Generating Symbolic and Natural Language Partial Solutions for Inclusion in Medical Plans

We describe the generation of partial solutions to Prolog queries posed during the design of medical treatment plans. Given a set of Prolog encoded safety principles, the queries request advise on plan revisions to conform with safety requirements. The user unfolds queries interactively, navigating a path through the solution search space by interacting with natural language representations of the Prolog terms. In this way, both symbolic and natural language representations of partial solutions can be generated. The former can be included in the plan, and the latter exported to a protocol document describing the plan. Hence, useful and informative partial solutions are still obtained despite incompleteness of the underlying knowledge base, which ordinarily would mean failure of a query. Furthermore, the user can avoid being overwhelmed by surplus solutions, and unfold to levels of detail suitable for different plans and their accompanying protocols. 1 Background and Introduction Previously, generic principles concerned with safety and efficacy of plans were abstracted from an empirical analysis of chemotherapy plans [5]. Table 1 shows textual representations of the principles at a high level of abstraction. Expressed in rule-based form, the principles have been used as constraints in software for the run-time management of complex, hazardous procedures such as chemotherapy treatment [6]. Here, we are interested in their use as obligations when devising symbolic representations of new treatment plans for export to such software; specifically in advising how a plan should be further specified, or revised, to conform to the obligations. We also focus on assisting the production of a protocol document which describes the plan, its associated hazards and their management. Protocols are required when new treatment plans are under trial or about to be introduced into clinical practice, and permission is required from ethical or regulatory bodies. They are also referenced by clinicians managing patients. Represented as Prolog rules, the safety principles can be specialised/ instantiated for inclusion in a plan and translation into useful natural language components of a protocol document. By way of illustration, the prevention principle can be represented as the following Prolog rule: Table 1. generic safety principles Exacerbation Avoid exacerbating Monitoring Monitor responses which herald anticipated hazards hazardous situations Diminution Avoid undermining the Efficacy Ensure that overall plans are benefits of essential efficacious in pursuing stated actions objectives Interaction Ensure that an action does not Sequencing Order (essential) actions temporally interact with another action for good effect and least harm to cause harm Reaction React appropriately to ameliorate Critiquing Critique proposal of certain detected hazards hazardous actions even if they are well motivated Warning Warn about hazards due to Prevention Prevent or ameliorate hazards incorrect execution of essential before executing an essential action actions advise plan( Action, Plan, Prevention, ameliorate(Action, Effect) ) :(1) part of plan(Action,Plan), effect(Action, Effect),