1. 1 LIDDEx web service for pre-signout validation of pathology reports John F. Madden, MD, PhD (Duke University) Shane Brown, PhD (PKS, Inc.) John Hamilton (University of Michigan) Fede Lopez (PKS, Inc.) Rob Manser, MSc (PKS, Inc.) Jeffrey Sica (University of Michigan) Ulysses Balis, MD (University of Michigan)

2. 2 Problem statement Cancer pathology report validation is complex, consensus driven, evolves quickly Pathologists need feedback prior to signout Some LIS systems support structured input, but flatten results to text → no easy validation possible

3. 3 Solution ‣ Report validation as a real-time web service ‣ Accept and parse text reports using rules ‣ Convert to standardized data format ‣ Validate content using centrally-curated rules ‣ Return error/warnings as response

4. 4 PKS/RippleDow n Server Knowledge Builder Clinical Knowledge Base Validator(s) Use Case: PKS-Rippledown ® “Enhanced” Pathology Reports Optional Manual Adjudication Rules Request Update Interpretation Request Interpretation Laboratory Information System Prior Test Results Current Test Results LIS Database Pathologist

5. 5 LIDDEx/PKS Realization Webservice Port Text Parser ValidatorOrchestration

6. 6 Implementation Service Orchestration ParserValidation This Pilot SOAP/WSDL PKS- Rippledown + Ant + XProc PKS- Rippledown Rules Engine PKS/ Relax NG/Schemat ron Other Options RESTVarious PKS- Rippledown Rules Engine PKS- Rippledown

7. 7 Service implementation ‣ Web Services Description Language (WSDL) ‣ Service calls are publicly-defined, included in LIDDEx service standard ‣ Payload is in SOAP Wrapper ‣ Incoming: Pathology report as semistructured text ‣ Outgoing: XML report of warnings and errors, using publicly-defined LIDDEx validation vocabulary

8. 8 Parsing of pathology report: PKS-Rippledown ®

9. 9 Conditional (hierarchical) rules

10. 10 XML Resulting from Parsing

11. 11 Validations All required sections & information items present All responses in CAP-prescribed valueset Any novel items reported with appropriate annotation No forbidden sections or items Structural validations

12. 12 Validations ‣ Calculations pTNM stage assignments correct? ‣ “Sanity checks” Tumor site subsets Specimen site ‣ Complex dependencies If any margin is (+), then no margin is “closest” Cross-element validations

13. 13 Validation implementation ‣ PKS-RippleDown very friendly to mashups with other technologies ‣ In this pilot, open-source Relax NG/Jing and Schematron/Saxon collaborate with PKS engine ‣ XProc/Calabash and Ant scripts collaborate with PKS on orchestration ‣ Cross-element constraints in Schematron especially suitable for translation into PKS- RippleDown (in progress)

14. 14 Validation output XML document using LIDDEx-defined reporting vocabulary English-language and encoded error/warning messages Error/warning location in parsed XML

15. 15 Orchestration ‣ PKS ‣ Server-side scripting ‣ XProc

16. 16 Challenges Context sensitivity in text-to-data processing Existing vendors can easily access validation web- service, but will have to integrate validation data into their GUI Who will be the custodian of the knowledge bases?