From Paper to Practice: Challenges in Implementing Clinical Guidelines

1
From Paper to PracticeChallenges in
Implementing Clinical Guidelines
William Salomon, MD

• Postdoctoral Fellow,
• Medical Informatics
• Decision Systems Group
• Harvard Medical School

• Medical Informatics Specialist
• Central Maine HealthCare

2
Guidelines a definition

• Synonyms du jour
• Protocol
• Practice Parameter
• Best Practice

3
Clinical Guidelines

• Recommendations on screening, diagnosis, workup,
  referral, or management of patients
• Consensus-based
• Evidence-based
• Recurring theme for at least thirty years
• Clinical algorithms in early 1970s
• Triage applications
• Early experiments with computer implementations
• Comparable to rules for clinical trials or
  research protocols

4
Evidence Categories

• Epidemiological
• Evidence stated in terms of probabilities or the
  likelihood in a population, such as estimated
  survival rate
• Biomedical
• Domain-specific knowledge of underlying process
• Clinical
• Domain-specific knowledge from clinical practice
• Observational
• Domain-general knowledge from everyday patient
  observation (e.g. from interacting with patients)
• Authority
• Knowledge derived from trust in someone else

5
Typical Concerns

• Cookbook medicine
• Guidelines, not absolute dogma
• Ambiguous
• Evidence often does not support a single approach
• Range of options often appropriate
• Run counter to accepted practices
• Evidence does not always support accepted
  practices

6
Why Physicians Dont Follow Guidelines

• Cabana, et al. JAMA 19992821458-65
• Lack of awareness
• Lack of familiarity
• External barriers (guideline, patient,
  environmental)
• Inertia of previous practice
• Lack of self-efficacy
• Lack of Agreement

7
Guideline Dissemination

• Conventional publication in journals and
  textbooks
• Mailing of monographs or guideline summaries to
  Clinicians
• Compilations of guidelines for reference
• Online resources
• Locally and regional (www.isci.org)
• National (www.guidelines.gov) and international
• Real-time integration with work flow (EMR)

8
Guideline Dissemination Issues

• Printed guidelines have little impact
• Cabana et al (JAMA, 1999) barriers to
  guideline adherence
• Web guidelines have little impact
• Oregon Health Sciences University study ( Stolte
  et al, AMIA, 1999)

9
Improving guideline compliance

• Distribute guidelines via Electronic Medical
  Record (EMR) systems
• Clinical information from existing data
• Recommendations automatically presented
• Proven in Clinical Practice
• Tang influenza (Northwestern University,Chicago,
  IL, 1998)
• Shojania Vancomycin (Brigham and Womens
  Hospital, 1998)
• Outpatient EMR usage 11 and growing rapidly

10
Guideline Creation andTranslation Process
Local Representation/ Application
Scientific papers
Paper guidelines
Computable Guideline
Encoder
Clinical Use
Manual
?
HUMAN
HUMAN
HUMAN
HUMAN
MACHINE
MACHINE
Authoring Tool
Clinical Use
Guideline Development Organization
MACHINE
General
Site -specific
11
A Common Feature in Clinical Journals
and Textbooks
12
Cough Chest 1998 Aug115(2 Suppl) 133S-181S

• Etc, etc (through a total of 17)
• 65 pages of text, 9 tables, etc.

13
Cough Chest 1998 Aug115(2 Suppl) 133S-181S
14
Analyzing Guidelines for Application to Patients

• Variety of guideline
• Intended users
• Diseases and problems
• Chronic, acute,
• Screening, chronic disease management, clinical
  trials ...
• ...
• Analytical tasks for modeling
• Decision-making
• Workflow specification
• Goal setting
• Data interpretation
• Alerts and reminders

15
Repertoire of Modeling Constructs

• Decision-making
• If-then-else condition or case step
• Rule-in/rule-out criteria for setting preferences
• Decision-analytic models
• Workflow specification
• ...
• Sequential/concurrent/iterative activities
• Organizational models
• ...
• Data interpretation
• Classification hierarchy
• Temporal abstraction
• ...

16
Modeling methodologies for Guidelines and
Decision Rules

• Clinical Guidelines Medical Decision
  Rules
• GLIF Arden Syntax
• PROforma GCARE
• PRODIGY
• PRESTIGE
• EON
• ASBRU

17
Guideline Interchange Format (GLIF 3.0)

• A guideline may be specified on 3 abstraction
  levels
• Conceptual flowchart a flowchart of clinical
  decisions and actions. The information within the
  steps is stored as strings
• Computable the specification may be verified
  for logical consistency and completeness
• Implementable incorporation into institutional
  information system environments

18
Case study stable angina guideline

ltaction stepgt
ltnon-automatic choice stepgt
ltautomatic conditional stepgt

19
Comparing Clinical Guidelines and Medical
Decision Rules
Decision Rules Clinical Guidelines

• Evoking method Event-driven
  Eligibility
  determination
• Temporal character Single point in time
  Prospective guidance
  episodic
• Support Consensus or Consensus
  or evidence-based
  evidence-based
• Typical result of Generation of
  Specification of processing warnings or alerts
  preferred actions
• Characteristics of Precise May be
  imprecise advice delivered recommendations
  and non-algorithmic

20
Arden Syntax What is it?

• A standard method of encoding rules
• Formalized in early 1990s
• Constructed of MLMs
• (Medical Logic Modules)
• Production rule / Procedural hybrid
• Each MLM behaves like a single rule
• Instructions within a MLM execute serially
• MLMs can call each other
• Procedural, not declarative or symbolic
• Penetration
• incorporated in vendor products (Eclipsys, HBOC,
  IBM, Micromedex, SMS, etc)

21
Thyroid Evaluation MLM (ACP)

• Maintenance
• title TSH Evaluation
• filename TSH_eval
• ...
• Library
• purpose Screen for thyroid disease
• ...
• Knowledge
• data
• tsh read damPDQRES2,constraintsC
  58363
• outpatient_visit event '32511','47638
• dob read damGYDAPMP HBASIC
  HBIRTHDT
• age (now - dob) / 1 year
• last_alert read damPDQDEC1
  mlmself
• sex read damGYDAPMP HPBASICHSEX

22
Thyroid MLM (contd)

• evoke outpatient_visit
• logic
• if (age gt 50) and (sex F) and (last_alert
  is null) then
• if (tsh is present) then
• if (time of (last tsh) is before (age - 50)
  years ago) then
• conclude true
• endif
• else conclude true
• endif
• else conclude false
• endif
• action write Patient is a woman older than 50
  years.
• Please obtain a screening TSH.
• end

23
How is it used?Event Monitor

• Events Clinically relevant occurrences
• Result reports Lab, radiology
• Patient oriented outpt. visit, hosp. admission
• Organization oriented inpatient unit transfer
• Output Knowledge Base (MLM) Inference Engine
  (Event Monitor) Patient Data
• Messaging, paging, ordering
• Routed as specified in MLM
• Demands integration with database(s) and
  communication pathways

24
For what is it used?Target Problems

• Usual Relatively simple decision processes
• Less Common More complex guidelines

25
Logician 5.4

• Medscape, Inc., Hillsboro, OR
• Based on Oracle 8 RDBMS
• Migrating to Oracle 8i in ver 5.5
• Uses proprietary scripting language MEL
• (MedicaLogic Expression Language)
• Based on C
• Has various add-in modules
• Scheduling
• Formulary
• Lab Interface (supports HL-7)
• External document interface (FAX, scanning)
• Does not include a billing system

26
Logician Protocols

• An example of a reminder system
• Universally invoked once created
• Can operate in three modes requiring more or less
  user participation
• Background
• requires user to remember to check Protocols
• Modal banner alert
• prompts user to check Protocols if one is active
  from patient header banner
• Display within Form Component
• entirely passive, requires no action of user

27
Logician Opening Screen
28
Invoking Protocols
29
Protocol Displayed
30
Protocol Setup

• This example has no Hx, Med, Dx criteria
• Note windows for BPs

31
Protocol Setup Selection
32
Protocol Syntax

• Representation in pseudocode
• IF
• DEMOGINFO (Sex AND AgeRange) AND
• REGINFO (String1 OR String2 ) AND
• PROBLEM (Modifer_Prob1 OR Modifier_Prob2 ) AND
  MEDICATION (Med1 OR Med2 ) AND
• OBSERVATION ((ObsVal1 AND ObsDate1) OR
• ((ObsVal2 AND ObsDate2) OR
• THEN
• IF EventObs1
• THEN Due on ??/??/??
• ELSE Due NOW
• IF EventObs2
• THEN Due on ??/??/??
• ELSE Due NOW
• ELSE
• Protocol Doesnt display

33
Protocol Shortcomings

• Universally invoked over Enterprise
• Primitive Boolean structure
• Demographics No ORs
• All others No ANDs
• (Ex. Dx of CHF AND Hypertension)
• No access to Medication/Allergy engine
• No branch points
• Branching accomplished by multiple instances of
  protocol simulating each branch
• (Ex. USPS Recommendations for Adult Women
  requires 6 instances of protocol)
• Is this a place that Arden would be useful?

34
Logician Encounter Forms

• Alternatives for Notes
• Free-text Entry
• Macros (Quick-text)
• Encounter Forms (Form Components)
• Encounter Forms allow use of
• Data Entry fields
• Selective display of data
• Extensive data manipulation and transformation
  using MEL (MedicaLogic Expression Language)
• A better method of implementing guidelines

35
Start Update (SexF, Age 45-49)
36
Option to use Protocol settings
37
Previous polyp now noted
38
Colonoscopy options displayed
39
GU note mammogram rec.
40
Colonoscopy now scheduled
41
Nml Cx Exam / Pap Smears noted
42
HRT? Risk of CAD,osteoporosis
43
HRT? Add in Family Hx
44
So what was the reference?
45
HRT discussed declined
46
HRT discussed accepted
47
GU (SexM, Age 40-50)
48
CV risk factors prev. recorded
49
Form-Based Advantages

• Guidelines are incorporated into standard
  workflow and requires no additional effort
• Every patient has access to the same level of
  care regardless of experience of provider or
  specialty
• Guidelines can be updated as a whole, some code
  can be reused (as opposed to Protocols)
• Reference materials can be added as desired
• Opt-out comments can be used, providing immediate
  feedback to improve guideline implementation

50
Form-Based Disadvantages

• Editing in Encounter Form Editor is cumbersome
  and time consuming
• MEL is still limited in clinical data retrieval
  functions
• No ability to selectively omit bad data
• Enforced structure of Observations across an
  Enterprise, regardless of specialty and need for
  detail
• Need for work-arounds to account for differences
  in recording history (Observations vs. Problems)
• Forms may not be sharable across Enterprises due
  to different data structures and workflows

51
Blood Pressure OBS (27/50)
52
Continuing Informatics Issues 1

• No well-defined vocabulary Observation term
  creation is ad-hoc (LOINC, UMLS?)
• Ill-defined semantic relationships between
  Observations (listed hierarchically) making Form
  creation difficult

53
Conflicting/Inconsistent Data Recording 1

• Personal History
• Patient-owned PROBLEM
• Personal hx of PROBLEM (ICD-V10-15.9)
• Hx OBSERVATION
• Family History
• Family Hx of PROBLEM
• Patient-owned PROBLEM (ICD-V16-19.8)
• Family Hx OBSERVATION

54
Family History OBS (34/107)
55
Conflicting/Inconsistent Data Recording 2

• Problems
• Primarily ICD-based (few use SNOMED)
• Irregular use of synonyms makes string searches
  difficult
• Attempts to solve by Custom Problems lists
  limited by interested individuals willing to
  construct them
• ICD-10 ??

56
Conflicting/Inconsistent Data Recording 3

• OBSERVATIONS
• All observations are stored as strings
  (characters) regardless of what you think they
  are
• Enforce use of numbers and date/time by using
  Encounter Form masks
• Each OBSERVATION type has a defined data type
  (lookup in OBSHEAD table)
• What to do with erroneous data

57
Conflicting/Inconsistent Data Recording 4

• Medications
• Uncoded cannot do interaction/allergy checks
• Delays in dissemination information due to
  quarterly updates

58
Conflicting/Inconsistent Data Recording 5

• Document Data
• Now cluttered by large amounts of RTF formatting
  strings

59
Continuing Informatics Issues - 2

• How should medical records that span long periods
  of time be managed so as to remain accessible,
  but not degrade system performance (selective
  archiving)
• Forms and Protocols are Reminders if Alerts
  are needed, where should they be implemented and
  how?

60
On-line, real-time activities

• Perform these when time value is critical
• Clinical Encounters Updates
• Critical alerts generated from imported lab
• High K, INR

61
Off-line, warehoused activities

• Periodic low-criticality reports
• Periodic screening reminders
• Recalls
• Utilization review
• Quality Assurance
• Ad-hoc database intensive queries
• Reporting
• Abstraction to other databases