How to make clinical decision
support work
- problems and solutions
Ilkka Kunnamo, MD, PhD
Editor-in-Chief, EBM Guidelines & EBMeDS
Adjunct Professor of General Practice, University of Helsinki
Disclosure: I am a salaried employees of Duodecim Medical Publications Ltd., the company that develops and licenses EBM Guidelines and the EBMeDS decision support service.
22
different ways how computers
caused medication errors
JAMA 2005;293:1197-1203 JAMA March 9, 2005Information technology in health care
– a problem or a solution?
22
different ways how computers
caused medication errors
JAMA 2005;293:1197-1203 JAMA March 9, 2005Do physicians follow alerts?
•
Primary care
physicians overrode
94.2%
of drug interaction and drug allergy alerts
Saul N et al. Physicians' Decisions to Override Computerized Drug Alerts in Primary Care. Arch Intern Med. 2003;163:2625-2631.
Clinical-decision support systems may offer a safety net by reminding
providers of clinical guidelines and catching errors before they cause harm. Evidence suggests that comprehensive EHR systems can improve adherence to clinical guidelines and reduce rates of medication errors. EHR users
overwhelmingly report improvement in the quality of care they provide.
Definition of clinical decision support
(CDS)
•
Providing health care professionals and
citizens
person-specific information and
guidance
based on data in electronic health
records or personal health records
•
Example:
The pa'ent has asthma and is using a non-‐
selec've beta-‐blocker – switch to selec've
Bolland MJ et al. N Z Med J 2007;120:U2804
How to cope with the flood of information?
50% 350%
7566 new systematic reviews were published in 2009
GenBank contents: 80 billion = 80 000 000 000 base pairs in 2008 http://www.ncbi.nlm.nih.gov/Genbank/ 2008 2003 2000 2006
The doubling
rate of
genome data
is 18 months
Exponential growth of medical knowledgeHistory and production of EBM Guidelines
and the EBMeDS decision support service
by Duodecim, Finland
• First version of electronic guidelines 1989
• CD-ROM 1991
• Liaison with the Cochrane Collaboration 2000
• Internet 2000
• Clinical decision support rules 2008
• Translations in English, German, Russian, Estonian,
Hungarian, Slovenian, French, Dutch
Level of evidence and strength of recommendation graded Guidance based on best available evidence
Duodecim is the first non-UK guideline developer that
obtained NHS Evidence
CDS is an opportunity to
summarize essential guidelines
•
Translation of ˜500
reminders
takes
about 45 hours
•
…
while the translation of
full guidelines
Knowledge resources Electronic
Health Record
Sends patient data (XML request message)
Receives decision support (XML message)
Peter Nyberg
The EBMeDS decision support service can be integrated with any electronic health record or personal health record (PHR)
EBMeDS
A clinical decision support service can be
integrated with any EHR or PHR via simple
XML messaging
Essential patient data for CDS
•
Problem list (diagnoses)
•
Medication list
•
Test results and measurements
•
Risk factors (e.g. smoking)
•
Procedures
•
Treatment plan
Essential patient data for CDS
•
Problem list (diagnoses)
•
Medication list
•
Test results and measurements
•
Risk factors (e.g. smoking)
•
Procedures
•
Treatment plan
How to get all data for the patient
from different providers
• Regional EHR (primary care and hospital care)
– Many regions in Finland, Scotland
• National archive for all health records (XML
documents)
– Estonia 2009; Finland 2014 ->
• Health information exchange (HIE)
– USA: CCR (Continuity of Care Record) XML
document
• Personal Health Records (PHRs)
– USA: Microsoft Health Vault
• National locator and viewing services
BMJ November 19th 2011
Standardization and
certification
•
National coding standards
•
Certification of electronic health records
–
USA: Meaningful use
•
Incentives for physicians up to
44 000 $
(Medicare) or
63 000 $
(Medicaid) in 6
years for using certified EHRs
–
Belgium: Support for users of certified
Meaningful use – USA
Requirements stage 1 – examples
•
Problem list
•
Medication list
•
Electronic prescribing
•
At least one decision support rule
•
Drug-drug interaction and allergy check
•
Reporting of clinical quality measures
•
Providing the patient an electronic summary
•
Capability to exchance information with other
What is needed for building a
generic CDS service?
•
Best available evidence
•
Shared ideas (free on the web)
•
Ability to use rules developed by others as
templates for local rules
•
Translating output into different languages
•
Mapping between EHR coding systems
Collaboration is needed in CDS
content development – also in
Europe
Multilingual, web-based
content development tool
Reminders
English
Finnish
Codes and aliases
Discussion blog for each rule
Demo website for clinical decision support
The EBMeDS decision support
service can be integrated with any
electronic health record or
How to make clinical support
acceptable and actionable
What clinicians want?
•
Summary of essential patient data
•
Reminders that really improve patient safety
•
Automation of routine work – saving time
Choices offered by the CDS system
Only 1 out of 9
existing US systems was able to offer all the choices
How to make reminders tolerable
•
Increase
thresholds
above those suggested
by guidelines
•
Use only data that has been
reliably coded
in
the EHRs
•
Make reminders of
strong
recommendations
that everyone would follow
= improve specificity (at the cost of sensitivity)
to decrease the number of reminders and
avoid alert fatigue
Tools for population
health, professional
development and
In a virtual health check all CDS rules are executed in a
population of patients, and resulting reminders are listed.
Number of reminders per person
3345 people out of 16143 (21%) got at least one reminder
607 people (3.8 %) got more than three reminders
Age distribution of people with
reminders
41 % of all reminders were triggered for people aged 70+
Two purposes for the VHC
•
Clinical
: find people who need interventions
and contact them
– Persons identified
•
Analytic
: create statistics about the target
group for clinical interventions or the quality of
care
Example of decision support and quality
reporting for a population of 16 000
• Cardioselective beta-blockers for patients with asthma:
No reminder (selective beta-blocker 32 in use)
Reminder: Asthma ‒ switch to selective 4
beta-blocker?
Quideline compliance = 0.89
(n = 36)
89% of patients with asthma and beta-blocker used the right type of beta-blocker.
Examples of reminders triggered in a
Virtual
Health Check
for a population of 16 000
from a set of
100 rules
• Antihypertensive drug not used in moderately high BP 396
and high CV risk
• ACEI/AT blocker/beta blocker not in use in heart failure 143
• LDL > 2.5 mmol/l in type 2 diabetes 69
• Metformin not in use in type 2 diabetes 61
NNTR = number needed to remind
NNTR = 5.9
= 1/(2.0/(2.0 + 9.9))
Tsurikova R. Clinical Decision Support Consortium 2011, Partners Health, USA
Performance
Pa'ent not seen during the month. No Performance
Reminder followed by performance
The lost population who could
benefit from an intervention
Performance
Pa'ent not seen during the month. No Performance
Reminder followed by performance
Reminder followed by no performance
Congratulations!
This month you and your team have
saved 2 lives, prevented 9
hospitalizations, and improved the
quality of life of 31 patients
What about a statistical
reward?
Keep it simple
Be flexible
Ten commandments for effective CDS
1. Speed is everything2. Anticipate needs and deliver in real time
3. Fit to user s workflow
4. Little things (usability) can make big difference
5. Do no stop clinicians actions
6. Change direction rather than stop
7. Simple interventions work best
8. Ask additional information only if really needed
9. Monitor impact
10. Manage and maintain knowledge base
Effective clinical decision support
•
Useful
•
Actionable
•
Targeted at the right person
•
At the right time
•
With the right presentation
NIMAC 18.4.2010
Target group: citizens
Your op-mal predic-on:
50 out of 100 will reach 85 years
Your current predic-on:
6 out of 100 will reach 85 years
The curve shows how many out of one hundred of your kind will reach the below indicated age
Diagnosis Not documented (%) Documented Hypothyroidism 331 80 % 85 Diabetes 193 21 % 739 Bipolar disorder 6 17 % 29 CHF or atrial fibrillation 26 5 % 488 Criteria
Hypothyroidism: thyroxine on medication list
Diabetes: insulin or oral antidiabetic on the medication list, HbA1c >= 6.5% Bipolar disorder: lithium on the medication list
CHF or atrial fibrillation: digoxin on the medication list
Diagnoses not documented in the EHR
The maximum remaining clinical care gap
Performance
Pa'ent not seen during the month. No Performance
Reminder followed by performance
Reminder followed by no performance
Tsurikova R. Clinical Decision Support Consortium 2011, Partners Health, USA
?
Performance
Pa'ent not seen during the month. No Performance
Reminder followed by performance
Reminder followed by no performance
If the NNTR would be the same for those
who are lost, an additional 8.3 % of the
population would get the intervention
Limitations and problems of the VHC
•
Missing or incorrect data
– All data are not available in one place
– Diagnoses not documented
– Wrong medications on the medication list
•
All treatments are not recorded
– Lifestyle interventions, patient education,
psychosocial interventions
•
Missing individual targets
– Might be different from recommendations in
Meaningful use – USA
Quality measures 2012 – examples of a
set of 36 measures
•
Use of appropriate medication for asthma
•
Appropriate testing for children with
pharyngitis
•
Low back pain: appropriate use of imaging
studies
•
Chlamydia screening for women
•
Pneumonia vaccination for the elderly
Decision Support Engine
Local EBMeDS Service Electronic Health Record
Work- station Patient Database
Global, National & Local Scripts
Link Table Patient Data Feedback Data Drug Tables Software Knowledge Databases & Electronic Forms
Editing & Customizing
Tools Database EBMeDS
EBMeDS Master File Package
Central EBMeDS Service
Lo
g
F
il
Decision Support
Core Engine
EBMeDS Local Service EHR Work station Local Patient Database
Global, National & Local Scripts Evidence Links
Function Library
Anonymous Patient Data
Response Data Drug Information • Best Practice • Contraindications Con versi on fi lt ers Gl u e c omp one nt Client Component Web Server Application Web Resources • EBM Guidelines • Evidence Summaries • Forms & Calculators
• EBMeDS Home Page Script Editing Tools Database EBMeDS
EBMeDS Master File
Package
EBMeDS Central Service
Administrator Tools
Testing tools Testing
tools
Table Editing Tools
JavaScript Interpreter Update Service Lo g F il es EBMeDS Engine Compiling Tools • Cochrane Library • Essential Evidence • Interactions
Patient Data Archive
Central Patient Database
Meaningful use – USA
Quality measures 2011
•
Hypertension – BP measurement
•
Tobacco use assessment and cessation
intervention
•
Adult weight screening and follow-up
•
Weight assessment and counseling for
children
•
Influenza immunization for adults > 50 years
Factors facilitating population-based
guideline implementation via CDS
•
Structuring and standardization of EHR data
•
Regional or national patient data repositories
– National eArchives in Finland and in Estonia
– Quality register in Denmark
•
Chronic care model
– Care plans, team work, empowered citizens
KNOWLEDGE Guidelines
Graded evidence Databases: drugs, laboratory, genome Images and videos for training of skills Cost-effectiveness Ethical summaries Patient information Patient data Genome map Database of all previous patients Probably beneficial therapy Simulation Individualized prediction of the effects of treatment Patient s values and choices Selection of treatment
Selection of medical interventions in 2020
Decision support Doctor s interpre- tation and experience Resource limits
Target group: nurses and other
health care professionals
•
More contacts than physicians
•
More time per contact
•
Responsible for care for chronic conditions
Ø
CDS must include guidance for all
Patient data analyzed
Real-time reminders triggered
Javascript is used as executable scripting language – works in all operating systems
Interactive algorithms show what is the best treatment option for this patient with current data
Interactive algorithms
•
A reminder (triggered by patient data)
contains link to an algorithm
•
The algorithm is populated by patient data
•
Example: radiotherapy for prostate cancer
Low risk
Intermediate risk Test site
Benefit exceed costs 2 years after implementation
According to HIMSS Analytics, it takes averagely 9 years to get return of
Factors and elements predicting the
success of a clinical DS system
• Using a computer to generate decision support
• Automatic provision of reminders as part of clinician
workflow
• Providing clear recommendations as opposed to
providing only assessment about the situation or patient s condition
• Providing decision support at the time and location of
decision making
• Of systems possessing all 4 features, 30 out of 32
(94%) improved the quality of patient care
Kawamoto K ym. Improving clinical practice using clinical decision support systems: a systematic review of trials to identify features critical to success. BMJ 2005;330:765-768
Types of health information
•
Medical knowledge
•
Patient data
•
Directory information (staff, services,
locations)
Decision support
Wyatt JC, Sullivan F. What is health information?
BMJ 2005;331:566-568
Standard codeset for Finnish electronic
health records
• Problems/diagnoses ICD-10, ICPC-2
• Medication ATC
• Test (orders and) results National codes
• Measurements (e.g. BP) LOINC (subset)
• Risk factors (e.g. smoking) National codes
• Procedures NCSP (Nordic)
– Procedures in primary care SPAT (NCSP extension)
• Treatment plan as above + national
The Finnish Medical Society Duodecim
Duodecim Medical Publications Ltd.
• Scientific society founded in1881
> 90% of the Finnish physicians as members
• Continuous Medical Education
• Clinical Practice Guidelines
• Medical terminology in Finnish language
• Awards and grants for young scientists
• 100% owned by the Finnish Medical Society
• Electronic publishing since 1989
• Publisher of the national health portal for both
