Design Considerations

When describing workflows, the information solution should be person-centric and ideally allow for the representation of relevant abstract concepts (Table 1). These abstract concepts ought to be delivered showing what are sequenced vs. not sequenced actions (i.e., chronological in order), planned vs. unplanned actions, while simultaneously being profession and care organization agnostic. It should also abide by conventions in clinical informatics concepts (17), prevailing standards, and secure data integrity (18). Another feature, often neglected, is that the more that the use of informatics principles mimic the structure of everyday language (without loss of precision), the more accessible it is for the practitioners.

The different concepts from modern health care development that have been targeted to be met with the methodology presented in this paper. Their respective impact on informatics was determined by the authors for later inclusion in the construction of the present framework. References: Person-Centered Care (19), Interprofessional care (20), Value based healthcare (21). Shared decision making (2), Embedded trials (22).

The concepts should be independent of the organizational structure and not lock procedures into the current organization of healthcare (visit types, profession specific activities, specific logistic procedures, etc.). Thus, in all components it should be based on a person centric aspect and all data that has been generated should be annotated to allow for full understanding of the semantic content of each data point (Figure 1: item 13). If this is achieved, multiple use of data (documentation, quality control, business intelligence, research, etc.) is possible without laborious mapping and post-hoc assignment of semantics. Each of these suggested uses of data has different requirements for the annotation and all aspects should be present in the information and annotation model. This minimizes the effort in extract, transform and load (ETL) procedures preparing for analysis as well as the storage of data (e.g., data-lake structures).

The ability to plan, execute and document a planned sequence of events in health care requires a richer data structure than conventional EMR systems typically provide, as the latter are dedicated mainly for documentation (Table 2). A care plan has components from multiple ontological domains that are normally not used in “documentation only” EMR systems. Such domains pertain to planning, logistics, assignments to health professions and individuals within the professional domain, reminders, warnings of deviation etc. A care plan is essentially a planned process and set of requirements subject to statistical process control (23) and quality. Hence, such requirements also enter into the specification of the annotation (24).

The three principal methods of obtaining clinical data in ordered processes for trials, quality control, quality registries and as real-world data.

Extracting data from the EMR necessitates implicit semantic assumptions of the reasons of the actions that are made along the care trajectory. In the EMR case, the reasons for single actions and decisions are consecutively documented post-procedure by the professional responsible for the action, and often stored in free text format. This strips each collected data point of semantic meaning as there is not a point-to-point reference between the free text level and the points of measurements.

When automatic data extractions are made from the EMR, most often the timeline is used as a basis for post extraction semantic assumptions (see Table 2). The procedure for a blood pressure recording in the beginning of the care period is technically the same procedure as at the end of a care episode. However, the first one is probably used for diagnosis, or for general health evaluation, while the second may have completely different reasons such as screening for complications or side effects, detection of the effect of a drug, etc. Such differences are often accounted for by using different terms (concepts) for procedures, observations and intention in order to load each data point with semantic framing to allow retrospective understanding of the data. Consequently, SNOMED CT ends up with 200 concepts for blood pressure; a terminology with high level of granularity for each clinical concept but essentially unusable for practical purposes because of an abundance of specialized terms with limited applicability (e.g., SNOMED CT 718125007 Blood pressure cuff, reprocessed). Therefore, we saw the need to define a common information and annotation model that adheres to current standards whilst remaining easily implementable (usable) (Table 3) and concomitantly yields high precision while adhering to the prevailing standards in health informatics.

An overview of the standards that the project considered in the development of a multipurpose care plan strategy.

The core instrument for holding medical information in health care is the electronic medical record (EMR) with the main function of supply non-changeable records as the basis for reimbursements and attribution of individual medical responsibilities. Thus, parts of the semantic content in the terms that are used are therefore lost once transferred to the EMR. In the current design, we therefore also considered how to perform programmed data collection as prescribed in quality registries and International Consortium for Health Outcome Measurements (25, 26). They have in common the logic of planned data collection and repeated measures obtained from the same individuals (before and after procedure or just cyclically repeated sampling in chronic conditions) which is the essence of data collection based on a care plan to build the evidence base (27).

A key issue that we addressed is the absence of a dominant standard for care plan derived data annotation that fully utilizes the semantic potential. The most elaborate framework to date is the HL7 FHIR based PlanDef, FHIR CarePlan and FHIR Observation standards. We therefore developed an information model and principles for annotation with reference to those resources.