Notes

Outline

Modelling clinical goals A corpus of examples and a tentative ontology John Fox 1, Alyssa Alabassi1,2, Elizabeth Black1, Chris Hurt1, Tony Rose1 1Advanced Computation Laboratory, Cancer Research UK 2Guy’s Hospital Breast Unit, London

Modelling clinical goals Models of goals: why and how Goals in PROforma Armchair ontology of goals Some proposals for goal models A corpus of examples (breast cancer) Revised ontology & goal model Conclusions

CREDO model of cancer care

“ A simple, flexible model” (with apologies to Gunther Schadow)

Why are goals important? Understanding the clinical process Rationale of clinical tasks  (Shahar) Critiquing, quality assessment (Advani) Management of clinical workflow Recovery from task failure Tasks become irrelevant/inappropriate Defractalisation of acts

Goals and tasks in PROforma

“ A simple, flexible model” (with apologies to Gunther Schadow)

Goal ontology (version 1) q Knowledge goals           Decide between alternative hypotheses about world ·          Detect e.g: presence/absence an abnormality ·          Classify e.g: which of N possible conditions is present ·          Stratify e.g: level of risk ·          Predict: e.g. diagnosis, prognosis   Acquire information about setting q Action goals           Achieve ·          Eradicate e.g: eradicate an infectious organism ·          Create e.g: create a sterile site           Control ·          Prevent e.g: prevent side-effect of a treatment ·          Limit e.g: maintain physiological parameter within limits           Communicate ·          Enquire e.g: request an appointment ·          Inform e.g: tell colleague results of test

Properties of goals Shahar (1998) Whether the intention is to achieve, maintain or avoid a situation; Whether the intention refers to a clinical state or action; Whether the intention holds during care (intermediate) or after it has been completed  (overall)

Properties of goals Hashmi, Boxwala et al (2004) Context in which goal is relevant Target e.g. state of disease or disorder Verb that specifies whether the target is to be achieved, avoided, etc Temporal constraints. Priority of the goal

“Formal” properties of goals Winikoff et al (2002) Known, goals must be explicit Consistent, goals must not conflict Persistent, while success conditions not satisfied Unachieved, drop as soon as satisfied Possible, abandon if impossible

Improving the ontology A large corpus of examples would help to identify the range of functions supported by goals Systematic classification of the examples would help to understand the main goal types and semantics CREDO provides one domain for developing such a corpus

CREDO model of cancer care

CREDO service description: triple assessment Clinical services Decision support (investigations, follow up, genetic risk) Tracking results and investigations. Management of follow up or discharge Patient services Personalised schedules Communication services Notifying physician of results, management, discharge plan Notifying patient of results and management plan. Inviting patients for follow up and investigations.

Review of CREDO corpus against Shahar model

Review of CREDO corpus against Hashmi, Boxwala model

Revised ontology Knowledge goals o        Acquire knowledge about specific setting [15 instances] §        … o        Decide between alternative hypotheses about the world [52 instances] §          Detect §          Classify         … §          Predict … Action goals o        Achieve some state of world [65 instances] §          Limit changes to current state §          Bring about required future state        … §          Decide between alternative interventions … o         Enact tasks [90 instances] §          Arrange service §          Investigate §          Communicate

Consensus model? <Situation> Context (Hashmi) Scenarios (e.g. Prodigy) Triggers and preconditions (e.g. PROforma) <Verb phrase><Noun phrase> Task, Focus (Huang et al) Verb, Object (Fox et al) Verb, Target (Hashmi et al) Verb phrase, Noun phrase (Kelly, safety goals) Performative, message (KQML, FIPA) <Constraints> Temporal (Shahar, Hashmi) Scheduling (Peleg et al) Cost and other resources? Other requirements? <Control> Priority (Hashmi) Urgency, importance, deontic

Example “if a patient presents with symptoms of possible breast cancer then it is obligatory that the patient is referred to see a specialist oncologist within two weeks”

Example <Situation> <Verb phrase> <Noun phrase> <Constraints> <Priority> if  patient presents with symptoms of possible breast cancer refer patient specialist oncologist within two weeks obligatory

Conclusions Guideline enactment systems should explicitly support clinical goals and intentions Several existing attempts to model goals have yielded proposals for possible structures and semantics Analysis of a corpus of examples in the domain of breast cancer suggests further refinements Analysis of other typical corpora would yield further refinements