Programme of Work : 2NRICH Project

Faculty of Technology and Environment

Quick Links

OverviewProject AimsProposed ResearchProgramme of WorkPartnersDeliverablesStatusStaffLinksContact

The proposed work will be organised into four work packages focused on reasoning models, to understand and take into account key aspects of current practice in decision making by clinicians, and decision support offered to patients; assembling the statistical evidence, which involves building and evaluating inference models for outcome of treatment, as well as generating data segments together with associated visualisation maps; development of specialised user interfaces, to make this complex information readily accessible at the requested level of detail; and, results generalisation and dissemination, which will drive the beginning and end of the project.

Understanding
We will investigate current clinicians' and patients? decision practices focusing on modelling decision processes, reasoning models, information/evidence gathering, objectives prioritisation and decisions validations. The novel aspect of this work will be an understanding of clinicians and patients decision and validation processes and a formal documentation of decisions processes together with multi-dimensional decision-making concerns.

Why do we need to study existing decision practices?
Cancer clinical decision-making is a risk-critical activity, where verification of the reasoning model by clinicians is key. Thus, it is essential to understand current decision processes to develop new health informatics system that are consistent with current decision-making practice, before embarking on the provision of augmented facilities for clinicians. In addition, it is not clear how different multi-dimensional decision concerns, for instance of the different members of the patient care team, interact together so that relevant evidence can be identified and presented to the user.

What will be the results of this workpackage?
A report describing clinicians and patient decision process map, which outline the decision process and support systems? requirements, including the following aspects:

Benchmarking a range of statistical methodologies for predicting survival and forming prognostic group allocations, across the participating clinical centres, using historical data.
Development of rule-based models of clinical reasoning in relating the patient condition to the choice of treatment.
Application of the prognostic group allocations and rule-based models to search the database, of patient records in order to recall the most relevant historical cases to inform on queries regarding an individual patient.
Specification, design, implementation and evaluation of GUIs for clinicians.
Initial controlled trials to evaluate system performance, its value for clinicians and its possible impact on decisions regarding treatment.

How will the work be done?
In a participative fashion our clinical partner will be involved in conducting a series of structured interviews with clinicians to understand the decision and validation models, and any tools and evidence used in relation to different type of decision dimensions. We will employ activity-network modelling based on the R?e and Activity Diagram (RAD). The models will be combined with a novel application of the Resources Description Framework (RDF) to describe decision models, semantics, data and knowledge sources, and objectives as systems resources.

Statistical evidence
A pilot study of a composite prognostic index of survival, has been carried out over the last 4 years in collaboration with Christie Hospital. This study benchmarked a standard method that is the workhorse of survival modelling in medical statistics, proportional hazards (Collett 1996) against generic non-linear approaches for the modelling of censored data with a fine time-scale, using the Partial Logistic Artificial Neural Network, PLANN (Lisboa et al 2000). In a monthly study of 5-year survival it was found that prognostic groups derived from PLANN for an ?operable? groups with small cancers and non distant metastasis were slightly more specific that those from proportional hazards, and comparable in the prediction of survivorship when validated retrospectively with records from a further 931 patients. In the remaining patients with higher mortality risk, a comparison between the linear and non-linear methods generated hypotheses about candidate pairwise interactions which, when tested with proportional hazards, show quite a different prognostic grouping from those generated by either of the previous models. In particular, retrospective validation with records from a further 334 patients appears to show that the composition of this high risk cohort changed from 1983-9 to 1990-3, which are the recruitment periods for the design and validation data, respectively (Lisboa et al 2001a, 2001b). These results show that prognostic groups can be reliably derived for patients recruited over a long period of time in one clinical centre, and are capable of providing powerful and robust evidence in support of prognostic inferences for these patients.
The novel aspects of this work will be a multi-centre evaluation of standard scoring systems together with inference models of increasing complexity, as well as the application of powerful and robust neural network methodologies to segment the database and to visualise it as a two-dimensional map.

What is the role of statistical evidence in this study?
Evidence is at the core of the advice given on expected survival outcome, but it is not enough to present clinicians with a large array of decision tools, models and relevant evidence without the provision of a flexible mechanism to enable them to trace back the consistency of the reasoning used by the model with established knowledge, and to construct alternative scenarios through changing the decision-support composition, i.e. the configuration of the reasoning model.

User interfaces
We will investigate improved methods for user interfaces to access, search and visualise high-dimensional data and its customised presentation for a range of users including; clinicians and patients. This will be informed by the process studies conducted in workpackage 1. The novel aspects of this work will be in the area of intelligent user interfaces for adaptive decision-support systems, and high-assurance use of distributed data and knowledge sources.

Why is it needed?
Evidently, decision-support of critical activities as in cancer care catering for potentially diverse range of IT and analytical skilled users, software systems usability is central to the development and acceptance of the proposed decision-support system.

Generalisation and Dissemination
The system specification is key stage of the project closely aligned with the initial stages of workpackage 1. However, an additional and important facet of the requirements specification is the consideration of the psychological effects of providing information regarding cancer treatment, to patients (Maguire 1999a, Maguire 1999b ). Specialist advice will be sought in this area.

We will undertake a series of systems evaluations and generalisation studies so that the results of the work with the aim of making the results applicable across a range of specialist cancer referral centres. Advice will be sought from the North West Public Health Observatory with relation to the design of performance evaluation measures, which will be combined with methodologies related to (Vellido et al 2000) that are currently being applied to formulate employee and customer decision metrics in business. During this period, we will also be involved in increased publicity for the project work, by dissemination at medical as well as computing conferences. Also, we will create a project Web site for the general dissemination with links to related sites and projects and initiatives.

Project Title

Towards a Disciplined Approach to Integrating Decision-Support Systems for Breast Cancer Care Activities