care and research domains. Recognizing that interoperability
frameworks must be continuously adapted to the user's needs and that the update can
hardly be fully automatized, a clear collaborative process is needed to efficiently
support the creation of new semantic resources scoped to any additional use case of
biomedical research.
The persistent challenges posed by data collection based on paradigms of primary use of data vs. secondary use remain a threat to achieving high quality/value data at the source. Currently considerable
resources are dedicated to secondary use of health data and to complex pipelines curating
and enriching data outside of the scope of patient care. Efforts to promote improved
data quality by design in clinical settings will result in higher quality data for
research and evidence generation. For ethical reasons and for supporting the implementation
of the Learning Heath System, ensuring the quality as well as the privacy-preserving accessibility of clinical
data for primary use should be considered a priority with major impacts on secondary
use and the research agenda.
3.3.3. Trends related to the technological pillar
Progress is being made in data quality management systems standardizing medical data extraction and quality control methods [[108]
[109]
[110]]. The development of methods and tools to promote the use of textual data and advanced NLP methods in real-world observational studies is still an active research area with important
contributions from such groups as the NLP Working Group at the Observational Health
Data Sciences and Informatics (OHDSI) consortium[[135]]. Major research efforts are also devoted to phenotyping. Clinical experts and computer scientists still experience a variety of challenges
when building phenotypes, including challenges to discerning key clinical events,
clinical reasoning, and temporal elements [[136]]. Standard representations for phenotypes are increasingly being promoted to support
the creation, localization and sharing of highly specific phenotype algorithms [[96],[137]]. Additional research is also needed for building secure and efficient infrastructures
for “Big data” management supporting, for instance, federated machine learning on
distributed data sets hosted on sovereign cloud environments [[127]].
3.3.4. Trends related to the ethical, legal and societal issues (ELSI)
Ethical, legal and societal research continues to be needed to define and ensure an
ethically-guided, fair and patient-centric approach for CRI. Such work is essential
to better understand and solve practical issues related to potential biases and inequities
arising from the use of RWD and AI [[186],[187]]. Specifically, strategies for appropriate constituent engagement are essential
to ensure that solutions like AI can meaningfully benefit patients and other end users.
As progress continues toward a new era of digital health driven by cloud data storage,
distributed computing, and machine learning, key stakeholders must collaborate to
ensure research findings and their downstream implementations are inclusive, representative
and accessible for all who should benefit, and work in CRI is key to this.
