Next generation terminology infrastructure to support interprofessional care planning

Journal of Biomedical Informatics 75 (2017) 22–34

Contents lists available at ScienceDirect

Journal of Biomedical Informatics journal homepage: www.elsevier.com/locate/yjbin

Next generation terminology infrastructure to support interprofessional care planning

MARK

⁎

Sarah Collinsa,b,c, , Stephanie Klinkenberg-Ramireza, Kira Tsivkina, Perry L. Mara,b,c, Dina Iskhakovaa, Hari Nandigama, Lipika Samalb,c, Roberto A. Rochaa,b,c a b c

Partners HealthCare System, Boston, MA, United States Brigham and Women’s Hospital, Boston, MA, United States Harvard Medical School, Boston, MA, United States

A R T I C L E I N F O

A B S T R A C T

Keywords: Interprofessional care planning Care coordination Terminology management Information modeling

Objective: Develop a prototype of an interprofessional terminology and information model infrastructure that can enable care planning applications to facilitate patient-centered care, learn care plan linkages and associations, provide decision support, and enable automated, prospective analytics. Design: The study steps included a 3 step approach: (1) Process model and clinical scenario development, and (2) Requirements analysis, and (3) Development and validation of information and terminology models. Results: Components of the terminology model include: Health Concerns, Goals, Decisions, Interventions, Assessments, and Evaluations. A terminology infrastructure should: (A) Include discrete care plan concepts; (B) Include sets of profession-specific concerns, decisions, and interventions; (C) Communicate rationales, anticipatory guidance, and guidelines that inform decisions among the care team; (D) Define semantic linkages across clinical events and professions; (E) Define sets of shared patient goals and sub-goals, including patient stated goals; (F) Capture evaluation toward achievement of goals. These requirements were mapped to AHRQ Care Coordination Measures Framework. Limitations: This study used a constrained set of clinician-validated clinical scenarios. Terminology models for goals and decisions are unavailable in SNOMED CT, limiting the ability to evaluate these aspects of the proposed infrastructure. Conclusions: Defining and linking subsets of care planning concepts appears to be feasible, but also essential to model interprofessional care planning for common co-occurring conditions and chronic diseases. We recommend the creation of goal dynamics and decision concepts in SNOMED CT to further enable the necessary models. Systems with flexible terminology management infrastructure may enable intelligent decision support to identify conflicting and aligned concerns, goals, decisions, and interventions in shared care plans, ultimately decreasing documentation effort and cognitive burden for clinicians and patients.

1. Introduction Care coordination requires communicating and tracking of clinical states, such as health concerns, health goals, care decisions, decision rationales, care delivered, outcomes of care, and continuous evaluation of outcomes [1]. Communication and tracking of complex patient care requires care coordination tools that surface summaries of patients’ clinical states to display linked, complementary, and conflicting health care concerns, goals, decisions, and evaluations. The current features of Electronic Health Records (EHR) are insufficient to handle the agile linking and association of coded clinical concepts across clinical professions. Traditionally, documentation of care planning by nurses was

⁎

implemented separately from care planning by physicians. Currently, most vendor-based EHRs cannot associate the medical problem of Congestive Heart Failure with the nursing problem of Impaired Gas Exchange. In many EHR systems these data are stored in separate modules, limiting the ability to establish necessary linkages or relationships. Such technical constraints serve as significant barriers to designing dynamic summaries of care that assist clinicians in understanding available data in context and in relation to linkages with other data, particularly for longitudinal care planning. Yet, as care models have evolved to patientcentered models of care planning, EHRs struggle to reconfigure shared care plans. One important limitation is the underlying terminology infrastructure within EHRs that are designed to be profession-specific

Corresponding author at: Harvard Medical School & Brigham and Women’s Hospital, Brigham Circle, 1620 Tremont Street, Suite OBC-3-002D, Boston, MA 02120-1613, United States. E-mail address: [email protected] (S. Collins).

http://dx.doi.org/10.1016/j.jbi.2017.09.007 Received 23 February 2017; Received in revised form 15 September 2017; Accepted 17 September 2017 Available online 20 September 2017 1532-0464/ © 2017 Elsevier Inc. All rights reserved.

Journal of Biomedical Informatics 75 (2017) 22–34

S. Collins et al.

The collaborative and cooperative activity of documenting on a shared care plan introduces interesting dynamics in that the documenting clinician may not be the direct beneficiary of the information in the future and differing perceptions of responsibility and rewards for completing documentation may exist [12]. HL7 defines care planning documentation as: (a) consensus-driven with prioritized concerns, goals, and planned interventions, (b) a blueprint to organize and guide care and integrate multiple interventions proposed by multiple providers and disciplines for multiple conditions, (c) an artifact that reconciles and resolves conflicts between various plans of care and treatment plans, and (d) a source of truth for a longitudinal coordination of care [15]. Clinical Document Architecture (CDA) representations aligned with the HL7 V3 CP DAM represent static exchanged care plans as a snap shot in time and do not represent dynamic care team participations or reconciliation of data [15]. Reconciliation of the care plans across professions, encounters and content areas, such as health problems/ concerns (which include allergies/intolerances), goals and interventions (which include medications) are critical to achieving coordinated care [15]. The HL7 Care Coordination Services (CCS) functional model does address team coordination actions (e.g., find, create, associate, change, close, read, share, synchronize, and publish) for care plans. However, neither of these HL7 models address mechanisms for reconciliation of care plan concepts at the terminology level. A care plan by design is a collaborative, shared and dynamic structure [15] and requires a comprehensive clinical ontology to handle the representation of interprofessional terminology concepts and modeling requirements for reconciling at the concept level to enable dynamic, shared, and consistent care plans across the continuum of care [9,10,16,17]. In this paper we add to the set of storyboards used to define the HL7 V3 CP DAM by defining new clinical care planning use cases, representing the terminology concepts for those use cases, and identifying the terminology infrastructure requirements to support dynamic reconciliation of data for those use cases.

and, consequently, unable to handle more dynamic interprofessional terminology requirements. EHRs need to implement a new approach to handle the representation of shared and distinct care planning concepts between professions, specialties, and patients across clinical settings [2]. Integrated interprofessional team-based approaches to care coordination are associated with better patient outcomes [3]. Care plans that align interprofessional care goals are a central component of integrated care delivery [4], and integrated care is associated with increased knowledge sharing between disciplines [5,6]. The design of care plans and decision support that fail to explicitly link interprofessional knowledge will propagate isolated care planning [4,7,8], leading to poor team communication and suboptimal patient outcomes [9–11]. Moreover, problem lists and care plans that fail to support linkages between care delivered (SNOMED CT) and administered processes (ICD-9 billing codes) will result in inefficiencies, billing errors, unrealistic expectations for provider documentation, and likely decreased hospital reimbursement. Effective care planning that is team-based and patient-centered hinges on the development of dynamic care plans with embedded functionality for interprofessional knowledge sharing [4]. We posit that a redesign of EHR terminology and knowledge representation infrastructures are necessary to produce effective patient summaries for continuous care planning that are interprofessional and consensus-driven, and able to promote shared understanding. EHR tools to engage with patient summaries should provide a flexible and dynamic “blueprint” to guide care, while leveraging reference terminologies to ensure interoperability and knowledge sharing. 2. Background The AHRQ Care Coordination Measures Atlas provides a framework for measuring care coordination that is organized by broad approaches (e.g., Health Information Technology) and coordination activities (e.g., communication and tracking of clinical states) with an important emphasis on measuring patient-centered care coordination from multiple participant perspectives (e.g., patient, provider, and system) [1]. The terminology requirements for clinical documentation in many clinical applications are based on requirements of a single discipline or profession, typically using one reference terminology, and with few linkages between concepts. However, the requirements for interprofessional care planning include support for multiple clinical professions with overlapping clinical terminology needs. The design of a care planning infrastructure must consider how it will be used as a cooperative, shared tool [12] and if existing terminologies are sufficiently robust to support interprofessional care planning content [13]. For example, a single activity may be represented from multiple perspectives including the patient, family, healthcare professionals, and population level metrics. The AHRQ framework provides a critical foundation to evaluate requirements for coordinated care planning tools. Recent advances related to care planning include the HL7 Version 3 Care Plan Domain Analysis Model (HL7 V3 CP DAM) and the Office of National Coordinator for Health Information Technology’s (ONC) Standards and Interoperability Framework Longitudinal Coordination of Care (LCC) Workgroup (WG) [14,15]. Yet, it is critical to recognize that the HL7 V3 CP DAM states that it is intended as an interim solution since “limitations in information system architecture, and healthcare cultural issues such as who ‘owns’ the care plan, how items are added, deleted updated etc. makes the near term implementation and use of dynamic care plans unlikely” [15]. In fact, the HL7 V3 CP DAM describes a vision for a collaborative care model where the care plan is dynamically updated and maintained as a flexible, accurate, and accessible tool with all information needed by patients and clinicians for cost-effective, high quality care. As noted in the HL7 V3 CP DAM, a standard for dynamic care planning would be ideal, but it is not feasible in the near term due to EHRs’ lack of infrastructure to support the terminology and modeling requirements [15].

3. Methods We used a 3 step approach: (1) Process model and clinical scenario development, (2) Requirements analysis, and (3) Development and validation of information and terminology models. Step 3 included the development of an Information Model using Object-Role Modeling (ORM) and a Terminology Model represented using Common Terminology Services 2 (CTS2) (see Fig. 1 and descriptions below). 3.1. Step 1: Development of process model and clinical scenarios An initial set of requirements for an interprofessional care planning process model were identified based on the HL7 V3 CP DAM, ISO Reference Terminology Model (RTM) for Nursing Diagnoses and Actions, and the AHRQ Care Coordination Measures Atlas. Next, we developed four outpatient clinical scenarios that were used to confirm process model concepts and generate requirements and three inpatient clinical scenarios that were used to validate requirements, i.e. seven clinical scenarios total. Scenarios were selected to reflect complex, but common clinical situations (i.e., such as patients with a chronic disease and socio-demographic risk factors) as these would provide a better basis for our model. The four clinical scenarios for the outpatient setting were: (1) Uncontrolled Diabetes Mellitus Type I, (2) Congestive Heart Failure related to Myocardial Infarction, (3) Diabetes Mellitus Type II and Uncontrolled Depression, and (4) Immune-Mediated Kidney Failure. In addition to the patient and family, four interprofessional roles were represented across the scenarios, specifically primary care physician (PCP), care coordinator (CC), licensed clinical social worker (LCSW), and pharmacist (PharmD). The content was based on care planning concepts from the HL7 V3 CP DAM and included assessment data, past medical history, health concerns, interventions, and goals, as 23

Journal of Biomedical Informatics 75 (2017) 22–34

S. Collins et al.

Fig. 1. Diagram of Methodological Steps.

3.3. Step 3: Development and validation of information and terminology models

documented by the different interprofessional roles. Subject matter experts (SMEs) were recruited to validate the outpatient clinical scenarios. A primary care physician, two care coordinators who were registered nurses, a psychiatrist and a pharmacist were interviewed and asked to validate the scenarios based on the clinical domains included in the scenarios. Interviews were thirty minutes to one hour long, during which a member of our team read through the clinical scenario with the SME and made changes and added notes where appropriate according to the SMEs’ feedback. At least two SMEs validated each scenario, based on the clinical domains addressed in the scenario. We conducted a systematic analysis of these clinical scenarios and identified interprofessional care planning requirements to iteratively refine the process model. A second set of three clinical scenarios was developed independently for the inpatient setting. The scenario topics were: (1) Cancer and at Risk for Fall, (2) Cancer and Sepsis, and (3) Orthopedic Requiring Anticoagulation. The first two scenarios underwent the same validation process as described above for the outpatient scenarios with an inpatient oncology nurse and a physician oncologist. The third inpatient clinical scenario was derived based on existing hospital protocols and anticoagulation management guidelines.

Each concept and associated attributes from the process model were defined in detail in an information model. The primary information modeling method used in this project was Object-Role Modeling (ORM) [18]. ORM utilizes a method for expressing fact types (predicates) about a domain of discourse. The design procedure of ORM involves analysis (generating example facts), design (identifying the fact types), and refinement (adding the constraints and modifying the structure). ORM models are expressed graphically in FORML (Formal ORM Language), which provides a formal graphical notation of ORM. An entity, the main type of object in ORM, is represented by a rounded rectangle, where the means used to identify instances of an entity may be indicated in parentheses. An association is represented by contiguous boxes, where one box is used for each role in the association. A uniqueness constraint on an association is represented by a line segment over a combination of roles in the association, in which case a maximum of one instance of that association in the information system may occur involving the corresponding combination of object instances. As an example, Fig. 5 shows the fact type that a decision occurred at an encounter. In that fact type, a uniqueness constraint is placed over the role that the decision entity plays in the association, which means that a given specific decision must have occurred at only one encounter. In other words, the model does not allow the same specific decision act at more than one encounter. To ensure that the information model was aligned with the project requirements, the requirements were first gathered and considered for what types of facts need to be tracked. These fact types are then expressed in the information model, primarily by associations that describe the relationships, semantics, and constraints in those fact types. For example, one of the requirements was the ability to track at which specific encounter a given decision of interest was made. This information can be tracked by including a fact type specifying that a decision occurred at an encounter. Therefore, this particular fact type was added to the model, as shown in Fig. 5. In this project the resulting information model was also used to identify specific locations where

3.2. Step 2: Iterative requirements analysis and identification We leveraged our prior work from Step 1 to define a care planning process model, including using HL7 V3 CP DAM, to inform an initial set of requirements for a terminology infrastructure. This approach was selected so that the scope of our requirements were based on activities in our clinical process model and so that any overlapping requirements were aligned with existing standards using the HL7 V3 CP DAM. Next, we analyzed each clinical scenario to identify terminology requirements related to care coordination and care planning. Identified requirements were iteratively refined as additional clinical scenarios were analyzed. Finally, the requirements were reconciled and mapped to the AHRQ Care Coordination Measures Framework. The reconciled list of requirements were used in step 3 as the information and terminology models were developed. 24

Journal of Biomedical Informatics 75 (2017) 22–34

S. Collins et al.

Approaches were based on prior team experience with development of Clinical Decision Support (CDS) infrastructure, our process model, and best practices for reusability, usability, implementation complexity and maintenance flexibility. We developed requirements that outline functions, processes, and methods needed to support interprofessional care planning. Our team considered the value of combining approaches to meet interprofessional care planning terminology concept representation and content management requirements with a primary emphasis on using SNOMED CT. Next, our models were validated using the second set of clinical scenarios from the inpatient setting. We chose to validate using clinical scenarios from a different clinical setting (i.e., inpatient setting) than the first set of scenarios to further the applicability of our derived terminology requirements (see Fig. 1). Specifically, this validation was performed by verifying that the extracted requirements and the process model were maintained in the validation scenarios. Given our selection of scenarios from different clinical settings, each scenario was not required to cover the complete set of requirements, rather the focus of the validation was to identify if there were any new requirements and to confirm that there were no conflicting requirements.

terminology was needed, by inspecting the information model for fact types that contain an entity whose values are specified in an applicable terminology. Those locations in the model show where terminology would be needed. Common Terminology Services 2 (CTS2) was selected as a platform to represent the terminology model for interprofessional care planning. We selected the entity and value set schemas of the CTS2 standard for the representing various reference terminologies. Common Terminology Services 2 (CTS2) is an Object Management Group standard that defines service interface requirements for the representation, access, and ongoing maintenance of terminology and ontology resources [19]. CTS2 specifies service descriptions and interfaces, along with models that define attributes and associations of common elements of structured terminologies. Examples of these common elements include models for concepts and value sets. The “entity” model provides generic attributes to represent concepts, classes of concepts, predicates, etc. Relevant model attributes include identifier, namespace, code system, designation, and status, along with the ability to define custom properties. The model also defines common concept associations (e.g. parents, children, equivalent concepts). The “value set” model provides attributes to represent groups of concepts typically created for classification purposes. Relevant model attributes include identifier, name, purpose, and members. Our use of CTS2 focused on SNOMED CT with consideration of other relevant reference terminologies, such as ICD-9-CM, ICD-10-CM and ICD-10-PCS, Current Procedural Terminology (CPT), and International Classification of Nursing Practice (ICNP) [20–23]. Based on the concepts defined in the process and information model our team evaluated a series of approaches for terminology representation and management.

4. Results 4.1. Step 1: Development of process model and clinical scenarios Our process model contains the concept types identified from our validated clinical scenarios and are displayed in a cyclical process Fig. 2). These concept types are: Health Concerns, Goals, Decisions, Interventions, Assessments and Evaluations. Importantly, the patient is

Fig. 2. Patient Centered Care Planning Process Model.

25

Journal of Biomedical Informatics 75 (2017) 22–34

S. Collins et al.

argument is the requirement that while goals are identified by distinct care team members and patients - and the goals identified by each individual may be unique or may overlap - the set of goals for a given patient should not conflict. In our scenarios we found that overlapping goals identified by different health professionals and patients were either synonyms or hierarchically related as goals and sub-goals (see Fig. 4 illustrating goals from Scenario 3). A goal statement information model and the lack of formal representation for goal statements in existing terminologies was described in detail by Bakken et al. in 2002 [29]. Our focus here is looking at the ‘readiness’ of SNOMED CT for terminology bindings. In attempting to map these goals within our terminology management model, we found that SNOMED CT does not formally represent goals structurally. Given the lack of formal representation of goals in SNOMED CT, we used a less than ideal workaround to explicitly define that the clinical concept was a goal-setting concept by post-coordinating with the SNOMED CT code “Identifying goal 225294001”. This approach is not ideal and resulted in post-coordination for every goal in our data set. Initially, we structured our post-coordination approach to include the following components: goal-setting, action qualifiers, clinical concepts, and timeframe qualifiers. SNOMED CT action qualifiers were insufficient to represent goal actions (e.g., adhere, secure, maintain) because action qualifiers are incomplete in SNOMED CT and other actions within SNOMED CT are unapproved. We identified the need to add more actions as qualifier values for complete representation of goal actions; however, doing so could explode the SNOMED CT hierarchy. Based on this analysis we decided that we would not include action qualifiers to represent goals. Timeframes are a critical component of a well-structured goal to support evaluation within the context of care planning process. Timeframe qualifiers should be used to represent goals, though the concepts in SNOMED CT are incomplete and require additional codes for robust representation of clinical timeframes (e.g., 6 h, 3 days, 2 weeks, 3 months). Our final structure to represent goals requiring post-coordination was: goal setting [i.e., “Identifying goal 225294001”] + clinical concept codes + timeframe qualifiers. Additional approaches to post-coordinating SNOMED CT concepts could have been used to represent goals in SNOMED CT. For example, the parent concept “Situation with explicit context [(situation) 243796009]” could be postcoordinated with definitional attributes, such as “has focus [(attribute) 363702006]”, “has intent [(attribute) 363703001]”, and “timing [(attribute) 246512002]” to convey the components of a goal. Additionally, concepts from the “Regime/therapy” sub-hierarchy may relate to goals for specific treatments, such as preventative care: “Diabetic foot ulcer prevention (regime/therapy)|713150007”. However, each of these approaches are also suboptimal given that SNOMED CT does not have a terminology model for goals and ultimately a hierarchy devoted to goals is required to meet the need. It is important to note that post-coordinating goals using SNOMED CT poses challenges because SNOMED CT concepts often pre-coordinate the “negative” aspect of a concept, where as goals are focused on moving toward positive states. As we move toward greater preventative population health management, reference terminology concepts goals will need to refocus from the negative to promoting the positive. We were not able to formally represent Decisions structurally using SNOMED CT. Decision concepts were represented by post-coordinating concepts in SNOMED CT. The post-coordinated structure used was: planned action + rationales (if applicable) + anticipatory guidance (if applicable) + clinical guideline concept (if applicable). However, there were two limiting factors: (1) inability to represent the temporal context of decisions as planned actions as opposed to completed actions when post-coordinating, and (2) there was a significant lack of concepts in SNOMED CT to capture Clinical Guidelines. For example, related to the first limitation, a decision from scenario 1 was to prescribe Coumadin for a patient that is at risk for stroke and has been diagnosed with Atrial Fibrillation. Our post-coordination of SNOMED CT concepts was:

at the center of the process model Fig. 2) in collaboration with clinicians and family/care partners as an intentional indication that everyone can play an active role at each step in the care process. As a concrete example, the broad term Health Concern includes any concern from the patient and family/care partner as well as types of diagnoses, such as medical diagnoses and nursing diagnoses. Each concept in our model is intended to include patient-defined concepts and family/care partner-defined concepts, in addition to clinician-defined concepts. This inclusive approach is aligned with our focus on all aspects of care, including patient and family-centered care, shared-decision making, and self-management. Table 1 includes brief descriptions and selected content most relevant to the primary care coordination issue identified for each scenario. The associated role or roles that identified each concept in the scenario are noted with the key at the bottom of Table 1 illustrating the collaborative nature of the care planning process. The scenarios included on average 12 health concern concepts, 14 goal concepts, 24 decision concepts, 24 intervention concepts, 12 evaluation concepts, and 4 clinical professions (Table 2). The selected content from each developed scenario in Table 1 illustrates several evolving clinical care plans that emphasize social determinants of health, communication, and education concepts with interprofessional linkages, in addition to medical conditions and disease states. For example, the concerns highlighted for scenario 1 were each identified by a different clinician: Poor diet identified by the care coordinator, at risk for readmission to hospital identified by the social worker, and Congestive Heart Failure identified by the primary care physician. These three concerns interrelate around the patient-identified and clinician-identified goals of enabling the patient to live at home and self-manage his Congestive Heart Failure by maintaining optimal fluid balance and blood pressure through diet changes. 4.2. Step 2: Iterative requirements analysis and identification Our analysis resulted in definition of 6 infrastructure requirements that a terminology infrastructure for shared care plans should: (A) Include discrete care plan concepts; (B) Include sets of profession-specific concerns, decisions, and interventions; (C) Communicate rationales, anticipatory guidance, and guidelines that informed decisions among the care team; (D) Define semantic linkages across clinical concepts and professions; (E) Define sets of shared patient goals and sub-goals, including patient stated goals; (F) Capture evaluation toward achievement of goals (see Table 3). These requirements were aligned to the AHRQ Care Coordination Measures framework to demonstrate how our proposed terminology infrastructure can enable HIT Specific Care Coordination Activities (Table 4). 4.3. Step 3: Development and validation of information and terminology models Our team found that identification of a health concern by a health care professional was the point of initiation in the process model. We considered each concept in the model and its dependencies to identify the best point of initiation. While the model is a continuous and collaborative process, a health concern serves as an anchor, both from a terminology and clinical perspective. This approach is aligned with problem based charting, a long-standing framework for documentation within the medical and nursing domains and is conducive to clinical decision-making, learning, and work within collaborative interprofessional care teams [24–28]. Analyses of each clinical scenario indicated that the goals identified by distinct care team members and patients could be linked, highlighting their role as a central component of patient-centered care planning. Based on this feature, we explored and defined the structure of the terminology model; goals emerged as the best underlying link connecting all concepts together (see Fig. 3). Most importantly to this 26

27

30 years-old with uncontrolled type 1 diabetes, diagnosed as child

4

To feel healthier by next appointment(PAT) To have more energy by next appointment(PAT) Exercise at least 3 times a week(PAT)

Diabetic mellitus uncontrolled(PCP) Medication non adherence(PCP, CC) Deficient knowledge(CC)

Patient had a normal physical exam Patient does not keep a food diary and struggles to exercise because of her long work hours

Uncontrolled type II diabetes mellitus(PCP) Major depressive disorder(PCP) Housing problems(PAT,

Poor self-management

Decreased feelings of worthlessness and sadness by next appointment(PAT, CCT) Adhere to anti-depressant with the hope of anti-depressant helping to diminish feelings of worthlessness and sadness for the duration of treatment(PAT, CCT) Ultimate goal HbA1c < 7; lower HbA1c to 7.5 by next appointment(PCP) Secure stable housing by next appointment(LCSW)

Kidney disease(PCP) Risk for infection(PCP) Transportation was a barrier impeding the patient’s ability to use community resources(LCSW)

K+ = 4.6 mEq/L BUN 65 mg/dL Creatinine 9.73 mg/dl Elevated blood pressure at 136/ 91 mm HG Current weight is 143 lbs Patient has psychosocial needs related to her self-management

Inadequate level of knowledge of coping strategies related to depression(CC)

LCSW)

Live at home for the duration of treatment(PAT) Increase functional status to maintain independence for the duration of treatment(PAT; CCT) Replace high-sodium snacks with healthy alternatives twice a week(PAT; CCT) Systolic blood pressure less than 140/90 by next appointment(PCP) Target weight of 195 lbs with fluid balance control by next appointment(PCP) Kt/V of at least 1.2 for 3x/week hemodialysis(PCP) Dialysis lines will remain free of infection for the duration of treatment(PCP) Follow dietary recommendation from nutritionist for the duration of treatment(CCT) Adhere to 3–6 month follow-up appointments with Primary Care Provider(PCP)

Poor diet(CC) At risk for readmission to hospital (LCSW) At risk for stroke(PCP)

Patient cares for self at home, is active, uses public transportation and is able to independently complete activities of daily living Snacking on high sodium foods Elevated blood pressure of 155/ 83 mm HG Current weight is 200 lbs, prior weight 205 lbs

Scored 15 out of 21 on the Beck Depression Inventory for Primary Care (BDI-PC), indicating severe depression symptoms Elevated HbA1C of 8.4% Patient has difficulty coping with multiple health and financial problems

Primary Goals

Top 3 Concerns

Assessment

Selected Content Most Relevant to Primary Care Coordination Issue Identified for Each Scenario

Difficulty coping with multiple health and socioeconomic problems

Started her dialysis last week, but has no scheduled renal clinic follow-up appointment and does not know who her main nephrologist is.

21 years-old with immune-mediated kidney failure post discharge from hospital and starting dialysis

2

47 years-old with uncontrolled depression, type 2 diabetes, and financial instability

Wife, recently deceased, was primary care coordinator

76 years-old with congestive heart failure (CHF), hypertension, past myocardial infarction and atrial fibrillation

1

3

Primary Care Coordination Issue

Primary Care Outpatient Scenarios

Table 1 Patient descriptions from clinical scenarios.

Recommend patient receive vaccines for influenza virus, pneumococcus, and hepatitis B to prevent infection)(PCP) Patient will call provider if infection or clot of dialysis line suspected)(PCP) Recommend the patient see a nutritionist since she has not yet gone to the dialysis clinic and seen a nutritionist(PCP) Decide to touch base with the dialysis care coordinator and social worker to discuss patient’s plan of care hospital(CC) Patient should stay on Zoloft for approximately 4–6 months to feel the full benefit of the medication(PCP) If common side effects of Zoloft occur, including nausea, abdominal pain and diarrhea, should stay on the medication as those side effects will typically disappear in a few days (PCP) If rare side effect of increased risk of suicidality occurs patient instructed to stop the Zoloft immediately and go to her closest emergency room(PCP) Recommend guidance and counseling for depression(LCSW) Recommend housing resources and financial resource assistance(LCSW) Endocrinology consult should evaluate need for insulin pump given blood glucose remains uncontrolled(PCP) Recommended HbA1c testing every

Patient agrees to adhere to a 2 mg low sodium diet replacing chips with healthy alternative snacks twice a week(CC) Recommend patient join a CHF program to remotely monitor his weight (PCP) Social worker should assess for selfmanagement needs and risk for readmission to hospital(CC) Recommend Coumadin secondary to stroke risk for Atrial Fibrillation

Shared Decisions or Clinical Recommendations/Guidelines

Referral to an endocrinologist(PCP) Prescribed Humalog subcutaneously before means and at bedtime for blood (continued on next page)

Diabetes self-management education(CC) Take Zoloft 25 mg by mouth daily(PCP) Referral to clinical social worker(PCP) Referral to individual weekly psychotherapy(LCSW) Depression education(CC) Assisted the patient in finding financial and housing resource assistance(LCSW)

Intermittent hemodialysis(PCP) Administer vaccinations(PCP) Patient educated on how to identify red flags that indicate infection or clot of the line and when to call provider Administer Transportation education, guidance, and counseling(LCSW)

Social worker referral(CC) Referral to CHF Telemanagement Program(PCP) Follow-up phone call and appointment for assessment and evaluation of goals(CC) Order Coumadin (5 mg by mouth daily)

Associated Interventions

S. Collins et al.

Journal of Biomedical Informatics 75 (2017) 22–34

28

85 years-old with multiple myeloma and at risk for falls admitted to Medical Intensive Care Unit (MICU) for sepsis

79 years-old post total Hip Replacement who lives alone.

7

Hemodynamic stabilization/ resuscitation with Mean Arterial Pressure (MAP) > 65 and Central Venous Pressure (CVP) > 8 during the first 6 h of resuscitation(CCT) Mixed venous oxygen saturation (Svo2) of ≥65% during the first 6 h of resuscitation(CCT)

Discharged to home after hospitalization(PAT) Patient will demonstrate independent and safe hip precautions while completing IADL activity for ten minutes before discharge (OT)

Septic shock(MD) Decreased cardiac output(RN)

Total hip replacement(MD) Impaired ability to carry out independent activities of daily living (IADLs) upon discharge(RN) Risk for deep vein thrombosis(RN, MD)

Alert and oriented to person Decreased activity tolerance Decreased performance in activities of daily living (ADL’s) Decreased performance in functional mobility

Physical therapy recovery and safe anticoagulation management post discharge

Patient will be free of sign and symptoms of deep vein thrombosis during hospitalization(MD, RN) Patient will be free of falls during hospitalization(RN)

RN)

To be cured To live longer(PAT, FAM, CCT) To be comfortable during hospitalization(PAT, FAM,CCT) Patient’s temperature will be less than 99.5 during hospitalization(MD,

(PAT, FAM)

Primary Goals

Gain glycemic control, measured by: morning fasting blood glucose range of 80–120; HgBA1C of < 8; postprandial blood glucose of less than 180; no episodes of hypoglycemia by next appointment(CCT)

Primary Goals

White blood cell count < 4000 cells/μL Temperature of 101.4 F (38.5 C) Mean Arterial Pressure = 48 mm HG Tachycardia at 120 beats per minute Mixed venous oxygen saturation = 49%

Multiple myeloma Risk for infection(MD) Risk for deep vein thrombosis (RN, MD) High risk of falling(RN) Alteration in coping(RN)

Alert and oriented to person and place Afebrile Marrow biopsy shows 30 percent involvement by abnormal appearing plasma cells

(MD)

Top 3 Concerns

Top 3 Concerns

Assessment

Elevated HbA1c of 9.5% Blood glucose of 146

Assessment

Selected Content Most Relevant to Primary Care Coordination Issue Identified for Each Scenario

Identification and management of critical condition upon unanticipated transfer to MICU

Primary Care Coordination Issue Establishing goals of care in setting of acute issues and risks

Acute and Critical Care Inpatient scenarios 5 85 years-old with multiple myeloma admitted to oncology unit after he fell at home

6

Primary Care Coordination Issue

Primary Care Outpatient Scenarios

Table 1 (continued)

Recommended 10 units of Humalog before light carbohydrate meals and 20 min before heavy carbohydrate meals(PharmD) Recommend 3 week follow-up with pharmacist to monitor progress towards medication adherence(CCT) Shared Decisions or Clinical Recommendations/Guidelines Plan to hold a Family Meeting to discuss patient prognosis and that myeloma is not curable(CCT) Lowered fever parameter to 99.5 to avoid patient already being septic at time of temperature spike given patient is on steroids and at risk for infection(CCT) Patient will continue current Myeloma therapy regimen until further discussion with outpatient oncologist and patient and family around goals of care considering toxicity risks from therapy and current co-morbidities(PAT, FAM, CCT) Treatment should follow institution’s Sepsis Management Guideline(MD, RN): – Fluid challenge boluses of 500 ml Normal Saline IV every 30 min if the CVP is below 8 mmHg, continue boluses until a CVP between 8–12 mmHg is achieved (12–15 mmHg for vented patients) or until patient is normotensive with a MAP > 65 mmHg – Notify physician if mixed venous oxygen saturation (Svo2) is less than 65% – Levophed to maintain blood pressure of mean arterial pressure MAP > 65 mmHg Recommend physical therapy for 8 weeks(MD) Recommend Anticoagulation Management Service (AMS) to manage anticoagulation for venous thromboembolism prophylaxis(MD) Recommend visiting nurse services

Home exercise program with repetition of each exercise 25x, 2–3x/day(PT) Warfarin 1 mg by mouth daily(MD) Patient education on warfarin risks and INR management(AMS) (continued on next page)

Collect Blood cultures(RN) Order Ceftriaxone 2 g IV, STAT(MD) Order Imipenem/Cilastatin(MD) Administer Fluid challenge boluses(RN) Titrate Levophed 2.5 to 15 mcg/ kg/minute IV to goal(RN)

Family meeting(PAT, FAM, CCT) Monitor patient for fever > 99.5(RN) Myeloma therapy regimen as prescribed by oncologist(MD) Administer Pneumatic boots(RN) Order for Lovenox(MD) Educates patient/families about the fall risk and safety plan(RN)

Associated Interventions

glucose management(PCP) Prescribed NPH insulin 30 units subcutaneously twice daily for blood glucose management(PCP) Referred to a registered dietician(PCP) Follow-up visits scheduled(CCT)

3 months for evaluation of blood glucose management(PCP) Diabetes care plan agreed upon(PAT, CC)

Associated Interventions

Shared Decisions or Clinical Recommendations/Guidelines

S. Collins et al.

Journal of Biomedical Informatics 75 (2017) 22–34

Journal of Biomedical Informatics 75 (2017) 22–34

Associated Interventions

Visiting nurse services referral upon discharge(MD) Order for INR level check in 3 days and INR tests twice a week for the duration of prophylaxis(MD)

Shared Decisions or Clinical Recommendations/Guidelines

daily for 4 weeks to perform INR test with a point of care machine(AMS) Recommend patient obtain MedicAlert necklace/bracelet, ID card, or other notification informing other medical caregivers of anticoagulation status(AMS)

Primary Goals

Daily performance of home exercise program for 12 weeks post discharge(MD) INR goal of 2.0–3.0 for duration of anticoagulation therapy with Warfarin(MD) Patient will be free of sign and symptoms deep vein thrombosis for the duration of treatment(CCT)

Top 3 Concerns Assessment

• Prescribe Coumadin (5 mg po QD) [(product) 48603004] secondary

Patient = PAT; Family = FAM; Primary Care Physician = PCP; Care Coordinator = CC; Licensed Clinical Social Worker = LCSW; Pharmacist = PharmD; Registered Nurse Inpatient Oncology (RN); Attending Physician Inpatient = MD; Collaborative Care Team = CCT.

Primary Care Outpatient Scenarios

Table 1 (continued)

Primary Care Coordination Issue

Selected Content Most Relevant to Primary Care Coordination Issue Identified for Each Scenario

S. Collins et al.

to stroke risk [(disorder) 230690007 + (qualifier) 30207005] for Atrial Fibrillation [(attribute) 246106000 + (disorder) 49436004].

These post-coordinated SNOMED codes capture the clinical concepts, but insufficiently represent the temporal context that this is an explicit decision for a future planned action and not an act that has already been completed. As an example of the second limitation, SNOMED CT includes a concept for “Implementation of pain guidelines (regime/therapy)”, but we were unable to find similar concepts for the three clinical guidelines in our scenarios (i.e., fall risk, sepsis, anticoagulation management). As part of our post-coordination we were able to represent the majority of Rationales and Anticipatory Guidance and Clinical Guideline concepts from our Decision model using SNOMED CT. For example, for the Decision example above, the rationales of stroke risk and Atrial Fibrillation convey why a decision to prescribe Coumadin was made at that point in time. The Anticipatory Guidance examples below from scenario 6 convey how SNOMED CT was used to represent these concepts and note two exceptions in which clinical concepts did not have an associated SNOMED CT code identified. For the Decision that the patient should receive a fluid challenge boluses of 500 ml Normal Saline IV every 30 min:

• Anticipatory Guidance = if the CVP is below 8 mmHg [decreased

central venous pressure (Finding, 38398005)], continue boluses [fluid balance therapy (Procedure, 276026009)] until a CVP between 8–12 mmHg is achieved (12–15 mmHg for vented patients) [normal central venous pressure (Finding, 91297005)] or until patient is normotensive [normal blood pressure (Finding, 2004005 with a MAP > 65 mmHg [no SNOMED CT code identified].

For the Decision to order and monitor a mixed venous oxygen saturation:

• Anticipatory Guidance = to notify the physician [no SNOMED CT code identified] if the mixed venous oxygen saturation is less than 65% [mixed venous oxygen saturation (observable entity, 442734002) + decreased (qualifier, 1250004)].

We were able to successfully represent Health Concerns, Interventions, Assessments, and Evaluations using SNOMED CT. Additionally, we found that health concerns and interventions were largely profession-specific in our scenarios. Health professionals identified distinct sets of health concerns and interventions, but explicit communication processes were used to coordinate interventions identified by other clinicians. Table 5 defines each concept type and its relation to SNOMED CT hierarchies or LOINC. Content from the second set of scenarios was used to validate our requirements and derived models and prototype. We confirmed that there were no conflicting requirements, but did iteratively refine requirements once the inpatient scenarios were introduced. It is noteworthy that the domains in the inpatient setting that were identified in our scenarios as requiring care coordination linked to interprofessional protocols and guidelines that were primarily focused on high risks domains (falls, anticoagulation), critical conditions (sepsis), and discharge readiness (physical rehabilitation, post-discharge services), rather than disease states. The domains requiring care coordination in our outpatient scenarios did not include any interprofessional protocols and guidelines identified by our subject matter experts, though it is possible that relevant guidelines exist. Interestingly, none of our subject matter experts identified protocolized care or guidelines aligned with social determinants of health concepts, such as the patients in our scenarios who were recently widowed or struggled with financial instability. The validation analysis demonstrated that the definition of the Health Concern, Goal, Intervention and Evaluation concept types were 29

Journal of Biomedical Informatics 75 (2017) 22–34

S. Collins et al.

Table 2 Count of HL7 care plan domain analysis model concepts per clinical scenario. Clinical scenario

Concerns

Goals

Decisions

Interventions

Evaluations

Clinical professions

1 2 3 4 5 6 7

9 7 21 15 20 1 13

7 7 13 10 22 5 33

18 21 28 30 24 20 29

18 21 28 30 24 20 29

7 7 13 3 22 5 33

4 4 4 5 4 3 6

Average Median Total

12.3 13 86

13.9 10 97

24.3 24 170

24.3 24 170

12.9 7 90

4.3 4 30

Table 3 Terminology infrastructure requirements and examples mapped to AHRQ framework. Terminology Infrastructure for shared care plans should: A B C

Include discrete care plan concepts Include sets of profession-specific concerns, decisions, and interventions Communicate rationales, anticipatory guidance, and guidelines that informed decisions among the care team

D

Define semantic Linkages across clinical concepts and professions

E

Define sets of shared patient goals and sub-goals, including patient stated goals Capture evaluation toward achievement of goals

F

Examples Selected from Scenario 4 Concerns, Goals, Decisions, Interventions and Evaluations • Health type I diabetes mellitus (Primary Care Provider Concern) • Uncontrolled knowledge related to glycemic control (Care Coordinator Concern) • Deficient = Referral to pharmacist • Decision for decision = to understand any reasons for lack of medication adherence and for medication • Rationale education

• ‘Gain glycemic control’ → ‘Eat a more nutritious diet’ → ‘Post-prandial blood glucose of less than 180’ • ‘Met’, ‘Not met’, ‘In progress’ entity with the following attributes: Rationales, Anticipatory Guidance and Clinical Guidelines. Inclusion of a separate Decision entity allowed for better capture of planned interventions and clarity of role responsibilities. For example, scenario 6 describes titration of intravenous vasopressors (medications to maintain a patient’s blood pressure) in the intensive care unit that requires both periodic decisions by the medical care team and continuous (minute to minute) decision making by the nurse. Specifically, this complex workflow requires a decision by the physician “to initiate the medication Levophed at a dose of 2.5–15 mcg/kg/minute IV” with the rationale “to maintain blood pressure of mean arterial blood pressure > 65 mmHg.” The workflow also requires the nurse to decide to increase, decrease, or maintain the Levophed dose based on the patient’s lack of response, over-response, or expected response in the mean arterial blood pressure > 65 mmHg every few minutes. In another example from Scenario 3, a social worker decided to discuss financial planning with a depressed patient at the patient’s next appointment if the patient is responding well to her antidepressant medication and appears capable of participating in the discussion. These decisions to act now or in the future following an anticipated plan provide critical information for other team members to understand the reasoning for dynamic changes in interventions and plans within a care plan.

stable given no significant changes were needed to capture the inpatient scenarios. However, the Decision concept was refined multiple times during the analysis of both outpatient and inpatient scenarios. The Decision concept emerged as a significant concept in care planning across our scenarios and its use as a formal concept that could be mapped to a reference terminology was novel. In our first set of clinical scenarios (outpatient scenarios), health professionals identified reasons for a particular intervention. Initially, only a Decision Modifier entity was created to account for why an Intervention was chosen and conducted. These Decision Modifiers consisted of Rationale, Anticipatory Guidance and Clinical Guidelines. Upon further examination of the highly protocolized inpatient scenarios that included interprofessional guidelines with anticipatory guidance we created a separate Decision Table 4 Terminology infrastructure requirements alignment with agency for Healthcare Research and Quality (AHRQ) care coordination framework. Requirements from Table 3

HIT Specific Care Coordination Activities

A, B, C, D, E, F A, B, C, D, E, F

Establish Accountability or Negotiate Responsibilities Communicate: Interpersonal communication & Information transfer Facilitate Transitions Assess Needs and Goals Create a Proactive Plan of Care Monitor, Follow Up, and Respond to Change Support Self-Management Goals Link to Community Resources Align Resources with Patient and Population Needs

A, B, C, D, E, F E, F A, B, C, D, E, F D, F E, F A C, E, F

5. Discussion Our proposed approach to terminology management by defining and linking subsets of care planning concepts appears to be feasible to 30

Journal of Biomedical Informatics 75 (2017) 22–34

S. Collins et al.

Fig. 3. Example Diagram of Portion Terminology Model for Clinical Scenario 3.

of

processes and outcomes, and (4) supporting linkages between care documentation and administrative processes. Examples of linkages within a discipline include establishing a connection between the SNOMED CT nursing diagnosis “deficient knowledge” and the nursing intervention “patient education” to determine if patients with those care planning concepts identified have lower readmission outcomes than patients without those concepts identified in their care plans [30–32]. An example of a linkage across disciplines to support advanced analytics would be evaluating if patients with the identified medical diagnosis of “noncompliance with medication regimen” and the nursing diagnosis of “ready for enhanced management of therapeutic regimen (finding)” have lower rates of hospital re-admission than patients with a diagnosis and related interventions identified by only 1 discipline. While these two diagnoses may appear to conflict, rather they reflect the orthogonal and complementary care provided by different health professionals that contribute to patient centered care plans. The medical diagnosis of “noncompliance with medication regimen” indicates the patient is not taking their medications. The

model interprofessional care planning knowledge bases for common cooccurring conditions and chronic diseases. To our knowledge, there are no published studies that have proposed this approach, though there is support for semantic linkages within health care data. We posit that as care delivery models become increasingly team-based and patientcentered, clinicians and patients will expect care coordination tools to be flexible for dynamic associations, visualizations, and summaries of any types of health data. Zhou et al. pointed out that with collaborative clinical documentation the documenting clinician may not be the direct beneficiary of the information in the future and differing perceptions of responsibility and rewards may exist [12]. Thus collaborative forms of documentation with lower individual perceptions of reward will require smarter systems to decrease burden on clinicians and incentivize use. Examples of desired functionality of smarter systems are: (1) optimization of individual search and user preferences, (2) supporting linkages among problems and interventions both across professions and within professions, (3) supporting reporting and advanced analytic requirements, such as associations between interprofessional clinical

Fig. 4. Goal mapping for Scenario 3 – patient with diabetes mellitus and uncontrolled depression. * Synonymous goals from different clinical professions (Figure borrowed from Klinkenberg-Ramirez, 2014) [38].

31

Journal of Biomedical Informatics 75 (2017) 22–34

S. Collins et al.

Fig. 5. Information Model for Decision Concept.

may prevent a patient from taking medications, such as financial barriers, logistical challenges getting to the pharmacy, or a lack of understanding, so that a shared plan to overcome these barriers can be established. As hospitals increasingly meet health IT requirements and incentive programs, the need to establish policies surrounding computer-based interprofessional problem and care plan documentation will follow. These policies will be different across institutions due to cultural and organizational differences. Therefore, a terminology infrastructure will need to be flexible to accommodate policy constraints as they arise. We found that capturing goals and decisions in care planning is complex. While the attributes of the Decision entity (Rationale, Anticipatory Guidance and Clinical Guidelines) were initially identified from outpatient scenarios, the structure of the decision concept was landed on only after evaluation with inpatient scenarios and the specific guidelines included in those scenarios. The iterative process used and refinements made may be an indicator of the complexity of the decision concept and the lack of prior terminology work in this space. There are many factors for both the goals and decision concepts that need to be modeled and this information may not be captured in the documentation. Additionally, SNOMED CT does not represent goals or decisions structurally. Creation of goal dynamics, including requirements for goals and sub-goals, in SNOMED CT will be an important step in the ability to manage this information. Creating goal dynamics in SNOMED CT will also facilitate compliance with Meaningful Use and the Joint Commission documentation requirements of goals. More importantly, the capture of structured goals may enable better care coordination and patient care through the design of decision support tools that drive alignment of multiple goals and promote timely evaluation of goals achievement. Likewise, the capture of the Rationale, Anticipatory

Table 5 Components of terminology model. Concept type

Definition

Terminology coding

Health Concern

Represents a current and active need for care and may be a condition or problem A future aim or desired result. Is Aligned with other goals. Central link between concern, intervention and evaluation The determination of a planned course of action. May have Rationale, Anticipatory Guidance, Guidelines An act (e.g., care act) planned and performed that is linked to a health concern goal, and decision An act of determining information about the importance or value of health issues through judgment. Synonymous with an observation

SNOMED CT Hierarchy: Findings

Health Goal

Decision

Intervention

Assessment

Evaluation

Interpretation of assessment data that indicates the state of progress of a health concern in relation to the associated goal

•

New Requirement: Goals should be defined in SNOMED CT

•

New Requirement: Decisions should be defined in SNOMED CT

•

SNOMED CT Hierarchies: Procedures, Substance administration SNOMED CT Hierarchies: Disorder, Situation, Procedure, Findings, Observable Entity LOINC SNOMED CT Hierarchy: Findings

• • •

nursing diagnosis of “ready for enhanced management of therapeutic regimen (finding)” indicates that the patient needs to reach a state of readiness to engage in a plan for how to better manage their medications. These combined diagnoses identify the problem while personalizing the patient’s situation, possibly identifying external barriers that 32

Journal of Biomedical Informatics 75 (2017) 22–34

S. Collins et al.

convey the significance of the event for consideration by other clinicians who are also planning and coordinating the patient’s care. Achievement of this balance will require iterative optimization and validation with end-users. Next steps in this work include further validation, such as simulations using real patient data, including patient-entered EHR data. In addition, to leverage value from goals and decisions represented structurally in SNOMED CT, redesign of EHR infrastructures will be needed that allow for linkages of interprofessional concepts, maintenance of interprofessional knowledge bases, and EHR tools and visualizations that emphasize goals of care and transparent, shared decision making.

Guidance and Clinical Guidelines attributes for Decisions will enable other health professionals working with the same patient to understand why a particular decision was made. The recognized need to formally represent Decisions structurally in SNOMED CT may increase as collaborative care documentation evolves. We are not aware of any current efforts by SNOMED International to focus on the semantic types of goals and decisions, but recommend the initiation of these efforts. Existing clinical practice guideline models, such as guideline elements model (GEM) [33], GuideLine Interchange Format (GLIF) [34], and GELLO Expression Language [35,36], provide information models for representing care guidelines and decision support, with each model serving a specific purpose. While these models could be used for our purposes, we believe it would add unnecessary complexity to representation of care planning concepts within an EHR. GEM is intended to facilitate translation of human-readable guidelines into a computerinterpretable format by means of tagging relevant information using pre-defined tags. The pieces of information are selected using the “GEM cutter” and then tagged using a simple predefined hierarchy. It can be used to represent document content throughout the entire guideline life cycle. GLIF is primarily used to model the execution flow of guidelines in terms of actions and decisions. GELLO is a standard query and expression language for decision support. Its syntax depends on the use of an object-oriented data model. GELLO is platform independent, and eliminates the need for curly braces or other implementation-specific encoding methods for information retrieval as part of knowledge content - a major limitation of the Arden Syntax. GLIF relies on expression languages such as Arden Syntax or GELLO for decisions, with GELLO as a preferred fit because it is object-oriented and platform independent. GELLO relies on a common data model (e.g., HL7 virtual medical record), however, if such a model is not used, it would still require mapping to a common model for information exchange. Therefore, the use of GEM, GLIF, and GELLO would act as intermediaries and require translation between the EHR and each model, as opposed to representing care planning directly within the EHR. In our proposed approach, CTS2 can be used for common data representation for the purpose of information exchange. The CTS2 representation could be removed from implementation, by representing the model in a separate knowledge management platform to declare all components, interactions, and expected behaviors and applied to an EHR using an ETL process (extraction, transformation, and loading). A digital infrastructure that “captures and delivers the core data elements and interoperability needed to support better care, system improvement, and the generation of new knowledge” is a recommendation from the National Academy of Medicine for a continuously learning health care system [37]. We propose the described terminology infrastructure as a step toward this goal. The ability to document and link care planning concepts will allow these data to be analyzed together, facilitating the generation of health outcomes related to the combination of diagnoses, goals and interventions set for the patient. The National Academy of Medicine also recommends care continuity, specifically the development of “coordination and transition processes, data sharing capabilities, and communication tools to ensure safe, seamless patient care.” [37] By linking care planning concepts of different health care professionals, our terminology infrastructure allows for the generation of one care plan for a patient consisting of the plans of care from all health professionals involved in that patient’s care. The ability to produce one care plan for a patient allows health professionals to visualize the plan inclusive of other health professionals’ decisions and actions. The intended sharing and coordination of these plans among care team members will necessitate a balance in identifying the threshold at which further coding a clinical statement results in loss of valuable contextualized information in human readable format. For example, the patient’s goal of “being able to attend daughter’s wedding” could be coded more generically to mean the goal of “attending a future event” while still capturing and displaying the valuable personal detail of “daughter’s wedding” in free text, serving to

5.1. Limitations This study is an exploratory research study using a constrained set of clinical scenarios validated by health professionals, but not by patients. Future work should apply our developed model to a broader set of patient-validated clinical scenarios for further testing. A main finding, that terminology models for goals and decisions are unavailable in SNOMED CT, is a limitation of this research. The lack of capability to fully define and implement the goals and decisions concepts of our proposed model in a system capable of parsing the SNOMED CT terminology models limits the evaluation. 6. Conclusion We applied information and terminology modeling techniques and defined a set of “next-generation” terminology infrastructure requirements consistent with AHRQ Care Coordination Measures Framework. These entity types in our terminology model are: Health Concerns, Goals, Decisions, Interventions, Assessments and Evaluations. Our requirements analysis and definition of terminology concepts for 7 clinical scenarios resulted in a set of content for an interprofessional knowledge base for common co-occurring conditions and chronic diseases. Further work in this area should focus on expanding this knowledge base to develop, implement and test a care coordination application integrated with existing EHRs for use in the clinical setting to manage patients with multiple co-morbidities and conditions. Based on our findings we recommend the creation of goal dynamics, including requirements for goals and sub-goals, and decision concepts in SNOMED CT. Requirements from Meaningful Use and the Joint Commission include documentation of goals, yet no controlled terminology provides a mechanism for coded documentation of these concepts. A learning health system will require the ability to semantically link and compute the clinical concepts defined in our model. As collaborative and shared documentation are increasingly adopted to enable patient centered models of care, the terminology requirements we defined in this project may drive design of systems to decrease documentation burden and enable intelligent decision support to identify conflicting and aligned concerns, goals, decisions, and interventions in shared summaries of patient care. Acknowledgements This study was funded by Partners-Siemens Research Council Grant: Knowledge Management Terminology Infrastructure to Support Interprofessional Plans of Care. The research team would like to thank all of the clinicians that participated in this study and Dr. Margarita Sordo for her expertise related to clinical practice guideline and decision support information models. Contributors All authors contributed to overall intellectual content and sections of writing. 33

Journal of Biomedical Informatics 75 (2017) 22–34

S. Collins et al.

[17] S. Henry, W.L. Holzemer, A comparison of problem lists generated by physicians, nurses, and patients: implications for CPR systems, in: Proc. Annu. Symp. Comput. Appl. Med. Care., 1995, pp. 382–386. [18] T. Halpin, T. Morgan, Information Modeling and Relational Databases, 2nd ed., Elsevier, Amsterdam, 2008. [19] Object Management Group (OMG) Inc. Common Terminology Services 2 – Version 1.2 [Internet] (cited 2017 Jun 24). Available from: < http://www.omg.org/spec/ CTS2/ > . [20] U.S. National Library of Medicine. SNOMED CT [Internet]. U.S. National Library of Medicine; 2017 (cited 2017 Feb 20). Available from: < https://www.nlm.nih.gov/ healthit/snomedct/ > . [21] International Classification for Nursing Practice (ICNP®) [Internet]. International Council of Nurses, 2015 (cited 2017 Feb 20). Available from: < http://www.icn. ch/what-we-do/international-classification-for-nursing-practice-icnpr/ > . [22] Center for Medicare & Medicaid Services. ICD-9-CM Diagnosis and Procedure Codes: Abbreviated and Full Code Titles [Internet], 2014 (cited 2017 Jun 24). Available from: < https://www.cms.gov/Medicare/Coding/ ICD9ProviderDiagnosticCodes/codes.html > . [23] Centers for Medicare & Medicaid Services. ICD-10 Overview [Internet], 2017 (cited 2017 Jun 24). Available from: < https://www.cms.gov/Medicare/Coding/ICD10/ index.html > . [24] M. Müller-Staub, I. Needham, M. Odenbreit, M.A. Lavin, T. van Achterberg, Implementing nursing diagnostics effectively: cluster randomized trial, J. Adv. Nurs. 63 (3) (2008 Aug) 291–301. [25] M. Muller-Staub, Evaluation of the implementation of nursing diagnoses, interventions, and outcomes, Int. J. Nurs. Terminol. Classif. 20 (1) (2009) 9–15. [26] M. Muller-Staub, M.A. Lavin, I. Needham, T. van Achterberg, Nursing diagnoses, interventions and outcomes? Application and impact on nursing practice: systematic review, J. Adv. Nurs. 56 (5) (2006 Dec) 514–531. [27] A. Wright, D.F. Sittig, J. McGowan, J.S. Ash, L.L. Weed, Bringing science to medicine: an interview with Larry Weed, inventor of the problem-oriented medical record, J. Am. Med. Inform. Assoc. 21 (6) (2014 Nov) 964–968. [28] S.M. Chowdhry, R.G. Mishuris, D. Mann, Problem-oriented charting: a review, Int. J. Med. Inform. 103 (2017 Jul) 95–102. [29] S. Bakken, J.J. Warren, A. Casey, D. Konicek, C. Lundberg, M. Pooke, Information model and terminology model issues related to goals, Proc. AMIA Symp. 17–21 (2002). [30] H.-A. Park, C. Lundberg, A. Coenen, D. Konicek, Mapping ICNP Version 1 concepts to SNOMED CT, Stud. Health Technol. Inform. 160 (Pt 2) (2010) 1109–1113. [31] T.Y. Kim, A. Coenen, N. Hardiker, Semantic mappings and locality of nursing diagnostic concepts in UMLS, J. Biomed. Inform. 45 (1) (2012 Feb) 93–100. [32] S. Matney, J. Warren, J. Evans, T.Y. Kim, A. Coenen, V. Auld, Development of the nursing problem list subset of SNOMED CT®, JBI 45 (4) (2012) 683–688. [33] N. Hajizadeh, N. Kashyap, G. Michel, R.N. Shiffman, GEM at 10: a decade’s experience with the Guideline Elements Model. AMIA . Annu. Symp. proceedings AMIA Symp., 2011, 2011, pp. 520–528. [34] A.A. Boxwala, M. Peleg, S. Tu, O. Ogunyemi, Q.T. Zeng, D. Wang, et al., GLIF3: a representation format for sharable computer-interpretable clinical practice guidelines, J. Biomed. Inform. 37 (3) (2004 Jun) 147–161. [35] HL7. HL7 Version 3 Standard: GELLO, A Common Expression Language, Release 2 [Internet], 2016 (cited 2017 Jun 26). Available from: < http://www.hl7.org/ implement/standards/product_brief.cfm?product_id=5 > . [36] M. Sordo, A.A. Boxwala, O. Ogunyemi, R.A. Greenes, Description and status update on GELLO: a proposed standardized object-oriented expression language for clinical decision support, Stud. Health Technol. Inform. 107 (Pt 1) (2004) 164–168. [37] (IOM) Institute of Medicine. Best Care at Lower Cost: The Path to Continuously Learning Health Care in America. Smith M, Saunders R, Stuckhardt L, McGinnis M, editors. Washington, DC: The National Academies Press, 2013. [38] S. Klinkenberg-Ramirez, K. Tsivkin, P. Mar, H. Nandigam, D. Iskhakova, R. Rocha, et al., Relating Health Concerns and Goals in Interprofessional Care Planning, in: AMIA Annual Symposium Proceedings, 2014, pp. 1467.

Competing interests None. Conflict of interest The authors have no conflicts of interest to declare. References [1] K. McDonald, E. Schultz, L. Albin, N. Pineda, J. Lonhart, V. Sundaram, et al., Care Coordination Atlas Version 3 (Prepared by Stanford University under subcontract to Battelle on Contract No. 290-04-0020). AHRQ Publication No. 11-0023-EF. Rockville, MD, 2010. [2] S. Collins, T. Hongsermeier, D. Dubois, H. Nandigam, R. Rocha, An Exploration of the Limitations of SNOMED CT to Support Sociotechnical Requirements of Interdisciplinary Problem Lists, in: AMIA Annual Symposium Proceedings, 2012. [3] J. Joynt, B. Kimball, Innovative Care Delivery Models: Identifying New Models that Effectively Leverage Nurses, 2008. [4] P.C. Dykes, R.R. Dadamio, D. Goldsmith, H.-E. Kim, K. Ohashi, V.K. Saba, Leveraging standards to support patient-centric interdisciplinary plans of care, in: AMIA Annu Symp Proc., 2011, pp. 356–363. [5] H. Boon, M. Verhoef, D. O’Hara, B. Findlay, From parallel practice to integrative health care: a conceptual framework, BMC Health Serv. Res. 4 (1) (2004 Jul 1) 15. [6] I. Gaboury, H. Boon, M. Verhoef, M. Bujold, L.M. Lapierre, D. Moher, Practitioners’ validation of framework of team-oriented practice models in integrative health care: a mixed methods study, BMC Health Serv. Res. 10 (1) (2010 Jan) 289. [7] S. van Langeveld, M. Post, F. van Asbeck, K. Postma, D. Ten Dam, K. Pons, Development of a classification of physical, occupational, and sports therapy interventions to document mobility and self-care in spinal cord injury rehabilitation, J. Neurol. Phys. Ther. 32 (1) (2008 Mar) 2–7. [8] H. Van Heukelom, V. Fraser, J. Koh, K. McQueen, K. Vogt, F. Johnson, Implementing nutrition diagnosis at a multisite health care organization, Can. J. Diet. Pract. Res. 72 (4) (2011 Jan) 178–180. [9] S. Collins, S. Bakken, D.K. Vawdrey, E. Coiera, L.M. Currie, Agreement between common goals discussed and documented in the ICU, J. Am. Med. Inf. Assoc. 18 (1) (2011 Jan 1) 45–50. [10] S. Collins, S. Bakken, D.K. Vawdrey, E. Coiera, L. Currie, Model development for EHR interdisciplinary information exchange of ICU common goals, Int. J. Med. Inform. 80 (8) (2011 Aug) e141–e149. [11] E. Larson, The impact of physician-nurse interaction on patient care, Holist. Nurs. Pr. 13 (2) (1999) 38–46. [12] X. Zhou, K. Zheng, M.S. Ackerman, D. Hanauer, Cooperative Documentation: The Patient Problem List as a Nexus in Electronic Health Records, in: CSCW 2012 ACM Conference February 11-15, 2012, Seattle, WA. 20p, pp. 911–920. [13] K.W. Fung, C. McDonald, S. Srinivasan, The UMLS-CORE project: a study of the problem list terminologies used in large healthcare institutions, J. Am. Med. Inf. Assoc. 17 (6) (2010) 675–680. [14] S & I Framework. Standards & Interoperability (S & I) Framework - Structured Data Capture Charter and Members [Internet]. S & I wiki, 2016 [cited 2016 Jan 28]. Available from: < http://wiki.siframework.org/Structured+Data+Capture +Charter+and+Members > . [15] Patient Care Workgroup. HL7 Version 3 Domain Analysis Model: Care Plan, Release 1 [Internet]. HL7 Informative Document. 2016 [cited 2017 Jun 24], p. 96. Available from: < http://www.hl7.org/implement/standards/product_brief.cfm?product_ id=435 > . [16] S. Henry, W. Holzemer, C. Reilly, K. Campbell, Terms used by nurses to describe patient problems: can SNOMED III represent nursing concepts in the patient record? JAMIA. 1 (1) (1994) 61–74.

34