Argumentation theory has evolved from its original primary context as a sub-discipline in philosophical logic, to emerge in the last ten years as an important area of logic-based Artificial Intelligence (AI). Within AI, research in Argumentation theory has produced significant contributions to the modelling and analysis of defeasible reasoning and the development of formal methods for negotiation and dialogue processes in multiagent systems. Within healthcare, argumentation methods have been used for decision making in a variety of areas (including prescribing, risk assessment and management and therapy selection).

The first area where argumentation methods were developed for use in healthcare was in decision-making. The classical approach to formal decision-making is based on expected utility theory, a quantitative technique for arriving at "rational" decisions which weigh costs and benefits of the outcomes of decisions, weighting these with the uncertainty that particular events and outcomes will actually occur, represented as probabilities. Despite its power and elegance, the difficulty with expected utility theory and similar approaches is that they have a voracious demand for data (to estimate the probabilities), and probabilistic and utility based inference are unnatural to clinicians and difficult to understand for patients. In the early to mid 1990s, John Fox and colleagues at Cancer Research UK therefore proposed an approach to decision-making under uncertainty in which logical methods are used to develop arguments for and against competing clinical hypotheses (e.g. diagnoses) or actions (e.g. therapies). Despite its simplicity, the approach has proved to be surprisingly effective for constructing practical decision support systems, and it is sufficiently natural that it has high acceptability to clinicians [see, for example, Fox et al, 2006].

LA, a formal logic of argument, also provides the foundation for decision making in the PROforma guideline modelling language. Argumentation techniques and PROforma have been used to develop a number of clinical applications, including CAPSULE (supporting prescribing in primary care), RAGs (support for genetic risk essessment in breast and ovarian cancer), REACT (support for the creation and enactment of complex care plans) and StAR (toxicological risk assessment) (see links below).

More recent models of Argumentation have become increasingly concerned not only with the formulation of individual arguments for beliefs or actions but also how arguments may interact, particularly how certain kinds of argument can attack and defeat other arguments. To illustrate, consider the following general argument which justifies an action (a natural language representation is given rather than a symbolic logical representation)

If it is believed that belief1, ……, beliefn is the case
Then we should do action a
Since this will result in effect e being the case
Which will realise our desired goal g.

The above generic argument scheme can be specialised?? to a specific medical context as follows (again a natural language rather than symbolic representation is given):

Argument A1 =
If it is believed that the patient has had a myocardial infarct
Then we should administer aspirin
Since this will result in reduced platelet adhesion
Which will prevent blood clotting.

One might also construct a conflicting argument justifying an alternative action for realising the same goal:

Argument A2 = If it is believed that the patient has had a myocardial infarct
Then we should administer chlopidogrel
Since this will result in reduced platelet adhesion
Which will prevent blood clotting.

A further argument against giving aspirin may then be introduced:

Argument A3 = If it is believed that the patient has a history of gastritis
So that administering aspirin
Will risk gastric bleeding.

The constructed arguments can now be organised into a network of arguments related by a defeat relation, allowing one to reason as follows:

A1 and A2 conflict with each other, but A1 is a stronger argument than A2 since a clinical trial indicates that aspirin is more efficacious at preventing blood clotting than chlopidogrel. Hence A2 defeats A1. However, there is an argument A3 that defeats A1 on the grounds that aspirin results in an unwanted side-effect. Hence, argument A2 is reinstated as the winning argument, i.e., chlopidogrel is the preferred choice of action.

Reasoning of the above type has essentially been axiomatised in terms of logical "calculi of opposition" that when applied to a network of constructed conflicting arguments, determines the winning arguments. Note that beliefs themselves are subject to argumentation. For example, the belief that the patient has had a myocardial infarct is itself the result of constructing arguments for alternative diagnoses and then determining the winning arguments and hence the preferred diagnoses.

The benefits of argumentation techniques have recently been reported on in a AAAI workshop on "Argumentation for Consumers of Healthcare" (see below).

Other recent work has included a growing body of research into formal theoretical models of argumentation and their application. For example, the EU ASPIC (Argumentation Services Platform with Integrated Components) project is formalising models of argumentation-based decision making, dialogue and learning, and migrating these models to software implementation for deployment in multi-agent and standalone applications. The ASPIC project is developing a number of medical applications in order to demonstrate the utility of argumentation-based reasoning.

In Spain, the CARREL project [Vazquez-Salceda J et al, 2003] is using the recent approaches to argumentation theory in a transplantation application. The aim of CARREL is to increase the number of human organs that are made available for transplantation from potential donors, by supporting argumentation-based deliberation over the viability of an organ. (Currently, if a physician representing the potential donor deems an organ to be non-viable then it is not transplanted. However this ignores the possibility that physicians representing a potential recipient may have a stronger argument claiming that the organ is viable).