The Learning Healthcare System: a European perspective Brendan Delaney Wolfson Professor of General Practice, King s College London
Challenges of the EBP Paradigm Clinical Research in crisis Hard to identify subjects Complex, costly CRFs with duplicate data entry RCTs not cost-effective Potential for evidence manipulation Diagnostic error 60% of litigation claims against GPs Failure of Decision Support Systems for Diagnosis 2 April, 2014 2
The Learning Healthcare System SAFER CLINICAL PRACTICE KNOWLEDGE TRANSLATION RESEARCH MORE RESEARCH EVIDENCE 2 April, 2014 3
A natural progression of EBP Past 2 decades ICT has taken centre stage in healthcare BIG DATA Genomics and stratified medicine Decision support Massive increase in clinical practice guidelines Prompts and alerts in health records 2 April, 2014 4
OK... BUT... Clinicians don t code And when they do, data stays in silos Alert fatigue and unknown cognitive effects DATA quality isn t good enough for research Big data Good data Next Generation Sequencing plus Millions of patients health record data = Peta bytes... 2 April, 2014 5
The informatics challenge What we say to dogs Larsen Meaning of data is lost in translation between Clinician and data entry One system and another (or simply left in silos) Clinical use and research use Research knowledge and translation 2 April, 2014 6
Requirements of the Learning Healthcare system Semantic integration tools (including embedded case report forms and prediction models), Workflow (alerts triggers and reminders), provenance Genomics Proteomics, Metaobolomics data TRUST EPRs CDW Primary Care DW Primary Care EPRs Patient portal Environ mental data Data linkage, cohort discovery, provenance, security. 2 April, 2014 7
The integrated clinical laboratory Patient safety and quality improvement RESEARCH COHORTS INTEGRATIVE INFORMATICS SYSTEM Cohorts, biomarkers, Genotypephenotype association studies, RCTs ACUTE AND TERTIARY DATA PRIMARY CARE DATA
Problems with routine data quality Reimbursement bias Why record a BMI in a thin person? Software bias System initiated UK ehrs don t allow negative values and <> Data errors 1% resurrection rate in one UK longitudinal study Myocardial infarction in code NOT in text. Different pick lists for terminologies and the use of non-standard representations e.g. BP!
Barriers Obtaining routine data from primary care ehrs is possible (EU sentinel networks) In most countries linkage via a trusted third party is also happening BUT this data is not collected for research purposes AND ehrs have mostly failed to provide interoperability or indeed to support research translation
TranSMART 2 April, 2014 11
FP7 TRANSFoRm Consortium 2 April, 2014 12
Aims of TRANSFoRm To develop methods, models, services, validated architectures and demonstrations to support: Epidemiological research using GP records, including genotype-phenotype studies and other record linkages Research workflow embedded in the EHR Decision support for diagnosis www.transformproject.eu 2 April, 2014 13
Use case 1: Type 2 Diabetes Research Question: In type 2 diabetic patients, are selected single nucleotide polymorphisms (SNPs) associated with variations in drug response to oral antidiabetic drugs (Sulfonylurea)? Design: Case-control study Data: primary care databases (phenotype data) and genomic databases (genetic risk factors) data federation 14
Use case 2: Gastro-oesophageal reflux disease (GORD) Research Question: What gives the best symptom relief and improvement in QoL: continuous or on demand PPI use? Design: Randomised Controlled Trial (RCT) Data: Collection through ehr & web based questionnaire - ecrf 15
Use case 3: Diagnostic Decision Support Alerting v prompting (assisting v correcting) in chest pain, abdominal pain and shortness of breath Clinical Prediction rule web service (with underlying ontology) Prototype DSS integrated with InPS EHR system Translational Research and Patient Safety in Europe 6
Overall requirements use cases 1+2 Requirement Note Use case Authorisation Explicit or general Cohort and case-control, RCT Consent Informed or explicit Cohort and case-control, RCT Linked phenotype Maintained and refreshed Cohort and case-control Genetic data Browsing and selection Cohort and case-control Recruitment ecrf Research subject portal Embedded real-time in ehr, manages contact and consent A functional tool rather than an CTDMS Patient Related Outcome Measures Cohort and RCT Cohort and RCT Cohort and RCT
Genotype-phenotype evaluations
Clinical Data Integration Model: Ontology-based Upper ontology: Basic Formal Ontology (BFO) Middle (domain) ontologies: OGMS (Ontology of General Medical Science) IAO (Information Artefact Ontology) VSO (Vital Signs Ontology) www.ifomis.org/bfo Biodynamic Ontology: Applying BFO in the Biomedical Domain, D. M. Pisanelli (ed.), Ontologies in Medicine, Amsterdam: IOS Press, 2004, 20 38 http://code.google.com/p/ogms R. H. Scheuermann et al, Toward an Ontological Treatment of Disease and Diagnosis, Proceedings of the 2009 AMIA Summit on Translational Bioinformatics, San Francisco, CA, 2009. p 116-120 http://code.google.com/p/vital-sign-ontology/ Albert Goldfain et al, Vital Sign Ontology, Proceedings of the Workshop on Bio- Ontologies, ISMB, Vienna, June 2011, 71-74 http://code.google.com/p/information-artifact-ontology/ 2 April, 2014 19
Study/Trial CRIM Reference Terminologies and mappings used by Workbench used by CDIM used by Middleware Provenance and security models used by Data node connector used by CDIM-DSM mappings db DS model (DSM) 2 April, 2014 20
Clinical finding -> Phys Exam Clinical finding -> Lab finding Diagnosis Prognosis length -> human height Mass -> dose phenotype -> gender pressure -> diastolic pressure Quality Dependent Thing Entity Continuant Clinical Data Integration Model (ontology) Independent Data item Information Content Material Measurement datum Systolic measurement Lab measurement Pulse rate measurement Document -> Rx Directive -> Act -> Rx item Directive -> Condition -> Rule Label -> Measurement unit -> Unit label Chemical -> Form -> Product Molecular -> DNA -> SNP Object -> Human -> Patient 2 April, 2014 21
GORD RCT evaluations
Agent-based Technology for real time recruitment Autonomous provides configurable flexibility adaptive to user requirements non-intrusive behaviour Asynchronous automation agents self-update their knowledge/registry configure for performance needs
Real-time recruitment and notification EHR systems Agent Pop up Central Control Service CPRD Study information server 2 April, 2014 24
Data Standards Data Elements ISO11179, 13606 IHE Profiles CRPC, RPE, RFDC Core Standards, CDISC, HL7, UMLS Brendan Delaney 2 April, 2014 25
CDISC Operational Data Model Standard for the description of metadata associated with a clinical trial. Allows exchange of datasets. Allows vendor extensions. Does not allow groups within groups on a form in its unextended format. ODM instance would be an xml document with bound terminology and descriptors for text, value, value range, code etc.
Archetypes A computable expression of a domain content model in the form of structured constraint statements based on a reference information model. Often encapsulated together in Templates. Sit between lower level knowledge resources and production systems Independent of interface and system Translational Research and Patient Safety in Europe
Archetypes and Forms 28 2 April, 2014
Diagnostic Learning Healthcare System 60% of litigation against GPs and A+E is for failure to diagnose We don t use Clinical Prediction Rules Stand alone DSS is ineffective How to integrate evidence with EHR? In a standardized way That integrates with clinical workflow That can be easily updated That helps generate new diagnostic evidence 2 April, 2014 29
Family Practice EHR CDIM based Data Connector Decision Support Tools Decision Support Tool Query Interface Diagnostic Evidence Models Clinical Evidence Service Clinical Evidence Ontology Literature Update Interface Data Mining Tools Data Mining Tools Analysis Research Repository Decision Support Components TRANSHIS Project 3
General model of evidence / hasredflaggroup Red Flag Group / hascue 4
Data Mining TransHIS Patient RFEs 1 Eposide of care Encounter 1 Diagnostic cues Clinician Diagnosis 1 Encounter 2 RFEs 2 Diagnosis 2 Diagnostic cues time RFEs n Encounter n Diagnostic cues Diagnosis n
Data Mining: Steps Encounter data TransHIS Encounter data KNIME tool 1 2 Derive association rules Calculate quality measures Web tool (clinical evidences) CSV Web tool (RuleViewer) Filter based on high quality rules Clinical review 3 4 Import XML Evidence transfer to ontology 5
Ontology Representation RFE -Frequency U02 RFE -Dysuria U01 RFE - Abdominal Pain D06 hasrfe hasdifferentialdiagnosis Quantification - Support x Confidence y Lift z Symptom - Fever A03 hassymptom Urinary Tract Infection U71 isquantificationof Demographic - Netherlands Female hasdemographic
Alice GREEN 16/06/1988 (F) (NHS No: 577 459 7164) 71 While Lion Walk, Leeds, Z99 9ZZ 36 New Consultation New consultation for: Alice GREEN 16/06/1988 (F) Reason for encounter / presenting complaint: Edit 1969.00 Abdominal pain Signs, symptoms and examinations: Temperature Y N X (Y) 19F..11 Diarrhea Just once yesterday (N) *Fy0.. Sleep disorders X Doesn t wake up at night, only to Possible Diagnoses: Appendicitis Urinary tract infection Bacterial enteritis Pyelonephritis Crohn s disease Ectopic Pregnancy Irritable bowel syndrome Ovarian cancer <Comments> Diagnosis: Save Done
Alice GREEN 16/06/1988 (F) (NHS No: 577 459 7164) 71 While Lion Walk, Leeds, Z99 9ZZ 37 New Consultation
Fitting it all together. We need to: Separate system components from knowledge Use domain ontology to understand clinical terminology and use it better Have EHR systems that can use ontology for clinical, research and knowledge translation purposes Have secure and fast middleware (plumbing) Be better educated about informatics 2 April, 2014 38
CEN 13606: independence of semantic representation. ehr Interface Clinical terminologies Semantic representation of clinical concepts Database
Conclusion: Collaboration is essential In UK In Europe Internationally Public Private IT industry Health Pharma Biotech Patients 2 April, 2014 40
Acknowledgments King s College London: Natassa Spiridou, Fennie Liang, Simon Miles, Adel Taweel Imperial College: Vasa Curcin University of Rennes: Jean Francois Ethier, Anita Burgin-Parenthoine University of Dundee: Mark McGilchrist University of Birmingham: Theodoros Arvanitis, James Rossiter, Lei Zhao RCSI, Dublin: Derek Corrigan Karolinska Institute: Anna Nixon Andreasson, Lars Agreus University of Antwerp: Paul van Royen, Hilde Bastiens, Johan Wens NIVEL: Robert Verheij CPRD: John Parkinson, Tjeerd van Staa Trinity College Dublin: Siobhan Clarke Brendan 2 April, 2014 Delaney 41