1 / 41

Automated Reasoning for Application of Clinical Guidelines

Computational Thinking to Support Clinicians and Biomedical Scientists June 21–22, 2011. Automated Reasoning for Application of Clinical Guidelines. Mark A. Musen, M.D., Ph.D. Mary K. Goldstein, M.D., M.Sc. Samson W. Tu , M.S. GLINDA: GuideLine INteraction Detection Architecture.

colum
Télécharger la présentation

Automated Reasoning for Application of Clinical Guidelines

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Computational Thinking to Support Clinicians and Biomedical Scientists June 21–22, 2011 Automated Reasoning forApplication of Clinical Guidelines Mark A. Musen, M.D., Ph.D. Mary K. Goldstein, M.D., M.Sc. Samson W. Tu, M.S.

  2. GLINDA: GuideLineINteraction Detection Architecture • Computational methods for reasoning about evidence-based practice • Mechanisms for dealing with the messiness of clinical situations • Application of multipleclinical-practice guidelines • Adjustments for patient co-morbidities • Adjustments for interactions among interventions

  3. GLINDA Project Team • Mark Musen, M.D., Ph.D.1 • Mary Goldstein, M.D., M.Sc.1,2 • Samson Tu, M.S.1 • Susana Martins, M.D., M.Sc.2 • CsongorNyulas, M.S.1 • Hyunggu Jung, M.S.1 • Pamela Kum,M.S.2 1 Stanford University, Stanford, CA 2 VA Palo Alto Health Care System, Palo Alto, CA

  4. Clinical Context of our Work • Populations are aging worldwide • Older adults tend to have multiple chronic conditions 75 million people in the US have two or more concurrent chronic conditions • Management of multiple co-morbidities presents challenging problems • Competing therapeutic goals • Interventions that interact • Difficulty achieving parsimonious treatment plans

  5. Role of Clinical Practice Guidelines • Clinical practice guidelines define evidence-based best practices • Lots of work on automating CPGs EON, InterMed (GLIF), SAGE, PROforma, Asbru, … • Almost all CPGs—and all systems to automate treatment in accordance with CPGs—focus on single diseases

  6. SYNTHETIC PATIENT DATA

  7. Simplified ATHENA Architecture Electronic Medical Record System Patient Data VISTA Hierarchical Database in M Treatment Recommendation CPRS Guideline Interpreter ATHENA HTN Guideline Knowledge Base SQL Server: Relational database Data Mediator

  8. ATHENA HTN KB

  9. ATHENA HTN Knowledge Base

  10. Goals Messages Action Choices SYNTHETIC PATIENT DATA ONLY

  11. ATHENA-HTN Evaluation Studies VISN 1 Study: 2008–2009 50+ Providers 7,000+ Patients 11,000+ Clinic visits VISN 1 sites: Bedford, MA Boston, MA Manchester, NH Providence, RI West Haven, CT San Francisco VA Three-Site Study: 2002–2003 50+ Providers 5,000+ Patients Almost 10,000 Clinic visits Palo Alto VA Durham VAMC, North Carolina

  12. Encoded Guidelines ATHENA Hypertension ATHENA Heart Failure ATHENA Hyperlipidemia ATHENA Diabetes ATHENA Kidney Disease ATHENA Opioid Therapy

  13. Limitations of Single-Disease Guidelines [Boyd et al. JAMA 2005] • Simultaneous application of multiple guidelines leads to suboptimal care • Hypothetical 79-year-old woman with chronic obstructive pulmonary disease, Type 2 diabetes, osteoporosis, hypertension, and osteoarthritis • If the relevant CPGs were followed, the hypothetical patient would be prescribed 12 medications and a complicated, pharmacologically inappropriate regimen • Application of CPGs needs to • Detect and repair conflicting interactions • Prioritize recommendations

  14. Recommendations for Hypertension

  15. Recommendations for Hyperlipidemia

  16. Recommendations for Diabetes

  17. Overview of GLINDA Approach • Incorporate our extensive experience with ATHENA CDS in an agent-oriented architecture • Use task–method decomposition to create agent-oriented model of procedural elements • Develop ontology of guideline interactions • Develop agents for detectingconflicts, repairing conflicts, prioritizing and integrating treatment recommendations

  18. GLINDA Task–Method Decomposition Multi-guideline CDS Get Data Select Guideline Apply Guideline Consolidate Advisories Detect Interactions Repair Prioritize DB query ATHENA ATHENA w/ Additional Knowledge Source Heuristic Rules based on Interaction Ontology Interaction-Specific Strategy Weight of Support Simple Aggregation Manual selection Goal satisfied? Combine Results Get KS Apply Guideline Framingham risk calculator ATHENA

  19. Modeling tasks and methods in Protégé

  20. Modeling tasks and methods in Protégé

  21. Modeling tasks and methods in Protégé

  22. Implementation of Tasks and Methods in an Agent-Oriented Architecture Task–method decomposition System agent Data Sources Blackboard Agent GLINDA workflow Task 8 Task 1 Task 3 Task 4 Task 6 Task 7 Task 5 Task 2 System agent Knowledge Sources

  23. Running GLINDA – Initializing … Task–method decomposition Configurator agent Patient Data Blackboard Agent GLINDA agent configuration Controller agent ATHENA HTN KB

  24. Running GLINDA – Creating Agents Task–method decomposition Configurator agent . . . Patient Data Blackboard Agent GLINDA agent configuration Prioritization agent GL Interaction agent ATHENA agent #n Select GL agent Get Data agent Repair agent Consolidator agent ATHENA HTN agent Controller agent ATHENA HTN KB

  25. Running GLINDA – Configuring Agents Task–method decomposition Configurator agent . . . Patient Data Blackboard Agent GLINDA agent configuration Prioritization agent GL Interaction agent ATHENA agent #n Select GL agent Get Data agent Repair agent Consolidator agent ATHENA HTN agent Controller agent ATHENA HTN KB

  26. Running GLINDA – Activating Agents Task–method decomposition Configurator agent . . . Patient Data Blackboard Agent GLINDA agent configuration Prioritization agent GL Interaction agent ATHENA agent #n Select GL agent Get Data agent Repair agent Consolidator agent ATHENA HTN agent YES/NO HF, HTN, CKD Controller agent ATHENA HTN KB 12345

  27. Running GLINDA – Get Data Task–method decomposition Configurator agent . . . Patient Data Blackboard Agent GLINDA agent configuration Prioritization agent GL Interaction agent ATHENA agent #n Select GL agent Get Data agent Repair agent Consolidator agent ATHENA HTN agent Controller agent <XML> ATHENA HTN KB

  28. Running GLINDA – Select Guideline Task–method decomposition Configurator agent . . . Patient Data Blackboard Agent GLINDA agent configuration YES/NO YES/NO YES/NO Prioritization agent GL Interaction agent ATHENA agent #n Select GL agent Get Data agent Repair agent Consolidator agent ATHENA HTN agent Controller agent ATHENA HTN KB

  29. Running GLINDA – Run ATHENA Agents Task–method decomposition Configurator agent Blackboard Agent GLINDA agent configuration Prioritization agent ATHENA HTN agent ATHENA CKD agent GL Interaction agent Repair agent Consolidator agent ATHENA HF agent ATHENA Lipid agent ATHENA DM agent Controller agent ATHENA HTN KB patient data [labs, probs.] goal patient classification action choices messages patient data [labs, probs.] goal patient classification action choices messages patient data [labs, probs.] goal patient classification action choices messages patient data [labs, probs.] goal patient classification action choices messages patient data [labs, probs.] goal patient classification action choices messages

  30. Running GLINDA – Consolidate Advisories patient data [labs, probs.] goal patient classification action choices messages Task–method decomposition Configurator agent Blackboard Agent GLINDA agent configuration Prioritization agent ATHENA HTN agent ATHENA CKD agent GL Interaction agent Repair agent Consolidator agent ATHENA HF agent ATHENA Lipid agent ATHENA DM agent Controller agent ATHENA HTN KB patient data [labs, probs.] goal patient classification action choices messages patient data [labs, probs.] goal patient classification action choices messages patient data [labs, probs.] goal patient classification action choices messages patient data [labs, probs.] goal patient classification action choices messages patient data [labs, probs.] goal patient classification action choices messages

  31. Running GLINDA – Calculate Interactions patient data [labs, probs.] goal patient classification action choices messages Task–method decomposition Configurator agent patient data [labs, probs.] goal patient classification action choices messages Blackboard Agent GLINDA agent configuration Prioritization agent ATHENA HTN agent ATHENA CKD agent GL Interaction agent Repair agent Consolidator agent ATHENA HF agent ATHENA Lipid agent ATHENA DM agent Controller agent ATHENA HTN KB ! ! !

  32. Running GLINDA – Repair and Prioritize Task–method decomposition Configurator agent patient data [labs, probs.] goal patient classification action choices messages Blackboard Agent GLINDA agent configuration patient data [labs, probs.] goal patient classification action choices messages 1. 2. 3. … Prioritization agent ATHENA HTN agent ATHENA CKD agent GL Interaction agent Repair agent Consolidator agent ATHENA HF agent ATHENA Lipid agent ATHENA DM agent Controller agent ATHENA HTN KB ! ! !

  33. Ontology of Cross-Guideline Interactions Among Recommendations

  34. Example 1: Contradictory Recommendations Repair agent ATHENA HTN agent GL Interaction agent ATHENA HF agent Contra-indication X

  35. Example 2: Inconsistent Patient Characterizations Conclude CKD if abnormal eGFR Conclude CKD if abnormal eGFRs > 3 months apart GL Interaction agent ATHENA HTN agent Repair agent ATHENA CKD agent Inconsistency Message

  36. Example 3: Cumulative Number of Interventions Repair agent ATHENA HTN agent GL Interaction agent ATHENA Lipid agent Constraint Violation Adjust strength of support for drug addition

  37. Use of patient data to drive our work • We extracted 2455 complex, deidentified patient cases from the Stanford Translational Research Integrated Database Environment (STRIDE) • We are applying our method for interaction detection to 226 selected cases selected for their combination of diseases and number of drugs • Formative evaluation of system performance drives knowledge-base evolution https://clinicalinformatics.stanford.edu/research/stride.html

  38. Conclusions • Systems that assist with guideline-based care need to address the messiness of actual clinical situations • An agent-oriented architecture allows for • Reasoning about comorbidities, application of multiple guidelines, and situation-specific interactions • Flexibility in experimenting with alternative computational workflows • Creating GLINDA will drive development of formal models for computational thinking about • Guideline interactions • Repair mechanisms • Prioritization of interventions

  39. GLINDA This work has been supported by the National Library of Medicine Any opinions expressed here are not necessarily those of the NLM or of the Department of Veterans Affairs

More Related