Pain Points in Health Care and the Semantic Web

2. The Pain Points in Health Care and the Semantic Web Advanced Clinical Application Research Group Dr. Dirk Colaert MD

3. Today Tomorrow Location Hospital Decentralized, at home Time Symptomatic, curative Preventive, lifetime Focus On the process and provider On the patient Scope Cure Patients Care for Citizens Methods Invasive Less invasive Healthcare is changing…

4. Order Process Manual Automated Experience Individual Best Practices The Process Fragmented, isolated disease mgt. Clinical Decisions Personal preferences Guide lines / evidence based De processes are changing … Information Fragmented, isolated Consolidated / complete Today Tomorrow

5. Data completeness Fragmented Consolidated Data integrity Manual/error prone Systematic mgt. and control Data access Limited, Difficult Any time, any place Technology Isolated systems Integrated systems IT is changing … Data availability Slow Real time Today Tomorrow

6. • Costs must decrease • Quality must increase – E.g. Medication errors : in the US 80.000 people died in 2004. (=8th cause of death) The health care is under pressure ...

7. The Hospital Medical Knowledge High Quality Cost Effective needs Activities Information Assessment needs produces

8. Healthcare as a Process Process Output Input Society subjective objective Medical Community Assesment operational Care Action Therapeutic Action Diagnostic Action Planning

9. Healthcare as a Process: pain points Isolated information Fragmented information Not accessable information Too much information Bad information presentation Only clinical data is kept (no knowledge) Some information is not computer usable (free text, image features, (genome in the future)) No feed back to medical community and society Complex desicions Lack of training Changing knowledge Medical errors Inefficient workflow Understaffing No operational information No infrastructure information No common language Input - Output Information Process Clinical Desicions Workflow Action Medical Community operational Society objective subjective Assesment Planning

10. Input - Output Information Process Clinical Desicions Workflow Cure for the pain points – wave 1 PAS: Patient Adminstration System HIS: Hospital Information System Result Distribution Action Medical Community operational Society objective subjective Assesment Planning Collect

11. Cure for the pain points – wave 2 PACS: Picture Archiving And Communication Sytem PAS: Patient Adminstration System HIS: Hospital Information System CIS: Clinical Information System Care Order Entry Medication prescription Result Distribution Input - Output Information Process Clinical Desicions Workflow Action Medical Community operational Society objective subjective Assesment Planning Collect Desicion support Optimization

12. Cure for the pain points – wave 3 Information filtering Decision support Semantic driven UI Clinical Pathways Evidence based medicine Clinical Trials (in- and exclusion criteria, data mining) Terminology feature extraction from unstructured or massive information (images, free text) Advanced connectivity Content Workflow optimization Intelligent patient portals Remote data capture Community HealthCare Input - Output Information Process Clinical Desicions Workflow Action Medical Community operational Society objective subjective Assesment Planning Knowledge Desicion support Optimization Common to all this is …

13. Connected Knowledge • Knowledge is a higher form of Information • Knowledge (meaning, understanding) begins when facts and concepts (information) are connected • Latin ‘intellectus’ comes from intelligere, inter + ligere = connect between • A formal description of a domain, using connected facts and concepts is called ‘an ontology’ • The W3C organization provides standards: RDF (Resource Definition Framework) , OWL (Ontology Web Language) • The “semantic web”: use the W3C standards and the inherent communication and linking properties of the WWW. • By linking ontologies they can be merged to “connected knowledge”: very powerfull but dangerous!

14. Salary Religion hobbies Simple ontology Me Audi Green owns has color Audi Opel Other Brands A3 A4 A6 Model of ABC 1234_567 Instance of

15. Knowledge: traditionally ‘assumed’ visit hypertension Tenormin Aspirin Lab Test ?

16. Connected Knowledge: explicit visit hypertension Tenormin Aspirin Lab Test Conclusion of threated by Indication for

17. Connected Knowledge: scalable

21. Connected Knowledge Examples of ontologies and rules: medical vocabulary, patient clinical data, infrastructural data Because ontologies are formaly described, computers can use them, take rules and reason about the concepts. Technologies, able to connect facts into ontologies, connect ontologies to each other and reason about it with rules gives us the means to improve vastly the current painfull processes in healthcare. Examples: Use of a Terminology Server for Controled Medical Vocabulary Decision support and clinical pathways

22. Terminology Server • Purpose: – Easy entry of data into the medical record keeping ‘freedom of speech’ and still be able to document in a uniquely defined and coded way. (e.g. ICD9) • Example – Data entry: “blindedarm onsteking” (Dutch) – Results in: ICD9 XYZ (“appendicitis”) – No single part of the search string is found in the result. This can only be achieved by a system ‘knowing’ the domain. Concept Appendix Term Appendix Term Blindedarm Concept Appendicitis Code XYZ inflamation of ICD9 code for Term for Term for

23. Decision Support and Clinical Pathways • Clinical Pathway: a way of treating a patient with a standardized procedure in order to enhance the efficiency, increase the quality and lower the costs. • Usually represented in a script book and/or flow chart diagram • Issues with conventional Clinical Pathways: – Not very dynamic: “one size fits all” • Not adapted 100% to the individual patient – Not mergeable • How can you enroll a patient into 2 pathways? – Difficult to maintain: mix op procedural and declarative knowledge

24. Agfa’s Advanced Clinical Workflow research • Combining – knowledge, declared in rules and concepts (the ontologies) • Medical domain • Clinical data about the patient • Operational (local policies) • Infrastructural (machines, people) • Workflow theory and ontology (pi-calculus) • Fuzzy sets theory and ontology • Calculating the procedure to follow: the next step(s) • After each action a recalculation is done

25. Adaptable Clinical Workflow Framework Society subjective objective Medical Community operational Assesment Care Action Therapeutic Action Diagnostic Action Planning

26. Adaptable Clinical Workflow (compare to GPS)

27. Adaptable Clinical Workflow (compare to GPS) After deviation from the calculated course the system adapts the itinerary

28. From pixel to community Guidelines Policies Clinical Data Events Requests (Local, Operational, Community, ...) Desicion support Human Interaction Recommendation Desicion Action The box is a fractal unit that can be scaled from “pixel to community”

29. Institution  Clinical Pathway Department  Order Workstation/User  Task Application  Event Region  Disease Management Country  World  Healthcare Management

30. Institution  Clinical Pathway Department  Order Workstation/User  Task Application  Event Region  Disease Management Country  World  Healthcare Management health monitoring process form generator clinical decision process workflow monitoring process task process scheduling process work list process communication and event bus: share knowledge and evidence

31. Issues when merging ontologies • Inconsistencies – Ontologies are build without other ontologies in mind. When merged they can contain contradictions. – This can be detected and brought to the attention of the user. • Semantic differences – See the example avove about “Audi” as a car and “Audi” as a brand. – Can be solved by using standard ontologies as much as possible (e.g. SNOMED in the medical domain) • Side effects – Duplicate examinations – Bad sequence – Wrong conclusions • Trust – When an external ontology is about to be merged the source must be trustworthy

32. Duplicate examinations • CP 1 – Day 1 CP1_Action1 – Day 2 Lab test: RBC – Day 3 CP1_Action3 – Day 4 CP1_Action4 • CP 2 – Day 1 CP2_Action1 – Day 2 CP2_Action2 – Day 3 Lab test: RBC – Day 4 CP2_Action4 • CP 1+2 – Day 1 • CP1_Action1 • CP2_Action1 – Day 2 • Lab test: RBC • CP2_Action2 – Day 3 • CP1_Action3 • Lab test: RBC – Day 4 • CP1_Action4 • CP2_Action4

33. Solution • By adding extra rules this can be solved. • “If the outcome of an examination is valid for x days than any duplicate examination within that period can be canceled” • These are “rules about rules” or “policies”

34. Bad sequences • CP 1 – Day 1 CP1_Action1 – Day 2 RX+contrast – Day 3 CP1_Action3 – Day 4 CP1_Action4 • CP 2 – Day 1 CP2_Action1 – Day 2 CP2_Action2 – Day 3 RX – Day 4 CP2_Action4 • CP 1+2 – Day 1 • CP1_Action1 • CP2_Action1 – Day 2 • RX+contrast • CP2_Action2 – Day 3 • CP1_Action3 • RX – Day 4 • CP1_Action4 • CP2_Action4

35. solution • Extra rule – “Examination X cannot be performed within x days after the administration of contrast medium Y” • Policy – Rules can be abstracted further into policies: – “All examinations must be checked against exclusion criteria”

36. Wrong conclusion • CP Rheuma – Rule x – Rule: If pain  Aspirine – Rule y • CP Gastric Ulcus – Rule a – Rule b – Rule … • CP Rheuma+GU – Rule x – Rule: If pain  Aspirine – Rule y – Rule a – Rule b – Rule …

37. Wrong conclusions • Because of the specific focus when making a clinical pathway, merging CP’s can potentially be dangerous. • Solution: – Detect possible patterns and add policies to cope with them. – For example: “For any medication prescription (outside the scope of the original CP), check interaction with the medical history and problems of the patient”

38. Trust • Inference engines can produce, as a side product, the proof that, what is concluded, is logically true. • We need standards to communicate and represent these proofs

39. Conclusion  Ontologies, together with theories (rules) can help health care providers to treat patients with better quality and less costs.  The intrinsic possibility of connecting ontologies and theories allow systems and people to use each others experience.  Extra policies can possibly detect and neutralize problem patterns within merged ontologies. Further research is needed here.  Scaling ontologies and theories outside the boundaries of the hospitals can be used to orchestrate effective community healthcare and regional healthcare programs.

40. Thanks