Guideline implementationTypes of CDSS
A.Hasman
Do physicians need support?
• In 2.3% of the 1.3 million patients (30.000 patients) preventable errors were made during their stay in a hospital in the Netherlands.
• About 10.000 patients suffered a permanent injury. This could have been prevented in 6.000 patients.
• For 4.1% of the 42.000 patients who died in the hospital death could have been prevented.
Conclusion
• Doctors are not infallible
• Support physicians and nurses both in repeating and difficult tasks
How to support physicians and nurses?
• Easy access to the scientific literature
• Guidelines
• Computer-based diagnostic systems
Medical decision support
• Already available for years– ECG/EEG analysis (signal analysis and parameter
interpretation)– Diagnostic systems (cardiology (congenital heart
disease diagnosis), radiology (bone tumor classification))
– Radiotherapy planning– Medication selection, dosing– Clinical algorithms (flowcharts on paper, for nurses
and ancillary personnel)– Guidelines
Methodology used by CDSS
– Decision trees– Statistical approaches
• Bayes’ rule• Discriminant analysis• Logistic regression
– Inference techniques• Rules• Logic• Semantic networks
– Etc.
Decision trees
BP lower than 140/90
Send patient home First visit?
YesNo
Yes No
x
x
xx
x
x
x
O
O
O
O
O
O
O
x
xO
Var 1
Var 2
Bayes’ rule
P(Dj |Si ) = P(Dj) * P(Si|Dj)/P(Si)
P(Dj) prior probability of disease j (prevalence)
P(Si) probability of symptom i in population
P(Si|Dj) conditional probability (sensitivity or specificity))
P(Dj|Si) posterior probability (predictive value)
Types of decision support
• Passive– Physician actively searches in the knowledge
base for relevant information. Information indexed
• Active– System pro-actively provides physician with
relevant information– System re-actively provides physician with
relevant information
Guidelines
• Systematically developed statements to assist practitioner and patient decisions about appropriate healthcare for specific clinical circumstances
• They provide information for various types of patients having some common problem
• Provide a common standard of care both within a healthcare organization and among different organizations
• Based on consensus or evidence-based
Use of guidelines
• May lead to a reduction of errors, practice variability and patient care costs, while improving patient care
• Narrative guidelines usually population-based, not patient specific
• Healthcare organizations pay more attention to guideline development than to guideline implementation, evidently hoping
• That clinicians will simply familiarize themselves with published guidelines and then apply them appropriately during the care of patients
Background: paper-based guidelines
Problems with guidelines
• Guidelines often contain ambiguities, vague sentences and ‘open ends’
• Leads to different interpretations
• This limits the use of guidelines
Guideline Dissemination
• Assumption: Practitioners will read the guidelines
• Assumption: Practitioners will internalize and there after follow guidelines
• Reality: Physicians do not use the guidelines or do not use them correctly
Why was decision support not accepted?
• Decision support systems were only applied in the institution where they were developed, if at all
• Computer systems were stand-alone systems: no integration, so double data entry
• Computer systems were slow and expensive• Initially physicians did not accept guidelines or clinical
algorithms (Cookbook medicine, patients differ, useful for ancillary personnel)
• Because of current emphasis on quality of care (evidence based medicine) guidelines are becoming relevant
Myth-1
• Diagnosis is the dominant decision-making issue in medicine – Typical questions are not “What does this
patient have?” but, rather, “What should I do for this patient?
Ted Shortliffe
Myth-2
• Clinicians will use knowledge-based systems if the programs can be shown to function at the level of experts– What do we know about “expertise” and
the associated cognitive factors?
The nature of clinical expertise• Tremendous variation in practice, even
among “experts”• Need to understand better how experts meld
personal heuristics and experience with data, and knowledge from the literature, in order to arrive at decisions (medical cognition)– Can we better teach such skills?– How could improved understanding affect the way
decision-support systems offer their advice or information?
– How will such insights affect our under-standing of clinicians as computer users?
Myth-3
• Clinicians will use stand-alone decision-support tools. – The death of the “Greek Oracle” model
→Integrated decision support in the context of routine workflow
Systematic review of Garg et al
• Systems that warn physicians have more effect on the physicians (success in 44/60 studies) than systems that have to be inititiated by physicians (17/36 studies)
• In the case of diagnostic systems 4 out 10 trials indicated that the use of a DSS leads to better results (an improvement for at least 50% of the the measured outcomes)
Reminder systems effective?
• For 16/21 trials concerning reminder systems for prevention using a DSS led to a better performance of the physicians (screening, test requests, drug prescription, etc.)
• Studies did not show a significant improvement in patient outcome
Computer interpretable guidelines
Computer-interpretable guidelines
• Guideline implementations best affect clinician behaviour if they deliver patient specific advice during patient encounters
• Computer-interpretable guidelines could provide such advice efficiently
• Computerized guideline systems are crucial elements in long-term strategies for promoting the use of guidelines (IOM)
Possibilities of ICT
• Computersystems can not only show guideline texts but can also reason with the information from the guideline. To do that information about an individual patient is necessary
• The combination of a formalized guideline and an EPR can lead to advice (pro-actively or reactively) concerning an individual patient
Medical Protocols
Methods and techniques20
System description
Guideline Base
EPRExecution
engine
Knowledge Base
Guideline / Knowledge Base editor
Physician
CIG ingredients
• Guideline model• Guideline expression language for expressing
decision criteria and eligibility criteria• Mapping of terminology in guideline to the
terminology used in EPR• Scheduling constraint specification language for
scheduling multiple steps• Guideline execution engine
Guideline modeling and representation
• System editor should provide– A domain ontology– A (visual) language for expressing the steps
in a guideline
Example: simple guideline
• Primitive: If … then
• Domain ontology: Digoxin, Potassium, …
If Digoxin and Potassium>3 mmol/l then “warning”
Phases in development process
• Select guideline to be formalized
• Formalize guideline
• Enter guideline into guideline system
• Guideline verification and testing
• Guideline execution
Acquiring guidelines
Acquiring guidelines
Acquiring guidelines
Acquiring guidelines
Acquiring guidelines5
Acquiring guidelines6
Complex guideline
• Primitives: Branch step, Synchronization step, Decision step, ….
• Domain ontology: Digoxin, ….
Guideline representation in GastonMode structure
Guideline representation in GastonMode structure
Guideline representation in GastonMode contents
IS ~ GP
Evaluation module
Re-active decision support
KB
Request module
ICPC-module Reminder
No reminder
Reminder 1:A sinus X-ray is not adviced for children
younger than 10 years of age
Executing guidelines