Integrating decision support with electronic referrals Jonathan Bury, Michael Humber, John Fox Advanced Computation Laboratory, Imperial Cancer Research Fund, London, UK Abstract The Early Referrals Application (ERA) is a web-based system for managing requests for urgent referrals from primary to secondary care. ERA has been designed in the context of the UK’s “2-week standard” for the referral of patients with suspected cancer. Our aim in developing ERA has been to explore mechanisms for integrating decision support into the data-gathering, electronic referral and appointment booking processes that may accompany an urgent referral. We have adopted a clientserver architecture, which enables both the electronic referral forms and the referral criteria used by the decision support module to be easily customised by secondary care centers to reflect local variations in referral policy and data gathering requirements. We believe the system described illustrates the potential advantages of the clientserver model in facilitating technical integration, data flow and clinical communication between primary and secondary care. Keywords: Clinical Decision Support Systems; Consultation; Primary Health Care; Cancer
Referral
and
Introduction The United Kingdom (UK) Government’s white paper “The new NHS – Modern, Dependable” [1] guaranteed that all patients with suspected cancer would be able to see a specialist within 2 weeks of their General Practitioner (GP) deciding that they need to be seen urgently and requesting an appointment. In their role as gatekeeper to the UK’s secondary care services, GPs must identify those patients with a significant possibility of having cancer, so that these patients can be referred while avoiding overloading hospital clinics with patients at lesser risk. The Department of Health (DoH) publication “Referral Guidelines for Suspected Cancer” [2] is intended to help GPs identify those patients at greatest risk. These guidelines specify, for each of twelve cancer groups, clear criteria defining which patients should be referred under the “two-week” standard. Clinicians are increasingly expected to use guidelines and
protocols in their decision making. However, the commissioning and creation of such material represents only the first step towards ensuring that the intended improvements in clinical outcomes are achieved. Compliance with guidelines is enhanced if material can be referred to during the consultation [3]. Interactive decision support tools represent one approach to making clinical practice guidelines accessible during the patient encounter. By engaging the user in a dialogue about the particular case under consideration, the computer can assist in interpreting complex patterns of clinical features and provide patient specific recommendations and guidance. During the consultation period that followed the publication of the draft version of “Referral Guidelines for Suspected Cancer”, many GPs expressed enthusiasm for electronically generated referral forms and direct electronic links between primary and secondary care. At the same time, secondary care centres have had to implement mechanisms for processing referrals efficiently, and for gathering clinical data on the referred patients for epidemiological and audit purposes. Our aim has been to explore mechanisms for integrating decision support alongside the data-gathering, electronic referral and appointment booking processes that may accompany an urgent referral.
Materials and Methods Integrating decision support into the referral process We sought to identify ways in which the clinical information required from the user for decision support purposes could be re-used by other systems. In consultation with General Practitioners and other interested parties, we identified 4 processes that could potentially use this information: •
Clinical information could be used in the hospital to which the patient will be referred, acting as the referral letter.
•
Online booking systems could utilise the decision support functions to prioritise appointments.
•
The data gathered could be passed to locally held electronic health records where such records exist.
•
Data could be gathered for epidemiological
monitoring, thus contributing to the refinement of the referral criteria. To support as many of these functions as possible, we have adopted a web-based architecture which we envisage being hosted by the centre to which patients will be referred. This client-server architecture means that the information GPs provide is potentially available to other systems at the hospital, including booking and data gathering systems, and gives receiving hospitals control over the decision support system’s referral criteria and design of the onscreen referral forms. Minimal hardware and software demands are made of Primary Care practices, with only a web-browser and appropriate network access being required. Representing clinical guidelines in PROforma We used the PROforma guideline representation format [4] to specify knowledge bases defining the referral criteria for each of the 12 cancer groups described the DoH’s published referral guidelines. PROforma models clinical processes in terms of the tasks that comprise that process, and the logical rules which govern the execution of those tasks. Four types of task are described – actions, decisions, plans and enquiries (data collection tasks). These 4 task classes share a number of common attributes, such as goals, preconditions and postconditions. Additional attributes describe properties specific to each class of task, such as the data items sought by an enquiry, or the procedure associated with an action. Tasks can be assembled into networks to enable the workflow of more complex guidelines
embodying the rules governing interpreting that data, and two mutually exclusive actions, one advising that the criteria for a two-week referral were met, the other advising they had not been met. This template was then populated with the specific data items and referral rules for each of the 12 cancer groups. A small number of these required the addition of a third action, such as advising a specific investigation or a less urgent referral. The PROforma knowledge representations were created using a graphical authoring environment the “Composer” (www.infermed.com). Making this Computer Assisted Software Engineering (CASE) toolkit available to hospitals running the system would enable them to modify the knowledge representation to reflect local variations in referral policies. Solo : delivering decision support over the web “Solo” is a communications interface that enables clinical guidelines driven by a PROforma enactment engine to be run over the web [5]. The technology is made up of three main components: the Solo server, Solo servlets and a Java wrapper which encapsulates each PROforma enactment engine. Together these components control and validate the communications flow between the web client and server and the PROforma enactment engine. •
The Solo server manages a pool of available PROforma enactment engines. It allocates an engine at the initiation of an enactment session, monitors its use over the course of that session and frees that instance at the conclusion of the session.
•
Solo Java servlets work alongside the web server, supporting data management and exchange between the web browser and a PROforma engine during guideline enactment.
•
A Java wrapper encapsulates each PROforma engine running a guideline application. It manages communications between the guideline and the Solo servlets. The Java wrappers and CORBA technology ensure interoperability between the web browser components and the Prolog-based PROforma guideline execution engine.
Results : the ERA system Figure 1 - The template ERA task network, displayed in the “Composer” authoring environment. Some cancer groups require the addition of a third action, such as an urgent referral for a particular investigation. to be described. A template PROforma representation of the guideline structure was developed. This comprises a single enquiry gathering the required clinical data items, a decision task
ERA is designed to be used during the clinical consultation. Streamlining the interaction between the clinician and computer is therefore crucial. ERA is accessed using a standard web-browser. The front page offers the clinician a choice of background information, referral criteria, or a referral form for each of the 12 cancer groups. The first two of those options are hyperlinks to an HTML representation of the appropriate section of the original published guidelines, whilst choosing the third option activates the PROforma engine. The first task requested by the engine is
the data-gathering “enquiry”, presented by Solo as a referral form modelled on the look and feel of the suggested paperbased referral forms circulated by the Department of Health (Figure 2). The typical minimum number of data items required by the ERA system is around a dozen, varying between cancer groups from 6 for sarcoma to 18 for lung cancers. Data entry on the part of the clinician is minimised through the use of context dependent fields, which are
confidentiality, without burdening the consultation with administrative processes or the need to write a further referral letter. During a typical session just four web pages are encountered from first accessing ERA to a referral being confirmed (figure 4).
Figure 2 – A typical ERA data entry screen
Figure 3 – Patient specific referral recommendations
Figure 4 - ERA user interface workflow greyed out unless relevant, and negative clinical findings being the default settings. Discussion Once the data gathering process has been completed, the decision task executes automatically and the Solo system dynamically generates a web-page stating whether or not the patient described meets the criteria for a two-week referral, together with a patient specific explanation (figure 3). It is likely that GPs may occasionally wish to refer patients who do not strictly meet the referral criteria. This is fully permitted, although the user is prompted for additional free-text information in such cases. Requesting a referral generates a patient specific “referral number”, to be quoted by the patient or receptionist on contacting the hospital for a booked appointment to enable the demographic information needed for the appointment to be matched with the clinical information already provided electronically. Under this model, the clinician does not have to spend time entering demographic details during the referral process or negotiating a specific appointment time, and no patient identifiable information has been transmitted over the network. We believe this model offers benefits in supporting patient empowerment and protecting
ERA’s client-server architecture is intended to facilitate closer integration with hospital services such as referrals databases, patient record systems, and appointments and booked admission systems. This model also gives hospitals direct control over the decision support module used to comment on the appropriateness of a referral for a given patient. Likewise, local variations in data gathering requirements can be reflected in locally modified referral forms. There is a recognised need for improved epidemiological data concerning the utility of particular combinations of symptoms and clinical signs in identifying patients at risk of cancer [6]. Indeed, the paucity of such evidence was cited as one difficulty in developing “Referral Guidelines for Suspected Cancer” [2]. Referral processing software based in receiving hospitals would be well placed to support early data gathering and thus contribute the epidemiological evidence base on which such guidelines are based.
The cost of the client server model is that integration with General Practice computing systems is more difficult. The automatic extraction of clinical and demographic information from local electronic patient records would reduce the data entry burden on GPs. Saving a record of the history of the interaction with the ERA system is also attractive. Technical solutions exist to both of these requirements and we are currently discussing frameworks for their implementation with vendors of Practice Information systems.
Acknowledgments
The impact of the system on the administrative aspects of the referral process is being evaluated in trials conducted by the DoH at a number of pilot sites throughout the UK. Key issues to be explored are the extent to which ERA affects delays between referrals being requested and patients being reviewed, the impact of electronic referral forms on data collection and quality, and the attitudes of clinical and clerical staff to the changes in working practices resultant from the introduction of electronic referral. We also hope these trials will provide an opportunity to investigate the psychosocial impact of the system on the interaction and communication between patient and primary care clinicians, and establish to what extent clinicians find the decision support offered valuable.
[1] Secretary of State for England. The new NHS. London: The Stationery Office, 1997.
Conclusion We have demonstrated the technical feasibility of describing a varied and complex set of referral guidelines using the PROforma format, and of using the resultant knowledge representations to support a web-based decision support tool. The prototype ERA system demonstrates the technical viability of combining decision support alongside electronic referral mechanisms. We believe the system described illustrates the potential advantages of a clientserver architecture in facilitating integration and data flow between primary and secondary care.
Who would we like to thank Huon Butterworth, David Sutton and Ali Rahmanzadeh for their work on the Solo web-delivery software and the PROforma decision support technology.
References
[2] Department of Health. Referral guidelines for suspected cancer. London: DoH, 2000. [3] Grimshaw JM, Russell IT. Effect of clinical guidelines on medical practice: a systematic review of rigorous evaluations. Lancet 1993; 342(8883):1317-22 [4] Humber M, Butterworth H, Fox J, Thomson R. Medical decision support using Internet technologies. Submitted to Medinfo 2000 [5] Fox J, Thomson R. Decision Support and Disease Management: A Logic Engineering Approach. IEEE Transactions on Information Technology in Biomedicine 1998; 2(4):217-228 [6] Summerton N. General practitioners and cancer. BMJ 2000 Apr 22;320(7242):1090-1 Address for correspondence Jonathan Bury Clinical Research Fellow Advanced Computation Laboratory Imperial Cancer Research Fund 61 Lincoln's Inn Fields London WC2A 3PX
[email protected] http://acl.icnet.uk/
