key: cord-0319928-axtzw4jw
authors: Lichtner, G.; Jurth, C.; Alper, B. S.; Spies, C.; Boeker, M.; Meerpohl, J. J.; von Dincklage, F.
title: Representation of evidence-based clinical practice guideline recommendations on FHIR
date: 2022-05-16
journal: nan
DOI: 10.1101/2022.05.16.22275120
sha: 445943df8c7d72a9fc31db68848497d1acf2ceee
doc_id: 319928
cord_uid: axtzw4jw

Background Various formalisms have been developed to represent clinical practice guideline recommendations in a computer-interpretable way. However, none of the existing formalisms leverage the structured and computable information that emerge from the evidence-based guideline development process. Thus, we here propose a FHIR-based guideline representation format that is structurally aligned to the knowledge artifacts emerging during the process of evidence-based guideline development. Methods We identified the information required to represent evidence-based clinical practice guideline recommendations and reviewed the knowledge artifacts emerging during the evidence-based guideline development process. Then we conducted a consensus-based design process with domain experts to develop an information model for guideline recommendation representation that is structurally aligned to the evidence-based guideline recommendation development process and a corresponding representation based on evidence-based medicine (EBM)-on-FHIR resources. Results The information model of clinical practice guideline recommendations and its EBMonFHIR-based representation contain the clinical contents of individual guideline recommendations, a set of metadata for the recommendations, the ratings for the recommendations (e.g., strength of recommendation, certainty of overall evidence), the ratings of certainty of evidence for individual outcomes (e.g., risk of bias) and links to the underlying evidence (systematic reviews based on primary studies). We created profiles and an implementation guide for all FHIR resources required to represent a complete clinical practice guideline and used the profiles to implement an exemplary clinical guideline recommendation. Conclusions Our EBMonFHIR-based representation of clinical practice guideline recommendations allows to directly link the evidence assessment process through systematic reviews and evidence grading, and the underlying evidence from primary studies to the resulting guideline recommendations. This not only allows to evaluate the evidence on which recommendations are based on transparently and critically, but also allows for a more direct and in future automatable way to generate computer-interpretable guideline recommendations based on computable evidence.

Background 25 Various formalisms have been developed to represent clinical practice guideline recommendations in 26 a computer-interpretable way. However, none of the existing formalisms leverage the structured and 27 computable information that emerge from the evidence-based guideline development process. Thus, 28

we here propose a FHIR-based guideline representation format that is structurally aligned to the 29 knowledge artifacts emerging during the process of evidence-based guideline development. 30

We identified the information required to represent evidence-based clinical practice guideline 32 recommendations and reviewed the knowledge artifacts emerging during the evidence-based 33 guideline development process. Then we conducted a consensus-based design process with domain 34 experts to develop an information model for guideline recommendation representation that is 35 structurally aligned to the evidence-based guideline recommendation development process and a 36 corresponding representation based on evidence-based medicine (EBM)-on-FHIR resources. 37

The information model of clinical practice guideline recommendations and its EBMonFHIR-based 39 representation contain the clinical contents of individual guideline recommendations, a set of 40 metadata for the recommendations, the ratings for the recommendations (e.g., strength of 41 recommendation, certainty of overall evidence), the ratings of certainty of evidence for individual 42 outcomes (e.g., risk of bias) and links to the underlying evidence (systematic reviews based on primary 43 studies). We created profiles and an implementation guide for all FHIR resources required to represent 44 a complete clinical practice guideline and used the profiles to implement an exemplary clinical 45 guideline recommendation. 46

Conclusions 47 Our EBMonFHIR-based representation of clinical practice guideline recommendations allows to 48 directly link the evidence assessment process through systematic reviews and evidence grading, and 49 the underlying evidence from primary studies to the resulting guideline recommendations. This not 50 only allows to evaluate the evidence on which recommendations are based on transparently and 51 critically, but also allows for a more direct and in future automatable way to generate computer-52 interpretable guideline recommendations based on computable evidence. 53

Introduction 54 A plethora of approaches have been developed to specify computer-interpretable representations of 55 clinical practice guideline recommendations, such as Asbru, EON, GLIF3, SAGE or GUIDE [1] [2] [3] [4] [5] [6] [7] . Despite 56 their many differences, all these representations have in common that they are designed based on the 57 concept that computer-interpretable guideline recommendations are derived by translating 58 unstructured recommendations into the formalism of the respective representation. However, with 59 the emergence of evidence-based medicine over the last decades [8] , guideline recommendations are 60 developed in a structured process in which knowledge artifacts, such as the effect size estimates from 61 primary studies or the grading of available evidence, are derived at each step. Leveraging these 62 structured knowledge artifacts for the computer-interpretable representation of clinical guideline 63 recommendations might require a reconsideration of the current computer-interpretable guideline 64 recommendation formalisms. 65

The systematic development process of evidence-based guideline recommendations for a specific 66 clinical question is based on a systematic review of studies providing data that address the clinical 67 question, potentially followed by aggregations of the study data in meta-analyses and finally balancing 68 and grading all available information in structured evidence-to-decision frameworks, resulting in 69 trustworthy recommendations [9-12]. With digitalization spreading, more and more structured data 70 becomes available from this process. Considering that this structured data is the basis to formulate 71 unstructured guideline recommendations, translating these unstructured guideline recommendations 72 back into structured recommendation representations induces unnecessary workload. As an 73 alternative approach, we propose a format to represent computer-interpretable guideline 74 recommendations in a structure that directly builds on the knowledge artifacts emerging during the 75 guideline recommendation development process. 76

To represent such knowledge artifacts that emerge during guideline recommendation development, 77 the HL7 Clinical Decision Support (CDS) Work Group-sponsored Evidence-based Medicine (EBM) on 78

Fast Healthcare Interoperable Resources (FHIR) project sub-work group (EBMonFHIR) created a 79 collection of FHIR resources [13] . These FHIR resources provide a standardized way of describing data 80 formats and elements that are related to both the evidence generation and evidence assessment parts 81 of evidence-based guideline development, such as the effect size for certain outcomes of a specific 82 intervention on a specific population, the grading of the certainty of evidence, and metadata such as 83 the group of authors or publication status and version of knowledge artifacts. 84

Based on these resources, we developed an interoperable, computer-interpretable representation of 85 clinical practice guideline recommendations that combines the structured data emerging during the 86 development of evidence-based guideline recommendations, from the evidence generation in primary 87 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 16, 2022. ; https://doi.org/10.1101/2022.05.16.22275120 doi: medRxiv preprint studies and systematic reviews and the structured evidence assessment in evidence-to-decision 88 frameworks to the resulting guideline recommendations. 89 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) To establish which information about the contents and the metadata of clinical practice guideline 92 recommendations need to be captured in the FHIR-based representation, we performed an iterative 93 consensus-based requirements engineering process. In this process, we included five clinical 94 stakeholders (health care professionals) and five guideline developers from German university 95 hospitals, medical societies and Cochrane Germany to identify both the required information for 96 practical use of clinical guideline recommendations and the metadata that is required to assess e.g. 97 the credibility and strength of recommendations, as well as the information required to connect 98 individual recommendations to their underlying evidence from systematic reviews of primary studies. 99

Participants of the requirements engineering process were recruited from the members of the COVID- In close collaboration with the EBMonFHIR maintainers, we derived a mapping of the information 116 model to existing EBMonFHIR resources. During this process, suboptimal applicability of the resources 117 to our use case were resolved by introducing differential changes to the EBMonFHIR resources in the 118 FHIR specification. We therefore developed our FHIR profiles and implementation guide based on the 119 latest available FHIR development build, which was at the time of writing the daily continuous 120 integration build of FHIR R5 as of April 1, 2022. 121 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) we required that at least one instance (i.e., example) was defined that instantiated that profile. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The ArtifactAssessment resource is newly introduced in FHIR Release 5 and is used to represent one or 254 more comments, ratings, or classifications about another resource. Here, that other resource is the 255

RecommendationPlan profile that describes the guideline recommendation. The profile includes the 256 individual ratings as justification for the recommendation in the content field and uses different value 257 sets and code systems for the individual components (Table 2) . 

RatingBenefitAndHarms EvidenceToDecisionCertaintyRating CEOsys small-net-benefit | substantial-net-benefit |important-harms

RatingPreferenceAndValues EvidenceToDecisionCertaintyRating CEOsys factor-not-considered |substantial-variability |no-substantial-variability | few-want-intervention

EvidenceToDecisionCertaintyRating CEOsys factor-not-considered |important-issues-or-not-investigated | no-important-issues | important-negative-issues

EvidenceToDecisionCertaintyRating CEOsys factor-not-considered |important-issues-or-not-investigated | no-important-issues | intervention-poorly-accepted

EvidenceToDecisionCertaintyRating CEOsys factor-not-considered |important-issues-or-not-investigated | no-important-issues | intervention-difficult-toimplement Risk of Bias RatingConcernDegree EvidenceCertaintyRating FHIR no-concern | serious-concern | very-serious-concern | extremely-serious-concern

RatingConcernDegree EvidenceCertaintyRating FHIR no-concern | serious-concern | very-serious-concern | extremely-serious-concern

RatingConcernDegree EvidenceCertaintyRating FHIR no-concern | serious-concern | very-serious-concern | extremely-serious-concern Imprecision RatingConcernDegree EvidenceCertaintyRating FHIR no-concern | serious-concern | very-serious-concern | extremely-serious-concern Publication Bias RatingConcernDegree EvidenceCertaintyRating FHIR no-concern | serious-concern | very-serious-concern | extremely-serious-concern Large Effect RatingUpratingTwoLevels EvidenceCertaintyRating FHIR no-change | upcode1 | upcode2

Plausible Confounders RatingUpratingOneLevel EvidenceCertaintyRating FHIR no-change | upcode1

RatingUpratingOneLevel EvidenceCertaintyRating FHIR no-change | upcode1 

The definition of an individual outcome (e.g., 30-day all-cause mortality), possibly including parameters 279 such as timing and method of outcome measurement, is represented using the OutcomeDefinition 280 profile of EvidenceVariable. 281

The evidence for individual outcomes, i.e., summary statistics from systematic reviews of clinical 282 studies or from primary studies, are represented by an instance of the StudyOutcomeEvidence profile 283 (for primary studies) or OutcomeEvidenceSynthesis (for systematic reviews) of the Evidence resource. 284

These instances hold the statistics associated with the particular outcome (e.g., the relative risk for 285 intervention vs. comparison and the associated 95% confidence interval). For that purpose, 286 To validate our proposed guideline recommendation structure, we implemented a recent 329 recommendation from a guideline for the treatment of COVID-19 intensive care patients [27-29]. As 330 this recommendation is part of a living guideline, it is ideally suited to validate the agility of our 331 proposed guideline recommendation representation as new evidence is added or certainty ratings for 332 specific outcomes or the actual recommendations changed. We selected a recommendation that 333

included both a treatment that should be performed on a specific population of patients and one that 334

should not be performed on another population of patients. Specifically, the guideline includes a 335 recommendation for treating ventilated COVID-19 patients with the systemic corticosteroid 336

Dexamethasone, applied for 10 days orally or intravenously, and a recommendation against treating 337 non-ventilated patients with any Dexamethasone. The recommendation is based on a systematic 338 review [30] , which is part of the evidence generation process. The different outcomes, most 339 importantly all-cause mortality, was evaluated as part of the evidence assessment process and the 340 recommendations were formulated based on the evaluated evidence (Figure 3) . We created instances 341 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) We here present an EBMonFHIR-based representation for computer-interpretable guideline 350 recommendations that leverages the structured data emerging during the development of evidence-351 based guidelines, ranging from primary evidence-generating studies over systematic reviews and the 352 evidence assessment using evidence-to-decision frameworks to the formulated recommendation. We 353 have developed the representation based on an iterative, consensus-based requirements engineering 354 process and mapping of the thereby derived information model items to EBMonFHIR resources. We 355 developed profiles for all used FHIR resources and created an implementation guide for the guideline 356 representation format. To evaluate the format, we implemented a recent guideline recommendation 357 in our representation format. 358

The here presented computer-interpretable representation of the whole evidence-based guideline 359 development process offers a set of advantages over just representing the final guideline 360 recommendations: 361

First, representing the evidence from primary studies and systematic reviews (i.e., effect size statistics) 362

and evidence-to-decision process artifacts in a computer-interpretable way allows them to be used for 363 semi-automated guideline recommendation formulation. Indeed, knowledge artifacts from evidence 364 generation and evidence assessment (i.e., ratings of existing evidence) in our representation format 365 may be published in repositories, e.g. on collaborative guideline development platforms such as 366

MAGICapp [32], and be reused for different guideline recommendation development processes. 367

Second, the representation of the evidence generation and evidence assessment artifacts may be 368 particularly valuable during the lifecycle of living guidelines that are regularly updated as new evidence 369 is published or existing evidence is re-evaluated in light of new findings. 370

Third, integrating a complete guideline recommendation representation into clinical decision support 371 or recommendation monitoring systems allows to close the loop from evidence to recommendation 372 back to evidence: Currently, the information flow is unidirectional from evidence generation via 373 evidence assessment to the formulated recommendation. However, when these recommendations 374 are implemented in a hospital setting by means of automated integration with electronic health 375 records (EHR), the treatment of patients according or not according to the recommendation in 376 connection with appropriate outcome data generates real-world evidence for (or against) the 377 intervention. This evidence, readily representable in the here proposed guideline recommendation 378 representation format, can be evaluated and used in an update process of the recommendation. 379

Fourth, representation of the full process of evidence-based recommendation development allows the 380 target audience of the guideline recommendations a direct and transparent assessment of the 381 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 16, 2022. ; https://doi.org/10.1101/2022.05.16.22275120 doi: medRxiv preprint GRADE Evidence to Decision (EtD) frameworks: a systematic and transparent approach to making well 515 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 16, 2022. ; https://doi.org/10.1101/2022.05.16.22275120 doi: medRxiv preprint

Combining diagnosis and treatment using asbru

Modeling Data and Knowledge in the EON Guideline Architecture, 489 MEDINFO

GLIF3: a representation format for sharable computer-interpretable clinical practice 492 guidelines

The SAGE Guideline Model: 495 Achievements and Overview

Architectures and tools for innovative Health 498 Information Systems: The Guide Project

Computer-interpretable clinical guidelines: A methodological review

Ten years of knowledge representation for health care

Clinical Practice

Guidelines We Can Trust

Introduction-GRADE evidence profiles and summary of findings tables

Achieving 525 evidence interoperability in the computer age: setting evidence on FHIR

Utilization of the PICO framework to 539 improve searching PubMed for clinical questions

Defining certainty of net benefit: a GRADE concept paper

International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity

for the Laboratory LOINC Developers, LOINC

The Unified Code for Units of Measure

Empfehlungen zur stationären Therapie von Patienten mit COVID-19

Guideline group

Clinical Practice Guideline: Recommendations on Inpatient Treatment of Patients with COVID-19

Key summary of German national treatment guidance 569 for hospitalized COVID-19 patients : Key pharmacologic recommendations from a national German 570 living guideline using an Evidence to Decision Framework

The copyright holder for this preprint this version posted May 16, 2022. ; https://doi.org/10.1101/2022.05.16.22275120 doi: medRxiv preprint

evidence generation and evidence assessment process that underlies the recommendations. It has to 382 be noted that the structured evidence generation and evidence assessment artifacts are generated in 383 any case within the context of a structured guideline recommendation development process; 384 representing them in a computer-interpretable way therefore comes at no or only little further cost 385 during guideline recommendation development. 386Even though some of the existing guideline recommendation formalisms that only represent the final 387 guideline recommendations would allow to represent all of the information, the effort to newly encode 388it from scratch appears as a relevant barrier. In contrast, the focus of our formalism is to enable using 389 data generated during the guideline recommendation development process, thus reducing the 390 required effort to include the additional data. However, as each representation formalism has its own 391 advantages and specialized scope of functions, our proposed representation might complement 392 existing formalisms instead of substituting them. In that way, mapping the treatment recommendation 393 part of our representation format to other formalisms would allow to close the gap between these 394 formalisms and the knowledge artifacts emerging during evidence-based guideline recommendation 395 development. 396 Apart from the previously described differences in content, in contrast to most previous guideline 397 recommendation formalisms, we have based our representation on FHIR, which currently might be 398 considered the most important standard for defining interoperable medical data exchange, with 399 growing support from EHR software vendors. There has been considerable work done on representing is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2022. ; https://doi.org/10.1101/2022.05.16.22275120 doi: medRxiv preprintWe expect that in the maturation process of resources from the Clinical Decision Support group, which 451 maintains the PlanDefinition resource, and EBMonFHIR, additional concepts and relationships of 452 declarative process modelling will be introduced to completely represent relationships between 453 recommendations and guidelines. 454The proposed EBMonFHIR-based guideline recommendation representation is currently being 455 implemented in several German university hospitals, where it is used as a computer-interpretable 456 guideline recommendation representation to be automatically integrated with standardized clinical 457 patient data to provide information about individual patient guideline recommendation applicability 458 and adherence. In the context of this project, and as the EBMonFHIR resources evolve, our guideline 459 recommendation representation will be continuously updated and improved where necessary. 460

To leverage the structured, computable knowledge artifacts that emerge during evidence-based 462 guideline recommendation development, we have developed a FHIR-based guideline 463 recommendation representation that is aligned with these knowledge artifacts. Thereby, our 464EBMonFHIR-based representation of clinical practice guideline recommendations allows to directly link 465 the systematic evidence assessment and the underlying evidence from systematic reviews and primary 466 studies to the resulting guideline recommendations. This not only allows for a transparent and critical 467 evaluation of the evidence on which recommendations are based, but also provides a more 468 straightforward and, in the future, automatable way to generate computer-interpretable guideline 469 recommendations from the available evidence. 470 471 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2022. ; https://doi.org/10.1101/2022.05.16.22275120 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted May 16, 2022. ; https://doi.org/10.1101/2022.05.16.22275120 doi: medRxiv preprint