Triple Assessment

Decision support for the initial assessment of patients with suspected breast cancer.

The Triple Assessment Decision Support (TADS) system was developed with two main aims: to investigate whether computer-enacted guidelines can significantly enhance breast clinicians' compliance with best practice, and to collect feedback on system usability.

Background

A key early part of the UK NHS breast care pathway is commonly referred to as Triple Assessment. Women with breast-related symptoms that raise suspicion of breast cancer are referred by their GP to designated hospital-based breast clinics called Triple Assessment clinics. The first element of the assessment consists of a clinical examination and the gathering of patient history and demographic data. If this initial examination reveals an abnormality, the patient is referred for imaging andneedle biopsy. […….incomplete - needs to fit with 4 decision points below?

Workflow

Four key inter-dependent decisions were identified in the triple assessment workflow for implementation in the system:

Risk assessment decision: genetic risk assessment (calculated as low, medium or high) covered as part of clinical history-taking.
Imaging decision: radiological investigations to perform such as mammogram, ultrasound, both or none.
Biopsy decision: method to perform biopsy such as FNA, core biopsy.
Management decision: whether to refer the patient to a multi-disciplinary team and/or to a geneticist, to discharge or to follow up (in high-risk surveillance cases).

Knowledge base

The system's knowledge base was built from conventional sources: national and international clinical practice guidelines and medical literature. The Triple Assessment pathway has wide scope and involves multiple, diverse tasks in areas such as genetics, imaging and pathology. No single clinical practice guideline (CPG) had been published in the breast cancer domain at the time of the system's development (DATE) covering all relevant knowledge. In consultation with domain experts, therefore, seven high quality evidence based guidelines were selected as the basis for specifying the logical argumentation for and against each candidate option, for each of the four decisions. The AGREE guideline appraisal instrument was used to determine the quality of each selected CPG. All guidelines scored highly on the AGREE scale suggesting a high level of rigour, but only two of them (NICE and NCCN) provided recommendations in a clear and easy-to-understand form. The other CPGs were more narrative in form so considerable clinical input was required to transform the knowledge they contained into a logical form that the system could use.

An expert panel was formed including four senior practicing consultants from four relevant disciplines (surgery, radiology, pathology and genetics) to review the knowledge base for its accuracy in encoding the guidelines. In all, a set of 125 arguments was specified to replicate exactly the expert panel's recommendations for the system's set of decisions.

Knowledge modelling

The knowledge required for the TADS system was modeled using CommonKADS [Schreiber et al, 2000]. Development work was split into three phases. The first phase focused on an analysis of the organisation in which TADS was to be used, describing the care processes, resources, and knowledge assets of the organisation and the expected impact of the CDS system on the organisation. Interviews were conducted with breast surgeons, a pathologist and a radiologist to gather inputs about clinical processes, workflows and functional requirements. In the second (conceptual analysis) phase, we abstracted the system's clinical knowledge together with the reasoning to be performed, and the interactions with users and other external agents. In the third phase, the TADS system itself was constructed using the PROforma decision and workflow modelling language [Sutton and Fox, 2003] and its associated Tallis process modelling technology.

A simplified PROforma view of the triple assessment workflow within TADS is shown in Figure 1.

Figure 1. A simplified view of triple assessment as a task network (as displayed in the Tallis
authoring tool). This shows the main plan which embodies a simple linear workflow with
conditional branches depending on the imaging modalities used (mammography and/or
ultrasound). The four decision nodes are represented by circles and are embedded at
various points in the workflow.

Decision modelling

A key feature of TADS is the argumentation-based approach supported by the PROforma decision model. PROforma provides a clinically-intuitive solution to the task ofidentifying and presenting clinical evidence to relate a specific patient state to aguideline recommendation (see Figure 2). A set of arguments 'for' and 'against' eachpatient management option is generated by the TADS decision engine, and thearguments are aggregated in a summary where options are listed in order of preference determined by an overall assessment of benefit or harm. The user can also see the medical reasons for andagainst each option, and can access the source evidence via hyperlinks. In this way TADS can deliveradvice based on guidelines while adding patient-specific decision supportand other data and knowledge management services. Theclinical user is always free to override system recommendations.Another advantage of the argumentation-based approach is in helping the user to takeproperly informed decisions by exposing any conflicts in the underlying evidence. This is not an uncommon scenario in medicine due to its inherent uncertainties and the constantpublication of new research findings and analyses. For example, two different CPGs may disagree about a particular management option due to differences in the views ofpanellists from different communities. In such a case, TADS can provide both thearguments and the underlying justification used by respective guidelines allowing the clinician to make a fully-informed judgement.

TADS screen

Figure 2. TADS screen showing decision options for the imaging for one case, to be taken after medical history and examination. The system recommends an ultrasound scan but recommends against mammography and against doing nothing. For the decision option 'Do an ultrasound of the affected area', argument has been expanded to show the justifying evidence (an option available to the clinician for all decisions, options and arguments). Links are provided to the relevant supporting literature, which can be accessed by the user if required (e.g. from PubMed)

Demonstration

Demonstration of Triple Assessment Decision Support

Study

The TADS system was tested in a crossover experiment with a balanced block designin which experienced breast surgeons were asked to consider a set of simulated patient cases, both with and without decision support. Twenty four senior breast clinicians, experienced in conducting triple assessment clinics, participated in the study. Fifteen simulated case histories were constructed from a larger number of real cases and reviewed by the expert panel for adequacy of data, internal consistency and representativeness. Three sets of five patients were created, each encapsulating a variety of clinical scenarios.Each participating clinician was asked to consider two different sets of cases: oneset with and one set without decision support. To control for differences in the difficulty of the case sets, the three sets of five cases were also balanced so that each set was included the same number of times in each arm of the trial. We referred to each patient taken through the four decisions by one clinician as a "patient journey", so in total 120 patient journeys (5 cases times' 24 clinicians) were made with decision support and 120 without.

The system proved to be very effective in improving adherence to best practice as defined in the literature and in guidelines. The number of deviations from bestpractice was significantly lower when TADS provided decision support to the clinicians.

An obvious limitation of the trial was the use of simulated cases rather than real patients, so the results need to be treated with caution.Our simulated cases were also deliberately more varied than one would find in a typical triple assessment clinic in order to test the effect of the software on differing scenarios.

The strict workflow-based design of TADS imposed a degree of rigidity by compelling study participants to take the four decisions in the same order. If we were to trial the software in a real clinical setting, users would be given the ability to enter data and make decisions in any order they choose. The TADS trial should be viewed as a proof-of-concept that a task-network technology, in this case PROforma, can support complex clinical workflows and provide guideline-based decision support in a way that is effective and usable.

References

Patkar V, Fox J. Clinical guidelines and care pathways: a case study applying PROforma decision support technology to the breast cancer care pathway. Stud Health Technol Inform, 2008;139:233-42.

V Patkar, C Hurt, R Steele, S Love, A Purushotham, M Williams, R Thomson, and J Fox. Evidence-based guidelines and decision support services: a discussion and evaluation in triple assessment of suspected breast cancer. Br J Cancer. 2006 December 4; 95(11): 1490–1496.

Patkar V, Hurt C, Steele R, Purushotham A, Love S, Williams M, Thomson R, Fox J. Impact of a computerised decision support system on compliance with evidence-based guidelines for triple assessment. Proc. AI techniques in healthcare: evidence-based guidelines and protocols, Riva del Garda, Italy, 29 August 2006.

Sutton DR, Fox J. The syntax and semantics of the PROforma guideline modeling language. J Am Med Inform Assoc. 2003 Sep–Oct; 10(5): 433–443.

Agree Instrument.

G. Schreiber, H. Akkermans, A. Anjewierden, R. de Hoog, N. Shadbolt, W. Van de Velde and W. Bob. Knowledge Engineering and Management: The CommonKADS Methodology, 1st edn, The MIT Press, 2000.