COSSAC

Risk, Events, Actions and their Consequences over Time

REACT is a technology for assisting in the creation and enactment of complex care plans.

The REACT user interface presents a planner which provides immediate feedback on the effects of events and actions over periods of care, based on constraints, interactions, and dependencies. It dynamically shows “what if” scenarios in terms of quantitative outcome measures, and presents the arguments for and against possible clinical interventions together with their justifications

Figure 1: The REACT plan manipulation system (phase II main screen). The system has four main panels. At the top is a timeline showing the events and actions making up a plan (the user manipulates this directly). Below this is a quantitative representation of risk over time. Next is a panel showing the arguments for and against any selected action. The lowest panel highlights conflicts or other alerts. As the user manipulates the plan in the top pane, the arguments and other information in the lower panels are updated dynamically.

Background

The problem of communicating and working with risk and uncertainty is a serious one for many fields, none more so than medicine. This project was motivated by the need to provide support for clinicians and patients involved in complex medical planning tasks.

Software exists for supporting clinical decision making, but generally only for a single, isolated decision - for example, what drug to prescribe, or whether to refer a patient to a specialist. Similarly theoretical and experimental work in cognitive psychology has largely concentrated on single decisions rather than the wider set of cognitive processes which are involved in framing decision-making as a part of planning and plan execution. Most decisions, however, are made in the context of plans of action, where they may interact or conflict with other planned actions or anticipated events. For example, the planning of medical interventions over a patient's lifetime is a complex proposition with many sources of information, interactions and dependencies between interventions.

For example, consider the case of a woman diagnosed as carrying a mutation to the gene BRCA1 which predisposes to breast cancer. Such a woman may have a lifetime risk of developing breast cancer approaching 80%. Following identification of the genetic factor, the woman will typically be seen by a genetic counsellor who explains what the impact of this will be, and what options are available for mitigating the risk. A number of options are available, including treatment with drugs, screening to detect tumours early in development, and surgery to remove the ovaries or breasts.

There is no "correct" plan of action in this situation, the care plan arrived at by the counsellor and patient will reflect the individual needs and plans of the patient. For example, she may be planning to have children, and will need to avoid ovarian surgery and certain drugs. She may plan to have ovarian surgery in a few years time but to wait until much later in her life, when breast cancer risk is higher, before considering breast surgery. Each decision involved in forming the plan is influenced by previous decisions, and can influence later decisions.

Experimental evidence (Klein, 1998; Smith-Spark et al. 2005) and consideration of the likely cognitive loads imposed by plan manipulation (Glasspool et al. 2003) indicate that working with medical care plans is a demanding task even for experts in the field, let alone the patients whose care is at stake.

Theory

Relatively little work has been done on understanding the cognitive processes involved in planning. The REACT project aims to 1) develop a theoretical understanding of these processes, and to link this to practical support techniques that could be provided by a computer system; 2) to design and develop computer software that could be used clinically, and finally 3) to use the software and psychological experimentation to empirically validate the theoretical position.

The project is developing a theoretical view of judgement, decision making and planning which encompasses the full set of processes surrounding these abilities, including the assessment of decision options, the bringing to bear of diverse bodies of knowledge in decision making, and the ability to reason and communicate about the decision making process.

Our approach is to relate the theoretical approach directly to practical means by which decision support software can assist the cognitive processes of planning, through consideration of the cognitive loads involved in the various processes of plan manipulation. Experimental work has established that the most significant of these cognitive loads did indeed play a part in the difficulty of planning and could be mitigated by appropriately designed computer feedback.

Based on these results we have proposed that the major sources of difficulty for an unaided person manipulating a complex plan involve:

1. Maintaining an evolving plan in memory as it develops
2. Identifying which options for action are available at each step in the plan
3. Deciding which of these options should actually be taken
4. Keeping track of constraints on, and dependencies between, planned actions
5. Keeping track of the effect of the plan as a whole with respect to its goals.

The REACT (Risk, Events, Actions and their Consequences over Time) software is designed to provide targeted support for these cognitive loads (see figure 1 above).

The user interface

The software provides an interactive planning chart on which a plan may be drawn and modified, and it provides immediate feedback, while the plan is being changed, of the consequences of any changes and any important information the user should be aware of. REACT can be applied in any situation where plans must be made in the context of a great deal of complex and potentially conflicting information, but we have applied it to medical care planning for the purposes of this project. The REACT software can be used to support an expert planner, or to support joint planning between an expert and a novice. This is the situation for example in the type of medical planning task discussed above, where a clinician and a patient jointly work on a care plan.

Evaluation

An evaluation study was carried out using actors to play patients in a simulated clinical situation like the one discussed. Clinicians used the REACT software to jointly work on a care plan with the simulated patients. The results were highly encouraging, with seven of the eight participating clinicians very positive about using the software clinically, and all feeling the software to be worthwhile.

Demonstration videos

Some demonstration videos made with an early version of the Phase II software are available here. Later versions of the software are significantly more capable but the basic concepts are illustrated in these videos.

Support

Phase I (1999-2001) supported by Cancer Research UK and by the UK Economic and Social Research Council programme in cognitive engineering (under the RAGs project).

Phase II (2001-2005) supported by the ESRC and EPSRC PACCIT Programme, award L328 25 3015 (design, analysis, psychological and technical underpinnings and evaluation and later software development) and by Cancer Research UK incubator funding (initial software development and clinical knowledge base authoring).

Phase III (2005 onwards) supported by Cancer Research UK and by the EPSRC "Exploration Studies for Grand Challenges within the Information-Driven Health Initiative" programme under the "Safe and Sound" project.

Publications: REACT-related publications listed in the COSSAC database.

Contact David Glasspool: dglasspo at inf.ed.ac.uk