Integrating Health Information Resources with Web-Based Clinical Information Systems James J. Cimino, MD Columbia University, New York, New York Guilherme Del Fiol, MD, MS Intermountain Health Care, Salt Lake City, Utah William R. Hersh, MD Oregon Health and Science University, Portland, Oregon J. Marc Overhage, MD, PhD Regenstrief Institute, Indianapolis, Indiana
Panel Outline Motivation History OHSU experience IHC experience Columbia experience Regenstrief experience Institution-independent solutions Open discussion
Motivation for Linking Clinical Systems and Resources Information needs arise with clinical systems The user is using a computer Questions can be anticipated based on context Resource can be anticipated based on question Retrieval can be automated using question, resource and context
History: the Pre-Web Years Yale: Hepatopix and Psychtopix - preconstructed topic-specific queries to Medline Duke: TMR-NLM link - acid-base nomogram to Medline using graphical input MGH: Interactive Query Workstation - UMLS to map to multiple resources - what can the resource answer? SUNY Buffalo: Medical Desktop - cut and paste into UMLS-based Term Linker Pittsburgh: Chartline - used UMLS to match text in reports to MeSH, then searched Medline
History: the Pre-Web Years (continued) Columbia: Medline Button - used ICD9 data, through UMLS, to search Medline Yale: NetMenu and Meta-Front End: UMLS translation to terminologies used by multiple resources Duke: Internet Gopher - hardwired links from Care Maps to Internet-based resources
How the Web Changed Things Local resources became widely available Clinical systems became Web-based HTML and URLs made integration easier
Challenges Integration without distraction Understand information needs Finding appropriate resources Capturing the clinical context for use in retrieval Institution-independence to allow sharing Patient confidentiality Copyright and license restrictions Maintenance
Linking Knowledge to Practice William Hersh, M.D. Department of Medical Informatics & Clinical Epidemiology Oregon Health & Science University
Overview SmartQuery – linking knowledge to practice at OHSU Toward interoperability of knowledge- based resources
SmartQuery (Price & Hersh, 2002) Linkage provided on top of NetAccess, Web- based interface to Siemens Medical Systems (SMS) Lifetime Clinical Record (LCR) repository Context of patient extracted from –Laboratory results – translation from lab system not simple, e.g., Meas ICA, Wh B → calcium, hypocalcemia, hypercalcemia –Diagnosis codes – ICD-9 codes translated to MeSH –Portions of clinical narrative
Resources in SmartQuery Available for free or licensed by OHSU –MEDLINE – via PubMed –Best Evidence (American College of Physicians) – collection of extended, structured abstracts of important journal literature –Harrison’s On-Line (McGraw-Hill) – electronic version of an internal medicine textbook –National Guidelines Clearinghouse (AHRQ) – collection of clinical practice guidelines –CliniWeb – collection of human-reviewed and MeSH-indexed Web pages
SmartQuery selects concepts and offers resources
Results offer linkages to resources
Current status Vendor acquired by another vendor; project on indefinite hold in transition Hard-wired access is not ideal; what we really need is interoperability of clinical information resources…
Rationale: Consider this scenario A primary care clinician of an elderly patient who has hypertension, congestive heart failure, sleep apnea, and obesity –Has charted pertinent information in electronic health record –Now wants recommendations from a guideline with overview of supporting evidence –Later wants to explore recommendations in more detail, including reading systematic review and some original clinical trials it has included –May want basic review of topics seen infrequently in practice
Some impediments for this clinician Cannot link directly from guideline to supporting or background information Wants to access pertinent section of a favorite textbook directly –Does not want to go to each Web site, log on, and use site search engine Would like to navigate across levels of evidence from compendium to systematic review to original clinical trial or other study May want to create personal digital library of preferred content
Impediments for others Publishers –Might desire to allow access to pieces of content but need assurances of revenue and intellectual property protection Content aggregators –Want to “mix and match” content that is “best of breed” but difficult to do across content of different publishers
The current problem: most information is in silos Content Metadata Content Web site Search Engine Other App (eg, EHR) User Most Few
Overcoming the impediments: Interoperability IEEE, 1990: “Ability of two or more systems…to exchange information or use the information that has been exchanged” Used in digital library community to describe seamless access and integration Required to facilitate IR interoperability are –Minimum set of metadata and interapplication interfaces –Cooperation among publishers, vendors, and others to agree upon standards
From silos to interoperability Content Metadata Content Metadata Search Interface Other Application User Site Search Site Search Metadata Repository
How might we achieve this? A starting point is Open Archives Initiative (OAI, OAI promotes the “exposure” of archives’ metadata such that systems can know what content is available and how it can be harvested (Lagoze, 2001) Each record in an OAI collection contains metadata –Protocol has “verbs” for metadata harvesting –Example:
Is this possible? Yes, look to the genomics community, e.g., dtabases of National Center for Biotechnology Information (NCBI) –Literature –Nucleotide and protein sequences –Protein structures –Textbooks and other textual resources –Genomes and map Doing this in clinical medicine will take agreement from publishers and others
The Columbia University Experience: Infobuttons and the Infobutton Manager James J. Cimino, M.D. Department of Biomedical Informatics Columbia University College of Physicians and Surgeons
Medline Button –Translated ICD9-CM to MeSH to search Medline –Hard to build –Did not satisfy users’ needs Columbia Experience Medline Button –Translated ICD9-CM to MeSH to search Medline –Hard to build –Did not satisfy users’ needs Infobuttons –Web-based –Easy to build –Require custom programming –Used preferentially in some settings
Columbia Experience Medline Button –Translated ICD9-CM to MeSH to search Medline –Hard to build –Did not satisfy users’ needs Infobuttons –Web-based –Easy to build –Used preferentially in some settings –Require custom programming Infobutton Manager –Standard set of context variables –Matches context to frequently-asked questions –Each question has corresponding solution –Table-driven Infobuttons –Web-based –Easy to build –Used preferentially in some settings –Require custom programming
34Drug Level InPatient Drugs Any ChildNYPH 33Drug Level InPatient Drugs Any AdultNYPH InfobuttonConceptTaskUserSexAgeInstitute Context Table QuestionURLInfobutton Give me pediatric information for… /pleaflets/… 34 Give me patient information for… /leaflets-english/… 33 Infobutton Table Infobutton Manager
Columbia Experience to Date Questions determined by empiric observation Resources found to answer questions Infobutton Manager links added to WebCIS
Columbia Experience – Next Steps Currently carrying out heuristic evaluation Usability study is next Roll-out to 4000 WebCIS users planned Follow-up observational study to determine: –Usefulness –Impact
Institution-Independent Solutions Contexts are common across institutions Information needs may be common Infobutton Manager is institution-independent “Institute” is a parameter –Questions can be customized by institution –Resources can differ by institution Questions and resources can be shared Terminology is the limiting factor
The Regenstrief Institute Experience J. Marc Overhage, MD, PhD Regenstrief Institute Indianapolis, Indiana
Abstract Many researchers are developing linkages between their clinical information systems and on-line information resources to provide context-sensitive decision support for medical decisions. The discussants of this panel, from the Regenstrief Institute, Oregon Health and Science University, Intermountain Health Car, and Columbia University are all working independently on such solution. They are identifying common problems and common solutions that may lead to collaborative development and resource sharing. The purpose of this panel is to discuss areas of commonality and invite discussion from attendees on their problems and solutions.
Folate
Gopher Example
Gopher
Gopher CPOE Integration 1.Notify one ring of the query 2.Construct and store URL 3.Send URL to browser DLL 4.Browser opens URL PC One Ring Shared Memory Messages DOS/Gopher Browser DLL
Context Institution User Patient demographics –Age –Gender Parameter of interest –Diagnosis (ICD-9) –Laboratory result (LOINC) –Radiology study (text name/local codes) –Question (text)
Logging/Tracking At source Aug 1 10:25:21 flux syslog: |Infomgr|jil7001^CPMC|vc5liu.cpmc.columbia.edu||med|ibutt1600 Aug 1 10:30:19 flux syslog: |Infomgr|jil7001^RMRS|vc5liu.cpmc.columbia.edu||med|ibutt^glucose Aug 1 10:31:33 flux syslog: |Infomgr|jil7001^RMRS|vc5liu.cpmc.columbia.edu||med|ibutt
Other challenges Lots of links increases size of web pages delivered Broken links due to changes in sites Access control –IP – proxy server –Password Fuzzy edges –Clinical sometimes adjunct faculty? –Fellows?
Electronic Health Information Resources at Intermountain Health Care Guilherme Del Fiol, MD, MS Jim Reichert, MD Paul D. Clayton, PhD
Intermountain Health Care (IHC) 21 hospitals, >90 outpatient facilities –4 teaching facilities 500 employed physicians 1.6 million individuals with computer- based clinical data 16,000 clinical system users
Lack of time Limited access to resources Difficulty learning and using multiple resources Variable quality of information Barriers to access health information resources
What we’re doing to lower the barriers Seamless access from the point-of-care –E-resources page: transparent access from clinical information systems –Infobuttons Build clinical questions using patient data and context Take user to the most relevant section within an information resource with a minimum number of mouse clicks In operation since September 2001
How does it work? What are the clinical manifestations of high serum potassium?
i 65 years old female physician lab results Question formulation Resource 3 Resource 1 Resource 2 ? Resource selection Answer retrieval Key factors: Structure Indexing Context
applicationContext = labResults searchString = “Serum potassium” clinicalConcept = terminology = “LOINC” labResult = “High” patientAge = 65 modifier = “Clinical Manifestations” Infobutton request
Arrhythmias: sinus bradycardia, sinus arrest, first degree AV block. Progressive ECG changes: peaked T waves Neuromuscular manifestations: depressed tendon reflexes...
IHC Infobuttons ResourceTerminologies Lab resultsClineguide, MDConsult LOINC, free-text MedicationsMicromedex, Clineguide NDC, free-text Problem listMDConsult, Clineguide, PubMed ICD-9CM, free text
2,247 hits 1,218 hits
Patient Education 13% Lab tests 31% Medications 51% Problems 5%
Challenges Development / Maintenance –Multiple clinical information systems in an institution may need Infobuttons –Adding new resources, changes on APIs –Need for an independent component Selection of resources Structure of resources frequently not driven to Infobuttons
Current status External content resources –Collaborations to improve content structure and Infobutton APIs Internal content resources –Conversion of policies, guidelines, and protocols to XML –Make resources accessible to Infobutton
HL7 Proposal Infobutton Standard API calls
Non-standard APIs POE system Lab results review Outpatient clinical information system i i i Infobutton Manager 1 Infobutton Manager 2 Resource 1 Resource 2 Resource 3 API
Infobutton Manager 1 Infobutton Manager 2 Standard APIs POE system Lab results review Outpatient clinical information system i i i Resource 1 Resource 2 Resource 3 HL7
HL7 proposal Common syntax and terminology for Infobutton managers and information resources API calls Parameters –Main search concept –Application context –Patient context –Modifiers
resource.com/search.cgi searchString= Pneumonia terminology= MeSH conceptOfInterest= D applicationContext= problems ageContext= 45 modifier = treatment What is the treatment for pneumonia ? 45 years old problems
Current Status Draft proposal at HL7 web site Continue discussion –Conference calls to be initiated –Improved draft to be presented at next HL7 meeting – San Diego, Jan 2004 Volunteers are welcome
Open Discussion Understanding how local solutions may be transferable to other settings Identifying challenges to shared solutions Discussion of a standard for evoking resources in context-specific ways Solutions to privacy and copyright issues Establishing mechanisms for collaboration Anything else…