Published on September 9, 2009
What is Clinical Decision Support? The Evidence For and Against CDS Current examples and R&D Projects from Partners The Clinical Decision Support Consortium
“What information consumes is rather obvious: it consumes the attention of its recipients. Hence a wealth of information creates a poverty of attention, and a need to allocate that attention efCiciently among the overabundance of information sources that might consume it.” Changing clinician roles: From Omniscient Oracle… to Knowledge Broker.
acted upon analyzed compiled After B Blum, 1984
Medical literature doubling every 19 years Doubles every 22 months for AIDS care 2 Million facts needed to practice Covell study of LA Internists: 2 unanswered clinical questions for every 3 pts • 40% were described as questions of fact, • 44% were questions of medical opinion, • 16% were questions of non‐medical information. Covell DG, Uman GC, Manning PR. Ann Intern Med. 1985 Oct;103(4):596-9
Generally, with direct observation, or interview immediately after clinical encounters, physicians have approximately one question for every 1‐2 patients Independent estimates: 0.6, and 0.62 Q/pt Holds across PCP and specialty care Holds across urban and rural Gorman, 1995 Gorman and Helfand 1995
An objective measure of the amount of literature generated by medical scientists annually
Original research Negative 18% variable results Dickersin, 1987 Submission 46% 0.5 year Kumar, 1992 17 years Acceptance 14% of to apply Koren, 1989 Negative results 0.6 year research knowledge Kumar, 1992 Publication 17:14 35% to patient care! 0.3 year Poyer, 1982 Balas, 1995 Lack of numbers Bibliographic databases 50% 6. 0 13.0 years Antman, 1992 Poynard, 1985 Reviews, guidelines, textbook Inconsistent 9.3 years indexing Patient Care Balas Yearbook Medical Informatics 2000gtre4, courtesy M Overhage
"...The curse of medical education is the excessive number of schools. The situation can improve only as weaker and superfluous schools are extinguished." “Society reaps at this moment but a small fraction of the advantage which current knowledge has the power to confer.” Abraham Flexner, Medical Education in the United States and Canada. Boston: Merrymount Press, 1910
“Instead of teaching doctors to be intelligent map readers, we have tried to teach every one to be a cartographer.” “We practice healthcare as if we never wrote anything down. It is a spectacle of fragmented intention.” Larry Weed, M.D. (father of “S.O.A.P.” note)
Prone to error Lots of information but no data Limited decision support, or quality measurement Does not integrate with eHealthcare Will not transform healthcare
Medical error, patient safety, and quality issues 98,000 deaths related to medical error 40% of outpatient prescriptions unnecessary Patients receive only 54.9% of recommended care Fractured healthcare delivery system Medicare benegiciaries see 1.3 – 13.8 unique providers annually, on average 6.4 different providers/yr Patient’s multiple records do not interoperate An ‘unwired’ system 90% of the 30B healthcare transactions in the US every year are conducted via mail, fax, or phone
“…driven primarily by local norms that tend towards heavier TEXAS use of discretionary services – El Paso such as diagnostic testing and surgical versus less invasive interventions – for which there are no clear clinical guidelines.” 790 mi., Peter Orszag, OMB Blog 1271 km http://www.whitehouse.gov/omb/ blog/ McAllen http://tr.im/sVLA
“A knowledge‐based system is an AI program whose performance depends more on the explicit presence of a large body of knowledge than on the presence of ingenious computational procedures…” Duda RO, Shortliffe EH. Expert systems research. Science. 1983 Apr 15;220(4594):261-8.
Algorithmic Inference Engine Statistical Pattern Matching Knowledge Base Rule‐based (Heuristic) Meta‐heuristic Fuzzy sets Neural nets Bayesian
A B Blois MS. Clinical judgment and computers. N Engl J Med. 1980 Jul 24;303(4):192‐7.
Formatting Results review, “pocket rounds” reports Interpreting EKG, PFTs, Pap, ABG Consulting QMR, DxPlain, Iliad, Meditel, Abd Pain, MI risk Monitoring Alerts: Critical labs, ABx/Surgery, ADEs Critiquing Vent mgmt, anesthesia mgmt, HTN Rx, Radiology test selection, Blood products ordering Kuperman GJ et al. J Hlth Info Mgmt (13)2, pg 81-96
CDS yields increased adherence to guideline‐based care, enhanced surveillance and monitoring, and decreased medication errors (Chaudhry et al., 2006) CDS, at the time of order entry in a computerized provider order entry system can help eliminate overuse, underuse, and misuse. (Bates et al., 2003; Austin et al., 1994; Linder, Bates and Lee, 2005; Tierney et al., 2003) For expensive radiologic tests and procedures this guidance at the point of ordering can guide physicians toward ordering the most appropriate and cost effective, radiologic tests. (Bates et al., 2003; Khorasani et al., 2003) Showing the cumulative charge display for all tests ordered, reminding about redundant tests ordered, providing counter‐detailing during order entry, and reminding about consequent or corollary orders may also impact resource utilization (Bates and Gawande, 2003; Bates, 2004; McDonald et al., 2004).
Savings potential: $44 billion reduced medication, radiology, laboratory, and ADE‐related expenses Advanced CDS systems Savings potential only with advanced CDS cost give times as much as basic CDS generate 12 times greater ginancial return A potential reduction of more than 2 million adverse drug events (ADEs) annually http://www.citl.org Johnston et al., 2003
Han YY (Pediatrics 116:6, Dec 2005) Analyzed data 13 prior, and 5 months post, implementation of CPOE in critical care Pre CPOE mortality rate 2.8%, Post 6.57% 3.28 Odds ratio after multivariate analysis adjusting for covariates Conclusion Order delay due to lack of pre‐register Up front time cost to enter orders Nurses away from bedside, at computer Altered interactions between ICU team members Delayed pharmacy administration Problems with order timing (subsequent doses)
Information Errors HCI/Workglow Errors Assumed dose Patient selection Med d/c failure Med selection Procedure‐linked med error Unclear log on/off Give now, and prn d/c error Meds after surgery Antibiotic renewal Post surgery suspended meds Diluent option error Time/data loss when CPOE Allergy display down Conglict or duplicate med Med delivery error Timing errors Delayed nursing documentation Rigid system design Koppel R et al. JAMA 293:10, Mar 2005
Record Review Proactive Templates/ Warnings/ Scheduling Alerts Guidelines & Update Reminders Order Sets Feedback Before the Encounter During the Clinical Encounter After the Encounter Patient Prepares History and End of Visit Results Arrive for the Visit Physical Patient Health Relevant Consequent Time-Based Communication Reminders Information Info Display Actions Checks Adapted from Osherorff JA, Pifer EA, Sittig DF, Jenders RA, and Teich JM. Clinical Decision Support Implementers' Workbook. 2004.
Bates et. al. JAMA 1998.
Secure Messaging Patient Lists Schedule Clinical Alerts Knowledge Links Population Task Management Management
Information Access Knowledge Linking
KnowledgeLink in the Workflow
Patient Disease Management
Smart View: Smart Smart Data Display Documentation Assessment, Orders, and Plan Assessment and recommendations generated from rules engine • Lipids • Anti‐platelet therapy • Blood pressure • Glucose control • Microalbuminuria • Immunizations • Smoking • Weight • Eye and foot examinations
Medication Orders Lab Orders Referrals Handouts/Education
Smart Form Used Control 0% 10% 20% 30% 40% 50% 60% 70% 80% Uptodate BP result <0.001 Change in BP therapy if above goal 0.05 Uptodate height and weight 0.004 Change in therapy if A1C above goal 0.006 Uptodate foot exam documented <0.001 Uptodate eye exam documented <0.001 # of deHiciencies addressed <0.001
Targets are 90th percentile for Red, yellow, and green indicators show HEDIS or for Partners providers adherence with targets Zero defect care: • Aspirin • Beta‐blockers • Blood pressure • Lipids
More medication changes in visits after diabetes journal submission: Grant RW et al. Practice-linked Online Personal Health Records for Type 2 Diabetes: A Randomized Controlled Trial. Arch Intern Med. 2008 Sep 8;168(16): 1776-82. .
New appreciation for potential unintended consequences of CDS Knowledge “hardwired” into applications Knowledge‐engineering tools assume authors know what to put into them Proprietary knowledge representation standards: not re‐usable, not easily shared Lack of healthcare leadership or resource investment in processes for knowledge acquisition and management
A Roadmap for Na,onal Ac,on on Clinical Decision Support “to ensure that op-mal, usable and eﬀec-ve clinical decision support is widely available to providers, pa-ents, and individuals where and when they need it to make health care decisions.”! Osheroff JA, Teich JM, Middleton B, Steen EB, Wright A, Detmer DE. J. Am. Med. Inform. Assoc. 2007;14(2):141-145.
To assess, deSine, demonstrate, and evaluate best practices for knowledge management and clinical decision support in healthcare information technology at scale – across multiple ambulatory care settings and EHR technology platforms. www.partners.org/cird/cdsc
How do we improve the translation of knowledge in clinical practice guidelines into actionable CDS in healthcare information technology? How do we optimally represent knowledge and data required to make actionable CDS content in both human and machine readable form? How do we collate, aggregate, and curate knowledge content for CDS in a knowledge portal used by members of the CDS Consortium? How may we use such a tool to support knowledge management and collaborative knowledge engineering for clinical decision support at scale, across multiple healthcare delivery organizations, and multiple domains of medicine? How do we demonstrate broad adoption of evidence‐based CDS at scale in a wide array of HIT products used in disparate ambulatory care delivery settings? Further, how do we deploy clinical decision support services in healthcare information technology in a manner that improves CDS impact? How do we take the learnings garnered through the course of these investigations and broadly disseminate them broadly to key stakeholders?
Decision Tables GEM Arden GEODE-CM ONCOCIN EON(T-Helper) GLIF2 GLIF3 MBTA EON2 Asbru PRODIGY PRODIGY3 Oxford System DILEMMA PROforma of Medicine PRESTIGE 1980 1990 2000 P. L. Elkin, M. Peleg, R. Lacson, E. Bernstam, S. Tu, A. Boxwala, R. Greenes, & E. H. Shortliffe. Toward Standardization of Electronic Guidelines. MD Computing 17(6):39-44, 2000
Shahar Y, et al. JBI 2004
Knowledge management lifecycle Knowledge specigication Knowledge Portal and Repository CDS Knowledge Content and Public Web Services Evaluation Dissemination 1. Knowledge Management Life Cycle 2. Knowledge 3. Knowledge Portal and 4. CDS Public Services Specification Repository and Dashboard 5. Evaluation Process for each CDS Assessment and Research Area 6. Dissemination Process for each Assessment and Research Area
Machine Execu$on Abstract Representa$on Semistructured Recommenda$on Narra$ve Guideline Narra,ve Recommenda,on layer Semi‐Structured Recommenda,on layer Narra$ve text of the recommenda$on from the published guideline. Abstract Representa,on layer Breaks down the text into various slots such as those for applicable Machine Executable layer clinical scenario, the recommended interven$on, and evidence Structures the recommenda$on for use in par$cular kinds of CDS tools • basis for the recommenda$on Knowledge encoded in a format that can be rapidly integrated into a Reminder and alert rules CDS tool on a speciﬁc HIT plaLorm Standard vocabulary codes for data and more precise criteria • Order sets (pseudocode) E.g., rule could be encoded in Arden Syntax A recommenda$on could have several diﬀerent ar$facts created in this layer, one for each kind of CDS tool A recommenda$on could have several diﬀerent ar$facts created in this layer, one for each of the diﬀerent HIT plaLorms
For each knowledge representation layer in CDS stack: Data standard (controlled medical terminology, concept deginitions, allowable values) Logic speciSication (statement of rule logic) Functional requirement (specigication of IT feature requirements for expression of rule, etc.) Report speciSication (description of method for CDS impact measurement and assessment)
Collaboration eRoom for Adult Primary Care
1 Oct 08 9:55pm • How does everyone feel about this? • Should we turn the reminder off for a shorter period of time if “Done Elsewhere” is chosen? 51
Personal Health Rule builder Science Informa,on Network Knowledge respository Clin. Inf. System Pa,ent Rule engine Medical Professional Community (”Crowd”) PeKer K. Risøe HSPH HPM512 2009
“I conclude that though the individual physician is not perfectible, the system of care is, and that the computer will play a major part in the perfection of future care systems.” Clem McDonald, MD NEJM 1976 Thank you! Blackford Middleton, MD firstname.lastname@example.org www.partners.org/cird www.citl.org
Calcification Inhibitors in CKD and Dialysis Patients
Dr. Blackford Middleton keynote address at MIE2009 in Sarajevo, Bosnia-Herzogovina, Sept. 1, 2009.
Share IBM Watson: Clinical Decision Support. ... MIE2009 Keynote Address: Clinical Decision Support. Login or Join. Processing Login successful.
MIE 2005 – Geneva, Switzerland. ... patient empowerment and Decision support and clinical ... site at www.mie2009.net 232 Papers MIE09 ...
The Clinical Decision Support Consortium* ... x Challenges in integrating decision support into the clinical workflow. ... address this challenge, ...
Clinical Decision Support. Articles, experts, jobs, ... Clinical Research (3,065 members) Medicine (2,298 members) Healthcare Consulting (2,207 members)
... related clinical decision support, ... Organised in collaboration with IMIA and the MIE2009 ... Priorities for decision support seem to have been ...
Interoperable Electronic Patient Records for Health Care ... * Extended abstract of a keynote ... advanced EHRs which advanced decision support