+
Durham & Darlington ERDIP Demonstrator Project
Information Models for Durham and Darlington EHR
Status: Final
Author: Prof. Ian Purves, Dr. Nick Booth.
Issue Date: 20th February 2002
Durham and Darlington EHR Project - Information Models
Why an EHR information model is needed
The EHR is concerned with electronic representation and communication of clinical information. In one sense this can be viewed as a simple progression from pre-existing current paper record systems and communications. Therefore, it is reasonable to question the need for an information model to support the EHR when there is no explicit model for these paper-based systems. This question can be addressed by contrasting the characteristics of paper records with the actual and potential characteristics of electronic records.
Paper record transfers are surprisingly limited in their direct contribution to communications surrounding the care of an individual. Research has found that records transfers represent only 10% of clinical communications. There are several reasons for this and some of these are liable to change in respect of electronic systems.
· The physical nature of the paper record means that it can only be in one place at a time. Thus for any particular paper record, a decision must be made about where it is stored and how it conveyed between those with competing needs for access. In contrast, the electronic record can be accessible simultaneously from any number of locations.
· Paper records have a single presentational form which is typically appropriate to the needs of one record user and not to another. The relatively terse life long GP record contrasts with detailed records of admission, hospital care and specialist intervention. In contrast, electronic records have the potential for multiple forms of presentation that meet differing needs.
· Paper clinical communications usually have a specific and limited objective. For example, to refer for a specialist opinion, to request an investigation or intervention or to hand over or share an aspect of the care of the patient. The type of communication that provides the entire patient record and asks the recipient to “see and advise” is quantitatively rich in information but is unlikely to be effective. The more focused brief communication that summarises the problem and asks the recipient for particular advice or intervention is more likely to achieve the intended outcome. This factor is also applicable to electronic communications. However, effective selective access to a shared electronic record may allow the recipient to augment the limited directed information. This may reduce the need for repetition of some details while enhancing the accuracy and timeliness of the available information.
· Paper communications are entirely subject to human interpretation. Recognition of relevant items information is dependent on human readers and is enhanced by their implicit understanding of the contextual relationships between various items of information. In contrast, an electronic record has the potential to be interpreted by a computer. The computerised interpretation may be limited to enabling alternative presentation or selective retrieval of information – as alluded to in the previous points. However, it may also be extended to more sophisticated processing. This may include processing for the benefit of the patient (decision support and access to appropriate reference materials), the population (epidemiology and research) and the service provider (clinical and managerial audit).
· A paper record system is merely a repository that stores information entered by people for subsequent physical retrieval. In contrast, a sophisticated computerised record system may itself become an active agent in the delivery of care. This active role is partially manifested through interpretation to support others delivering or managing health care (as noted in the previous point). However, the role of the computer as an agent may also encompass direct interaction with the patient and automated contribution to the record itself (e.g. to record a decision support interpretation or a computer generated care plan).
Electronic health records present new opportunities for collaboration and co-ordination of care, supported by more effective communication of clinical information. However, the extent to which these opportunities can be realised depends upon characteristics of the design of electronic systems. It is relatively easy to design electronic systems that echo the characteristics of a paper based system rather than realising the potential advantages. It is thus a simple matter to reach a situation in which:
· There is no physical or logical mechanism for making the record available outside the location covered by a single implementation of an application.
· A single presentational form is available or at best a selection of presentational forms suited to a single discipline or specialty.
· Directed communications are possible only as printouts or electronic documents in which there no opportunities for integration with and links into a wider health record.
· Information is stored in ways in which it can only be safely interpreted by a human reader. Thus it cannot be reliably processed for decision support or analysis. If applications support selective retrieval for these purposes on locally recorded data, these functions cannot be utilised safely on information contributed by other source systems.
· The system acts only as a repository for a collection of human mediated contributions with the computer playing no role in contributing to the quality and completeness of the record.
This outcome is the almost inevitable consequence of isolated development of different record systems. If the EHR is to assist in realising the potential benefits of electronic patient records it must encourage the development and implementation of systems that can overcome these results of insular design. To achieve this there is a need to share a common understanding of various characteristics of individual applications and the architectures within which they work together. Models exist to allow the essential characteristics of systems to be expressed and shared.
Before moving on it is worth sounding a word of caution about modelling and modellers. At best models are a way of allowing people working together to share ideas. At worst models and different modelling techniques become a quasi-religious quagmire where the correct use of jargon becomes more important than the objective. This paper attempts to distinguish various useful types of model that can be usefully applied to EHR. In this context it states what this paper means by an “EHR information model” and then takes an intentionally pragmatic view of the options for expressing this. This pragmatism may well cut through some purist views about information modelling. Partly this may be due to the brevity of the paper and partly to a conscious attempt to minimise jargon. Where jargon is used it attempts to reflect usage within the most influential parts of the international health informatics community (e.g. HL7, CEN TC251 and ISO TC215).
Models addressing EHR requirements
The essential characteristics of an EHR can be expressed using many different types of models. Three general types of model are outlined below. Each of these looks at the EHR from a different perspective and addresses the requirements that enable different potential benefits discussed in the previous section to be delivered.
· A network model concerned with how systems are connected to one another.
o This can be viewed at several levels - from the hardware level up to the application software level. Clearly it is a prerequisite for clinical communication that an effective and appropriate connectivity exists. Furthermore, different models may affect the functional capabilities of the EHR. However, this is not the topic of this paper.
· A functional model or input, communication and access to clinical information.
o This is concerned with where and when information is collected, communicated and made available to support processes associated with the delivery and management of health care.
o This can be addressed from a high-level relatively abstract architectural perspective and at the detailed nuts and bolts level. At the abstract level general characteristics can be considered such as the requirements and technical mechanisms for:
§ Applications that act as sources, which contribute to or modify a record.
§ Application that act as clients accessing a common record.
§ Shared access to information
§ Directed messaging style communications.
§ Appropriate guarantees of confidentiality
§ Adequate accessibility to meet clinical and other needs
§ Validation of provenance of information in a record
§ Management of changes to records due to clinical process (e.g. resolution of a clinical problem, completion of a requested activity, cancellation or postponement of or requested activity)
§ Management of changes to records due to correction and other maintenance activities.
o At the more detailed level such a model can address the ways in which electronic systems can support the design and implementation of specific care pathways.
o An understanding of functional requirements arising from changing models of health care delivery and the way in which these can be met by an electronic record system is a vital part of the specification of an EHR. However, this is not the topic of this paper.
· An information model concerned with the representation of clinical information
o The requirement for this type of model arises because
§ There are many ways of representing the same information.
§ Shared or communicated information can only be interpreted by a computer if there is either
· A single agreed way of representing information; or
· An agreed reliable and reproducible equivalence between a set of agreed ways of representing information.
o This paper explores the variety of representations of similar information and suggests an approach to reaching sufficient alignment to allow effective use of information in different systems.
o This paper also summarises the relationship between this type of information model and different functional model of the EHR.
o Like the models discussed above this model can be discussed at a high-level relatively abstract architectural view or from a nuts and bolts system design perspective. This paper is limited to the high-level view but also identifies sources for more detailed views and strategies for developing these.
In addition, more advanced clinical systems can contribute as agents in care provision through inclusion of or access to decision support systems. The role of electronic records in communication differs from the role of paper communications. Different tasks are supported the receiving computer, an agent involved in interpretation of the communicated record, is unable to comprehend natural human language. To realise the potential that computers offer for selective retrieval, aggregation and decision support the communicated information must therefore contain formally specified structures. These structures must represent the essential context and semantics of the record in a manner that is predictable and unambiguous.
Human understanding and computer interpretation
Communication between humans involves a shared understanding of a language. This includes both a vocabulary, the grammatical rules within which this vocabulary is used and the relationships between concepts represented by words in the vocabulary. The requirements for computerised interpretation of the meaning of a communication are similar. However, much of what is implicit in human communication must be made explicit to avoid misinterpretation.
Consider the following sentence that might appear in a patient record.
Examination of the fundus shows signs of haemorrhage.
A clinician reading this statement would have to consider an ambiguity – does this refer to the fundus of the stomach, the fundus of the eye or possibly the fundus of the uterus. An ambiguity of this type would be implicitly resolved by the context – if this is a gastro-enterologist’s report of a gastroscopy only one interpretation is reasonable.
Faced with the same statement a computer either needs an explicit distinction between the different fundi or a rule that explicit rule that enables distinguish these interpretations by context. Future systems may be able to “learn” these implicit rules but today this is not speculation rather than sound reality. Representing all such rules is a complex and probably open-ended process and this is only the beginning of the problem with this sentence. Consider the following questions (assuming we have stated that this is the gastric fundus).
· Is the stomach normal?
o To answer this requires the knowledge that
§ “signs of haemorrhage in the fundus” are not “normal”;
§ “the gastric fundus” is part of the “stomach”.
· Has the patient had any gastro-intestinal tract bleeding following ingestion of non-steroidal anti-inflammatory drugs?
o To answer this requires the knowledge that
§ “bleeding” and “haemorrhage” are synonymous terms
§ “the “gastric fundus” is part of the “gastro-intestinal tract”
o It also requires a check on a temporal relationship in the record between this item of information and information about ingestion of drugs. This check requires knowledge of
§ Which drugs are classified as “non-steroidal anti-inflammatory drugs”.
§ What constituted “ingestion” – presumably this excludes injections, suppositories and perhaps some other routes.
§ What is meant by “following” – does it imply a particular limited time offset?
§ How certain must the information be to record a positive response to this question. Endoscopy – as implied by the earlier statement – is fairly conclusive but what about an uncorroborated report by the patient of passing a dark stool or a fleck of fleck of blood in a vomit shortly after taking half an aspirin?
§ Is the question time limited or does it mean has this coincidence of events ever occurred in this patient?
This is just one fairly simple example of the type of problem to be addressed. A complete solution depends on many factors in the framing of questions and the completeness of recording. These factors are outside the scope of an information model. The objective of an EHR information model should be to ensure correct interpretation within the limits of the completeness of the record and the precision of the question. Given these parameters the result should be predictable, consistent (independent of the source of the information), complete (no false negatives) and precise (no false positive).
Correct interpretation of clinical information is essential to avoid inappropriate advice that may endanger a patient or incorrect analysis that may lead to inappropriate and potentially wasteful service planning.