mmi 402 fall 2010 - medical informatics...
TRANSCRIPT
MMI 402 FALL 2010
Problem Solving in Clinical Medicine: From Data to Diagnosis Paul Cutler, MD 3rd Edition, 1998 ISBN: 0-
683-30167-5
Contents Article: Agenda Setting ............................................................................................................................. 3
Article: Greeting Patients .......................................................................................................................... 3
Article: Computational Technology for effective health care:.................................................................. 3
Chapter 1: Dissection of Symptoms .......................................................................................................... 8
Chapter 2: Data collection, data processing, problem lists ...................................................................... 9
Chapter 3: Problem solving Methods ..................................................................................................... 12
Chapter 4: Clues: the Building Block ....................................................................................................... 16
Article: Where is the beef ....................................................................................................................... 18
Article: Clinical skills textbooks fail evidence based examination .......................................................... 19
Article: Stimulating the adoption of Health IT ........................................................................................ 19
Article: What is Wegener’s Granutomatosis .......................................................................................... 20
Chapter 9: Impact of new technology .................................................................................................... 21
Chapter 10: Into the Solver’s Mind ......................................................................................................... 23
Chapter 11: From Patient to Paper ......................................................................................................... 23
Article: A look at Google Scholar ............................................................................................................ 24
Article: Google Scholar: A Source for clinicians ...................................................................................... 25
Article: The Medical Literature ............................................................................................................... 26
Article: BMC Medical Informatics and Decision Making: Software askMEDLINE ................................... 28
Chapter 7: Digits, Decimals, Doctors ....................................................................................................... 28
Chapter 8: How to use test ..................................................................................................................... 31
Article: Prediction of Pulmonary Embolism in the Emergency Department .......................................... 34
Article: Using Clinical Prediction Rule ..................................................................................................... 35
Sensitivity and Specificity ........................................................................................................................ 35
Algorithm for the prediction of disease .................................................................................................. 35
Chapter 5: Data Resolution Skills ............................................................................................................ 36
Chapter 6: More Clinical Tools ................................................................................................................ 37
Article: Likelihood ratios: Getting Diagnostic testing into perspective .................................................. 39
Article: Failure Modes and Effects Analysis (FMEA) ............................................................................... 41
www.rapid-diagnostics.org/accuracy.htm.............................................................................................. 43
Med Rec .................................................................................................................................................. 43
Article: The Epidemiology of Prescribing Errors: The potential impact of computerized Prescriber
Order Entry (CPOE) ................................................................................................................................. 44
Article: Clinical Decision Support on Electronic Prescribing: Recommendation and an Action Plan ..... 45
Article: Electronic Prescribing: Toward Maximum Value and Rapid Adoption ...................................... 46
Article: Health Information Technology: Fallacies and sober realities ................................................... 47
Absolute Risk Reduction ......................................................................................................................... 50
Article: Collective Statistically Illiteracy .................................................................................................. 50
Article: Fumbled handoffs: One dropped ball after another .................................................................. 50
Article: Assessing the value of a diagnostic test ..................................................................................... 51
Article: The Wrong Patient ..................................................................................................................... 52
Article: Triage of patients with Acute chest paint and possible cardiac Ischemia: The elusive search for
diagnostic perfection .............................................................................................................................. 53
Article: Yield of Diagnostic tests in Evaluating Syncopal Episodes in Older Patients ............................. 54
Article: Computerization can create safety hazards: A bar coding near miss ........................................ 54
Article: The Cognitive Psychology of Missed Diagnoses ......................................................................... 56
Article: Randomized comparison of Guaiac-Based vs Immunochemical FOBT ...................................... 56
Article: Prescription Errors and outcomes related to inconsistent information transmitted through
computerized order entry ....................................................................................................................... 57
Article: Health Care Technology: A Cloud Around the Silver Lining? ...................................................... 57
Index........................................................................................................................................................ 60
Article: Agenda Setting
Complete Interview:
1) Single Chief Symptom
2) Further elaboration of the history of that symptom
3) Family History (FH)
4) Personal Medical History (PMH)
5) Personal Social History (PSH)
6) Drug and Allergy History (DAH)
7) Systems Review
Key to establishing the Agenda:
1) What are the patient’s main concern for today
2) What are the clinician’s concerns about this patient
3) What are the patient’s specific requests
4) How much of the patient’s or doctor’s concerns need to be addressed today? And which ones
can be deferred to a subsequent contact
5) What disagreements about the priorities exist? How will they be negotiated?
- MD use an “empathic bridge” to bring the conversation back
- Invest the time to develop a mutual agenda at the beginning of the visit helps agree on mutually
important concerns
Article: Greeting Patients - MD should be encouraged to shake hands with patients but remain sensitive to non verbal cues
- MD initially use patient’s first and last name and introduce themselves using their own first and
last names
Article: Computational Technology for effective health care:
2 Basic challenges:
1) Use the best technologies today to build and deploy systems in the short term
2) Identify the gaps between the best of today’s technology and what is ultimately needed to
improve health care
- Success will require greater emphasis on providing cognitive support for health care provider
- Computer based tools and systems that offer MD and patients assistance for thinking about
solving problems related to the specific instances of healthcare
Current HC 3 areas:
1) Tasks and workflow of HC
a. Complex care is increasingly provided to patients in a time and resource pressured
environment because of the need to contain costs
2) The institution and economics of HC
a. Large # of HC payers and coverage plans, each with own rules
3) The nature of HC IT as it is currently implemented
a. Often implemented in systems in a monolithic fashion that makes even small changes
hard to introduce
b. Often designed to mimic existing paper based forms and provide little support for the
cognitive tasks of MD
c. Applications do not take advantage of human computer interaction principles
HC Persistent Problem:
- Intellectual complexity of HC as a whole
- Environment not structured to help avoid mistakes or improve decision making
- Administrative fragmentation
- Complex unclear authority
IOM: HC quality
- The degree to which health services for individuals and populations increase the likelihood of
desired health outcomes and are consistent with current professional knowledge
- HC Should be:
1) Safe
2) Effective
3) Patient oriented
4) Timely
5) Efficient
6) Equitable
Achieving the vision entails:
1) How to pay for HC
2) Disease prevention
3) Effective use of Information
7 information intensive aspects of the IOM vision for 21th century HC:
1) Comprehensive data on patient’s conditions, treatments, and outcomes
2) Cognitive support for HC professionals and patients to help integrate patient-specific data
3) Cognitive support HC professionals to help integrate evidence based practices guidelines
4) Instruments that allow MD to manage a portfolio of patients and to highlight problems as they
arise both for an individual and within population
5) Rapid integration of new instrumentation biological knowledge, into a “learning” HC system that
encourages early adoption of promising methods but also analyzes all patient experience as
data
6) Accommodation of growing heterogeneity of locales for provision of care
7) Empowerment of patients and their families in effective management of HC decision and
implementation
To cross the HC IT chasm:
1) Re-balancing the portfolio of investments in HC IT to greater emphasis on providing cognitive
support for HC providers and patients
2) Observing proven principles for success and designing and implementing iT
3) Accelerating research to HC in the computer and social sciences and in health / biomedical
informatics
Principles for Evolutionary Change:
1) Focus on improvements in care (technology is secondary)
2) Seek incremental gain from incremental effort
3) Record available data so that today’s biomedical knowledge can be used to interpret the data to
drive care, process improvement and research
4) Design for human and organizational factors so that social and institutional processes will not
pose barriers
5) Support cognitive functions of all caregiviers
Principles for radical change:
1) Architect information and workflow system to accommodate disruptive change
2) Archive data for subsequent re-interpretation in anticipation of future advances in biomedical
knowledge
3) Seek and develop technologies that identify and eliminate ineffective work process
4) Seek and develop technologies that clarify the context of data
Research challenges:
1) The extent to which new fundamental, general purpose research is needed
2) The extent to which new research specific to HC and biomedicine is needed
a. Some for improving CS problems
b. Some for health care
3) Grand challenge: Patient centered cognitive support
Much of HC is transactional
- Understanding of the patient and a set of goals and plans for patient
- Clinical spend a lot of time researching through raw data about patients and try to integrate the
data with medical knowledge
Vision of patient centered support cognitive support:
- The MD interacts with models of the patient that place the raw data into context and synthesize
them with medical knowledge
- Enable patient specific parameterization and multi-component alerts
- Medical logic links the raw data to abstract model
- Computer provide decision support
- The content of MD interactions would be captured in self documenting environment
Other challenges:
1) Modeling various subsystems within a real patient and show how they interact
2) Automation: Automated systems do not work harmoniously with each other
3) Data sharing and collaboration
a. Data integration entails a major and costly effort
4) Data management at scale
5) Automated full capture of MD-patients interactions
*** MD drawn to IT only if it can be shown to enable them to do their jobs more effectively ***
6 recommendations for the government:
1) Incentivize MD performance gains rather than acquisition of IT per se
2) Encourage initiatives to empower iterative process improvements and small scale optimization
3) Encourage development of standards and measures of HC IT performance related to cognitive
support for the MD and patients
4) Encourage interdisciplinary research in 3 critical areas
a. Organizational systems level research into the design of HC systems. Processes and
workflow
b. Computable knowledge of structures and models for medicine needed to make sense of
available patient data
c. Human computer interaction
5) Encourage efforts by the HC organization to aggregate data about HC people, processes, from all
sources
6) Support additional education and training efforts at the interactive of HC CS and health /
biomedical informatics
CS community should
1) Engage as coequal intellectual partners with HC experts in health / biomedical informatics
2) Develop institutional mechanisms for rewarding work at the HC/CS interface
3) Support educational and retraining efforts for CS researchers in HC
HC organizations should
1) Organize incentives to encourage clinical performance gains
2) Balance the institution’s IT portfolio among automation, connectivity, decision support, and
data mining
3) Develop the necessary data infrastructure for HC improvement by aggregating data from all
sources
4) Insist that vendors supply IT that permits the separation of data from applications and facilitate
data transfers to and from other non vendor applications
5) Seek IT solutions that yield incremental gains from incremental efforts
Better IT systems:
1) Coordinate patient’s care with other providers
2) Share needed information
3) Monitor compliance with prevention and disease management guidelines
4) Measure and improve performance
Key findings:
1) Properly implemented and widely adopted health IT would save money and significantly
improve HC quality
2) Annual savings from efficiency alone could be > $77B
3) Health and safety benefits could double the saving while reducing illness and prolong life
4) Implementation would cost around $8B / year assuming adoption by 90% of hospitals and
doctors offices over 15 years
5) Obstacles include market disincentives: Those who pay for health IT do not receive the related
savings
6) The government should act now to overcome obstacles and realize benefits
- HIT includes a variety of integrated data sources
- HIT to provide timely access to patient information and can communicate health information to
other providers
- Creating and maintaining such systems is complex
- Benefits can include
1) Dramatic efficiency savings: When the same work is performed from with fewer resources
2) Increased Safety: From alerts and reminders generated by computerized physicians Order
entry (CPOE) systems for mediations
3) Health Benefits:
a. Disease prevention
b. Chronic disease management
c. Scan patients records for risks factors and recommend preventive services
d. Cost $9 M / year
Obstacles:
1) Few providers have access to HIT
2) Connectivity – the ability to share information
3) The disconnect between who pays for HIT and who profits from HIT
a. Providers pay in higher cost to implement HIT and lower revenues after implementation
Government should act now:
1) Continue current efforts
a. Support for development of uniform standards
b. HIT certification processes
c. Common performance metrics
d. Expand liability protection for hospitals using HIT
e. Subsidize HIT system for doctors
2) Accelerate market forces
a. Set up a pay for use program for providers using HIT systems
b. Create national performance reporting infrastructure to receive and report comparative
performance data
c. Fund research on pay for performance incentive
d. Educate consumers about the value of HIT in improving their ability to manage their
own health
e. Pay for use program followed by broad based pay for performance programs
3) Subsidize change
a. Institute grants to encourage the development of organizational tools, and best
practices to help IT success
b. Convincing individual MD and their patients of the value and safety of networking
confidential data will be critical
Chapter 1: Dissection of Symptoms 1) Questions asked by ER
a. How it began
b. How it progressed
c. The Accompanying symptoms
d. The patient’s identifying data
e. The Past medical history
2) ER decided what was the most likely diagnosis out of a group of contending hypotheses
3) Gather additional information to uphold or reject
4) Rule out contenders
5) Re rank the order of likelihoods
- Ask carefully selected questions
- Performs a subset of PE the focused on the area of concern
- Once reasonably certain, ordered tests to sustain primary impression and rule out others
ER needs to elicit meaningful info, with understanding of the basic Pathophysiologic processes that
might be operating in patient’s chief complaint
In seeking the cause, need to determine:
1) How and when it began
2) What has been its progression since its onset
3) Whether it is continuous or intermittent
4) Whether it is related to a body position
5) What makes it better and worse
6) What are the associated symptoms
7) Whether it occurs at rest, exertion or both
As a problem solver, you must
1) Know and understand the variable circumstances in which the illness occurs in each instance
2) Familiar with the associated symptoms
- Use associated signs from PE that focuses only on areas that may be concerned
- A much more focuses initial examination may be done if, on the basis of history, one diagnostic
impression seems almost certain
- Chief complaint + associated symptoms + physical signs
- For more certainty, may request high yield tests to be done
Chapter 2: Data collection, data processing, problem lists Principal initial steps in the problem solving processes:
1) Collection of patient data
2) Evaluation of data
3) Formation of a differential diagnosis or problem list
Every bit of information collected is interpreted as being relevant or non relevant to the chief complaint
Collection of patient data:
1) The History
2) PE
3) Tests that include blood chemistry, complete blood count, urinalysis,
- In the great majority of the case, the history is the most important and most revealing portion of
the database.
- Relative importance to diagnosis:
o History 70%
o PE 20%
o Lab Test 10%
- History taking has been the most neglected skill
- Often ask a few questions, do a cursory PE, and order a lot of test
- Learn how to approach a patient with sympathy, politeness, humility, and a smile
- A good history taker must be a good listener
- At the beginning asks open ended questions for patients to talk
- Specific closed ended questions come later
- Body language is as important as what doctor say
- Difficulties may arise in getting the truth
Students starting to take history steps:
1) Identification data
2) Chief Complaint
3) History of present illness
4) Family history
5) Systems review
But MD uses a flexible approach and a conversational style that change order, bypass some
questions, and asking other questions
- The problem solver forms one or more early provisional diagnosis and asks questions with high
sensitivity and questions with high specificity
Steps:
1) 2 minutes in:
a. Initial concept is based on patient’s age, gender, chief complaint, and one or 2 bits of
information
b. Generate initial hypotheses
i. A presumptive diagnosis that the problem solver thinks may be the explanation
for patient’s complaints
2) Exclude or confirm:
a. Hypothetical-deductive method: Interviews to gather information to excludes most
possibilities and confirm one
b. Searches for clues that present if the hypothesis is true, e.g. look for clues in history, PE,
tests
3) Strengthening impressions
a. MD sorts information into meaningful groups
4) Complete history must be done
a. The patient’s history must be taken by the MD, the MD assistant, or the RN
b. Not acceptable for patients to fill out a form
c. To establish rapport while taking history
d. To observer patient’s reaction to questions
PE: 2 types:
1) Selective or ad hoc exam if know what is wrong
2) Complete exam if not know
a. Orderliness, thoroughness, and attention to one thing at a time
- Palpation : Information obtained by touching a feeling
- Know what is normal vs abnormal
- Examination needs many separate actions in order to be properly executed
- Need proper positions for examinations
- Aim for a minimum of change in position
- A tremendous array of tests and procedures is available but could be too much
- Test may be needed
- Additional special tests may need
- Should the test be done?
Data processing: The step by which patient’s data is transformed into problem list
- Does the clue relate?
- How to select clues that relate
- Irrelevant material must be weeded out but not discarded
- Fitting clues together
- Both positive and negative information can be of value
- The absence of a clue is as important as the presence of a clue
- Processing of data depends on the acquisition of data, and on the data collector’s ability to
selectively pursue only certain subsets of information
Problem List: Represent the highest possible resolution of the patient’s problems based on the data
accumulated so far
- Sometimes diagnosis stated
- Others only syndrome listed
- Also included isolated abnormalities
- Psychosocial and socioeconomic problems also included
Problem list can contain:
1) Diagnosis
2) Syndrome
3) Pathophysiologic state
4) Clusters of clues
5) Isolated Abnormality
6) Psychosocio economic state
- Each diagnosis or problem can be raised to the highest level of resolution with the information
at hand
Possible errors:
1) Fail to assemble clues properly into one recognizable group and list each symptoms and sign as
a separate problem
2) Fail to raise the problem to as high a resolution as possible
3) Make premature closure of a diagnosis and unjustified decisions before adequate information is
available
- Initial Problem list is based on the first 24 hours of information and may need adjustment as the
case develop
- Each problem assigned a number and remained permanent
- By the events or by the serious of problem
- Each item assessed
- A Plan formed
- Each problem dealt with separately
- With assessment include items in the database that led you to label the problem
- The urgency of the problem
- Pathophysiology of each problem
- Possible relationship between problems
The Plan is the proposed management for each problem
1) Diagnostic studies / tests and why
2) Proposed treatments
3) What to accomplish for patient education
Final Problem list is completed when the patient if ready for discharge from the hospital
Chapter 3: Problem solving Methods - Problem solving in clinical evidence is the process whereby the doctor finds out what is wrong
with the patient
- Given a chief complaint, the doctor then gathers additional information what guides him
towards a solution
- On the basis of the chief complaint, and the patient’s gender, age, and occupation, the doctor
establishes a list of possible diagnosis (hypotheses), and the asks additional questions to support
or exclude the diagnosis
- After 2 minutes of dialogue, one diagnosis usually emerges as a leading contender.
- The doctor then processes to a selected hypotheses driven subset of PE, aiming to raise the
most likely diagnosis to a level of virtual certainty
- The problem solver must have some knowledge of diseases, their pathophysiological state, and
their symptoms and signs
- He can then ask the correct questions and know where to search for additional clues
Traditionally, students have been taught to gather a data base, and solve problems in separate
blocks:
1) Get a complete history
2) Do a complete PE
3) Order routine lab test, chest x ray, ECG
4) Select the important clues from each of the 3 sources
5) Put them together to fit a known diagnostic pattern
6) Extend the problem list for a clue that does not fit
In real life, experienced MD solve the problem with an subconscious manner
A complete database must usually still be fleshed out to:
1) Make sure the hypothesis is correct
2) Discover possible co existence of other diseases
3) Stumble on clues that don’t fit
4) Rework the order of likelihood
5) Establish a baseline of normality for future comparison
Students must acquire many skills:
1) Interpersonal skills
2) Take history and perform PE
3) Evaluate and process information
4) Formulation of a problem list
5) Assessing the problem list and devising a diagnostic game plan
6) Transform data into written records
7) Internet / Computer skills
8) Present the case to others
- Precise criteria for diagnosing diseases do not exist
- Disease cannot be diagnosed with certainty because they cannot be defined
- We can describe a disease, but cannot define it
- Patients do not conform to textbooks and common features are often absent
- Disease differ: Symptoms or signs can be different in different people
Diagnosis and treatment may proceed simultaneously, determined by
1) the seriousness of the case
2) The time required for more studies
3) The urgency for needed action
4) The level of certainty
- Seemingly simple problems can turn into complex problems
- Is the principal problem limiting to one organ, or one system, or involves several systems, or if
more than one disease exists
- The distinction between a multisystem disease and multiple diseases is not simple
- The prevalence of a particular disease is represented by the # of cases of the disease per
population unit at any particular time
- The probability that a disease exists in a particular patient is a % estimate based on available
data
- Both prevalence and probability of a disease are expressed by the Bayesian symbol P(D)
- The Probability of 50/50 of a patient having a disease is P(D) = 0.5
- Should increase the likelihood before treatment with more clinical information and test
- A revised P(D) must be calculated after each additional clue or test is done
- Provided the sensitivity and specificity of the new information is known
- But physician revises probability by simply adding clinical information
Diagnosis Ways:
1) Early Hypothesis Generation: The single most commonly used and most effective method of
solving medical problems based on
o Age, gender and chief complaint
o Combination of symptoms
o Mention of a cause
o Put the clues together
- The experienced MD develop fixed deductive patterns
- These patterns are integrated as templates into his cognitive structure and can be recalled on
demand for any predicated hypothesis
- If the patient has disease A, then find 5 clues
- If B, find a another 5 clues
- To distinguish between A and B, find 3 studies
2) Clusters: The experienced doctors often used classical groups, triads, or tetrads of clues to
generated hypotheses and make diagnosis
- Error may occur if more than one disease can cause a cluster
- To formulate a cluster, decide if all the symptoms and signs in the cluster apply to a single
diagnosis or several
- The cluster components must be clearly related
3) Pattern recognition: Allows to make diagnosis with a single glance
- Recognition by gestalt
4) Syndromes: A concurrence of manifestations seen together often enough to be more than a
chance relationship
5) The Key Clue: Look for a key clue that is known to occur in a specified variety of clinical situation,
and track it down by seeking further information
- A pivotal clue is the deciding factor around which hypotheses are formed and diagnosis are
made in very complicated cases
- From a list of info, select a few that add up to a table
6) Forming a deferential: The differential diagnosis: A list of possible causes for the patient’s chief
complaint
- It includes all of the medical diseases that may possibly explain the patient’s chief complaint
- A novice may not be able to form an all-inclusive list
- The illness remained undiagnosed because it was not considered
- Should construct a differential diagrams that has been selected out of the classification of
diseases categories:
a) Infectious
b) Neoplastic
c) Endocrine-Metabolic
d) Neuro psychiatric
e) Special organs
f) Connective tissue and autoimmune
g) Hematologic
h) Genetic
i) Traumatic
j) Nutritional
k) Latrogenic and drug induced
- Then must prove one diagnosis and eliminate the others
7) Suppression of information: Separate pertinent data from unrelated info
8) Hunch and intuition: An educated deduction made almost subconsciously on the basis of a
single observation
- Intuition: Based on more solid evidence
- The direct perception of a truth or fact independent of any reasoning process
9) Symptomless Disease: Problems are discovered that are not yet causing symptoms
- An Abnormality
10) Rashes, lumps, and bumps: May be an omen of an underlying disease
11) Tactics with clues:
a. Patterns building: Positive clues are added until a picture that is suggestive of the
suspected disease is constructed
i. Highly sensitive clues
ii. Highly specific clues
b. Template matching: Build the cues that coincides with the textbook template as closely
as possible
c. Weight of evidence: In favor of a diagnosis
i. More positive clues, more weight
12) Algorithms: A printed format that uses a branching pattern to solve a problem
- A test can eliminate large section and directs to follow one path
- In time, algorithmic patterns become imprinted in the minds of MD
- The principal values of algorithms
1) teach the Student
2) Help the MD to think and decide
3) Improve care
4) Decrease the cost of medical care
5) Act as consultants
6) Make chart audit easier
13) Flowcharts: Guide through questions with “yes/no” answers until an endpoint is reached
- Good MD generate their own flowchart for each patient
- Not following printed path ways
14) Data Paths: A pathway that is constructed by the experienced problem solver
- Asking limited questions examining a few areas, getting a few tests
15) Decision Trees: A modified algorithm that depicts the various option and outcomes of a clinical
situation that requires a decision
- Whether to test or treat
- Which treatment has the best outcome
- Risk-cost-benefit analysis
- In constructing the tree, deal with
a. the probability that the problem exists,
b. the treatment options
c. the test’s sensitivity and specificity
d. the morbidity and mortality for each option
e. the mathematical representation of expected outcomes
16) Computers: Match the patient’s symptoms and signs with its stored database, and suggest
diagnostic likelihoods, treatments, requests for additional information and progress
For patients whose clues cannot be easily found, it is best to proceed the traditional way:
1) Gather a complete data base
2) Seek out clues
3) Synthesize them and diagnose disease probability
4) Prove hypotheses
Chapter 4: Clues: the Building Block Types of Clues:
1) Positive Clues: A Clue is positive if it is present and abnormal
- Must consider a clue that is positive that are known to be positive in a suspected disease
- The more positive clues there are, the more likely the disease
2) Negative clues: include items in the medical data base that are either not present or are normal
- Negative clues can be as important as positive clues
- Negative clues serve best to eliminate diagnosis possibilities
- When a student presents a case, a list of pertinent negative findings usually follows the chief
complaint and history, to rule out some of the competing hypotheses
- Negative clues may also tell you how far a disease has progressed
3) Key clues: The key clues may be a symptom, a sign, or a laboratory abnormality
- Each clue prompts you to think of a list of possible causes
- To find a unique approach to solve the problem
4) Decisive clues: A clue that leads the problem solve directly and decisively to the correct
diagnosis.
- It is often found in family history, social history, or the patient profile
5) Primary and secondary clues:
a. Primary clues are those that are related to the disease process in the situ
b. Secondary clues occur at a distant site and are pathophysiologically related
6) False clues; A clue is found, but not related to the clinical picture
- 2 clinical clues may be independent, interdependent, or mutually exclusive
- Often a clue will point directly to the site of disease
- Pinpoint the organ
If a clue does not fit, then
1) Doctor on the wrong track
2) Clue incorrectly obtained
3) Part of a separate problem
4) Just “one of those things”
The intersection of clues: One disease can cause both key clues
- Track down each key clue and determine if they intersect at a common diagnosis
3 properties of clues:
1) Sensitivity: A Quantitative statement of how often this clue is present in patients who have the
disease.
- The more sensitive a clue, the more its absence will exclude a disease
- If a clue sensitivity is 96%, this means 96% of patients with the disease have the clue
- 4% of patients with the disease do not
2) Specificity: A quantitative statement of the frequency with which the clue is absence in person
who do not have the disease
- If a clue is 90% specific for a disease, then 90% of persons not having the disease do not exhibits
the clue
- The clue will be present in 10% of person who do not have the disease
- The more specific a clue, the more its presence will diagnose a disease
3) Relative Importance of a clue: An indication of how significant a role the clue plays in the
pathophysiology of the disease under consideration as well as the weight you attach to its
presence or absence
The ideal clue has a high sensitivity, high specificity, and high relative importance
- Clues may come and go. The stage of the disease is important
- The relationships, timings, and sequence of events weights or logic: If 2 symptoms develop
close together in time, they are probably related
- The sequence is important
- Symptoms occurring in the same system are apt to be related
- If they occur in different systems, they may indicate the presence of 2 different diseases, a
multisystem disease or a disease that spreads
- The clinical picture of a single disease may be altered if another disease becomes superimposed
- All clues point to a diagnosis but the patient has a different disease
- Medicine and diagnosis are not precise sciences that can be solved by machines
- If a clue is present, may need to determine to what degree the clue is present: Quantifiable
- Accurate quantification of a clue and qualification, is usually of great value, and is required to
make the diagnosis, to determine the stage and severity of the disease, its prognosis, and
treatment
Article: Where is the beef - Treatment recommendation In textbooks are out of date
- Textbook descriptions of disease often omit information about the frequency and temporal
characteristics of clinical manifestations in patients with the disorder
- Omission of quantitative information from textbooks descriptions of disease
In studies:
- Clinical findings with the # and % of patients who have each finding
- Details of qualitative features
- Temporal characteristics
- Clusters and patterns of findings
For validity, 4 issues are important:
1) How the diagnoses were verified
2) How the study samples relate to all patients with the disease
3) How the clinical findings were sought
4) How the clinical findings were characterized
- Each clinical manifestation of a disease
- = a test for the presence of that disease
- = the sensitivity of the finding
- = how frequently the test / clue is positive in patients with the disorder
- Clinical studies on the frequency of clues in one disease does not include the frequency of the
same clue in other diseases
Doctors should review evidence about the frequency of clues when
1) Encountering patients with atypical presentation of disorders
2) Making the diagnosis of any condition we see less frequently
3) When the definitions of a disease or its diagnostic criteria are changed
4) When important new research about the frequency is published
5) Teaching others about the diagnosis of a condition
Article: Clinical skills textbooks fail evidence based examination
- Students can learn to solve problems more expertly by using schema that assist storage and
retrieval of clinical knowledge than by rote memorization of lists and “dispersed knowledge”
- Diagnosis involves gathering clinical information and then refining the probability of a particular
diagnosis after acquiring each piece of evidence
- The history and PE can be considered as individual tests
- Textbooks describe only traditional approach in which the student is expected to take a
complete history in an unspecified timeframe, then wait for inspiration to find a diagnosis
- None gave reasons for selecting these particular questions or exams a s basis set for each
system
- No guides for students to distinguish clinical features with high likelihood ratios
- Most omit conditions commonly seen in primary care
- They discussed the clues of disease X rather than show how a symptoms complex can be
analyzed to make the diagnosis
- Fail to integrate lessons from medical education research and available evidence about the
effectiveness of the aspects of PE
- Textbooks fail to report the precision and accuracy of clinical signs
- Time efficient, selective clinical examination, without cutting corners by using the principles of
clinical reasoning and problem solving
Article: Stimulating the adoption of Health IT
- American recovery and reinvestment Act of 2009 (ARRA)
1) Prevents dramatic state cuts in Medicaid
2) Expands funding for preventive HC services and HC research
3) Helps the unemployed buy health insurance
4) $19B to promote the use of HIT and EHR
- The HIT components of ARRA – HITECH – EHR are essential to improve the health and HC of
Americans
- Face challenges:
- Few doctors and hospitals have basic EHR
- Barriers for adoption:
1) Substantial cost
2) Perceived lack of ROI
3) Technical and logistics challenge
4) Concerns on security and privacy
HITECH most important feature
- Its clarity of purpose: as a mean of improving the quality of HC, the health of populations, and
the efficiency of HC systems
The Office of the National Coordination of HIT (ONCHIT) to create a strategic plan for a
nationwide interoperable HIT, a plan that must be updated annually
- $17B in financial incentives to adopt and use HER
- Doctors receive extra medicare payments for the “meaningful use” of a “certified” EHR that can
exchange data with other parts of the HC system
- Starting in 2011 could collect $44000 over 5 years
- Meaningful use of EHR in 2011 will earn hospitals a one time bonus payment of $2M plus an
add-on to the diagnosis related group (DRG)
- Physicians who are not using EHR by 2015 will lose medicare fees
- ONCHIT sets aside $200M to support the development of health information exchange
capabilities at the regional and state levels
- HITECH extends the HIPAA to health information vendors not previously covered by laws
- Promptly notify patients when health data have been compromised
- Limits the commercial use of the information
Major hurdles:
1) ONCHIT operating on a very tight schedule
2) Infrastructure to support HIT adoption in place before 2011
3) Define 2 critical terms: “meaningful use” and “certified EHR”
a. But many “Certified EHR” are not user friendly or meet the HITECH goal of
improving quality and efficiency in HC system
b. Physicians will have to use them effectively
c. Take advantage of embedded decision support
4) Requirements cannot set too high
5) Depends on changing the HC system overall payment incentives so providers benefit from
improving the quality and efficiency of their services
Article: What is Wegener’s Granutomatosis
- Diagnostic errors are hard to fix
- The best we can do is improve information flow try to prevent handoff errors, and teach
ourselves to perform meta cognition: Think about our own thinking, so that we are aware of
common pitfalls and catch them before we pull our diagnostic trigger
- Human response to illness is virtually limitless, as are the illness themselves
- Medical AI program, most cannot handle unexpected or extraneous data
- Programs are clunky and expensive
- Need work to enter the necessary information into the computer program to generate output
- Experts tend to be skeptical of computers smarter than they are
- Breakthrough when computers began to process statistical correlation, learning that some
words often keep each other company in literature
- Computers mine correlations from online sources of information for answers
- Run > 100 algorithms to try to answer a question
- The results of these algorithms are back tested for plausibility
- When one of the answers crosses the plausibility threshold, the computer rings in
- Mimic how the brain function: Instantaneously sorting through thousands of possibilities,
testing them against known patterns, ultimately setting on the most plausible matches
- For MD, to perform iterative hypotheses testing, developing a list of potential diagnosis that
might fit a given set of facts and then a testing strategy to render some of the possibilities more
likely
- Each additional diagnostic test has false negatives and positives and needs to be interpreted n
the light of prior possibilities: Applying the Theorem of Bayes
- The MD settles on an answer when the probability of one of the diagnoses crosses a magical
threshold
- Medical AI will learn from their experience
- To learn that other patients like this one turned out to have the disease
- And adjust its algorithm accordingly
- It is up to human to screen through the suggestions
- Machine is not capable of judgment or empathy
Chapter 9: Impact of new technology
- Numerous new technologic procedures developed during the past 3 decades have substantially
shortened and simplified the diagnostic process
- In some cases, previous testing sequence have been made obsolete
- New procedures must be first be evaluated for reliability
- Cost benefit risk factors need to be appreciated
- What is considered correct diagnostic traffic flow may be detoured
- Many formally used diagnostics plans have been radically altered by new imaging techniques
- This reduces costs by reducing the # of hospital days needed
- But false positive results may require repeat studies, and additional studies may increase HC
costs
- Different doctors and hospitals may pursue varied procedural sequences
Ultrasound: images obtained from the reflection of high frequency sound waves
- Inexpensive, non invasive, harmless
- The reliability of ultrasound is dependent on the expertise of the technician, the interpreter, and
the quality of the equipment
- The M mode method detects and measure motion of the ventricular wall
- 2D ultrasound uses a sound beam that sweeps across the chest
- Good in cardiac diagnosis where motion is a factor and anatomic and functional abnormalities
can be detected
Color Doppler: Ultrasound study integrates the anatomic information provided by 2D with color
coded flow patterns within the heart
- Moving objects reflect and alter the pitch of sound waves
Computed Tomography (CT)
- X ray signals from a body encircling source are collected and computer synthesized into cross
sectional image of a body slice
- Different tissues have different shades of gray, highlighted by injection of iodinated radiographic
contrast materials
- The sensitivity and specificity of the procedure are increased
- CT decreases the need for more invasive tests
- CT should not be used to supply repetitive or redundant information
- Should be goal-directed with a specific problem in mind
Magnetic Resonance Imaging (MRI): Offers images that reflect anatomy and biochemistry
- Gives structural and biochemical information about tissues
- Essentially no radiation exposure
- MRA: Magnetic Resonance Angiograph
Nuclear imaging: Chemicals tagged with radioactive isotopes are administered intravenously
- Radioactivity is measured over the special site to which the chemical is selectively attracted
Biopsies: Use with CT scan to quickly direct a probing biopsy needle to a tiny target
- Expensive procedure
Digital Subtraction Angiography: The computerized system for visualizing arterial trees by the
intravenous injection of a contrast material
- Computer stores the digitized pre injection image and subtract it from the post injection image
leaving only the dye in the arteries to be seen
Positron Emission Tomography: Tracer amounts of biochemically important substrates are labeled
with very short half life positron emitting radionuclide
- The in – vivo distribution of these tracers is measured by special scanning devices
- Limited to large institution
- Positrons must be manufactured onsite with nuclear accelerators
- Most new technology involve the radiology department
- Are we better off?
- Always have imperfect information
- Skill of the performer is variable
- Conflicting or indeterminate results may result in another test
- What should be the appropriate test given?
- When results from 3 tests are not the same
- False positive and false negative results exist
In general:
1) Diagnosis more easily
2) More quickly
3) More correctly
4) More cheaply
5) Surgery simplified
6) Exploration avoided
7) More diagnoses made
8) Disease process is better categorized and understood
In some cases, we can unequivocally state the best procedure for a prospective diagnosis
Chapter 10: Into the Solver’s Mind
- 2 minutes of conversation should be enough to establish a list of leading diagnostic contenders
- Rank them in the order of likelihood and establish one or 2 possibilities as being the most likely
- There can be distracting bits of information that mandate the consideration of other hypotheses
Chapter 11: From Patient to Paper
Procedures
1) Dialogue between patient and MD
2) Patient did most of the talking
3) MD follows no rigid format
4) Misses little format
5) Tracks positive clues
6) Goes off on tangents
7) Then returns to the point
8) At the conclusion of interview, PE is presented
9) Try to formulate a problem list based on the PE and history
10) Case Written up
11) Any missing information, ask the patient
12) The initial problem list
13) Assessment
14) Plan for each problem
15) And a final problem list following the written history and PE
Written Record:
1) History of present illness
2) Past medical history
3) Patient profile
4) Family history
5) Systems review: General, head, skin, eyes
6) PE Results
7) Problem List
8) Assessment of Problem list and plan
9) Studies
10) Final Problem List
a. Combine problems
b. Escalation of problem priority
c. New Problems found
11) Reassessment
Article: A look at Google Scholar
Why Google is popular
1) Simplicity, speed and coverage. Users have faith in Google Branding
2) Google does index a lot of content
However:
1) PubMed records in GS are a year out of date
2) Some major health science publishers are not crawled by Google
3) Some major Canadian content is inadequately indexed
4) Canadian health content from recognized web sites are not crawled
5) “Grey” Literature are not indexed
6) GS lists articles by how relevant and popular they are, not by how current
7) Ranking of older research is a big problem
8) Lack of re sorting options
9) Filtering of results is difficult
- GS search results significantly different and less
- Users should search publisher sites directly
- To maximize recall, search PubMed by keyword and MeSH simultaneously from the homepage
- GS coverage is incomplete, retrieving fewer unique citations than publisher’s site or PubMed
- GS overall performance is robust and comparable than other specialty health search engines
- GS capable of getting bibliographic information from references at the end of articles
- GS link the PubMed records showing the URL
- “Cited by” linking users to related research. GS provides for free
- Article linking products are compatible with GS
- GS offers linking options under preferences
- Scirus: lists its content sources
- Provides focused channel searching by content provider
- Improved customization and flexibility for more precise searching
- In PubMed, a class of drug can be searched by “exploding” a subject heading and its narrower
items
- GS and Scirus not able to limit searches by publication type or research methodology
- GS or Scirus cannot search by age and gender
Best: OVID MEDLINE: Best functionality, and flexibility for building and manipulating sets developed
using PICO
- PubMed interface not as intuitive
- Clinical queries on PubMed and OVID MEDLINE
- PubMed very current links to open web and growing free content
- Saved searches for later retrieval
GS :
- Easy means to Access health literature
- Simple browsing
- Known item searching
- Link to Free materials
MEDLINE: Where literature review are required (grants, clinical trials, or systematic reviews)
- Clinical queries must be answered by replacing requests in context
Article: Google Scholar: A Source for clinicians
- Ideal tool for finding clinical info should be:
1) Fast engine that provides the best hits from scholarly journal literature and clinical resource
2) Succinct reviews and best evidence
3) Links to key papers
4) Balance popularity with relevance and quality
5) Enable search refinement
6) Provide integrated powerful access to many sources
- But results emphasize pages that are cited more often
- Creates a bias toward older literature
- No similar pages
- Only 1/15 of PubMed Records access
- No “Did you mean” feature
- Only major health database used in MEDLINE
Competitors to GS:
1) TRIP database
2) Cochrane Library
3) Info Retriever
4) UpToDate
5) Clinical Evidence
6) MDConsult
7) Access Medicine
8) STAT!Ref
- MedLIne = Map of Medical literature
- Same for PubMed
- GS Not useful choice for Clinicians
Article: The Medical Literature
- Practicing evidence based medicine involves integrating individual clinical expertise with the
best available evidence from systematic research
- First step in the search for evidence
- Identify uncertainties in patient care and formulate these into questions
- To answer focus clinical questions, the most efficient approach is to begin with a “pre-filtered”
evidence based medicine resource such as :
o Cochrane Library
o Clinical Evidence
- That are updated with studies on a regular basis and easy to search
- More general medical questions use electronic medical textbooks
o UpToDate
o Scientific American Medicine
o MEDLINE: For both focused and background medical questions: Search difficult
o Best Evidence: Good place to start looking for answers to focused clinical questions
150 medical journals are systematically search on a regular basis to identify
studies that are both methodologically sound and clinically relevant
Substantially smaller and easier to search
Relevant trial may publish before 1991
If study pertain to subspecialty care than general internal medicine
o Cochrane Library: Focuses primarily on systematic reviews of controlled trials of
therapeutic interventions
Provide little help in other areas
3 main sections
1) Cochrane Database of Systematic Reviews (CDSR)
a. Include complete reports for all the systematic reviews
prepared by members of the Cochrane collaboration
2) Database of Reviews of Effectiveness (DARE)
a. Includes systematic reviews published outside the collaboration
3) Cochrane Controlled Trials Registry (CCTR)
a. References to trials Cochrance investigators found by searching
MEDLINE, EMBASE, etc.
b. Create more complex search strategies that include Medical
Subject Heading (MeSH) terms and logical operators
o UptoDate: Electronic textbook updated every 4 months
Very well referenced
No Methodological quality criteria
o MEDLINE: Maintained by US National Library of Medicine
9M citations to both clinical and pre-clinical studies
Pre Medline: Includes citation but not yet indexed
Search free using PubMed
Searching requires careful thought
Know how to use MeSH terms, textword searching and exploding with logical
operators
Tutorials at www.docnet.org.uk/drfelix/medtut.html
- MEDLINE indexes choose MeSH terms for each article
- Indexer reference articles under the most specific subject heading available
- Use a more general heading risks missing many articles of interest
- Use Explode command identifies all articles that have been indexed using a given MeSH Term
- Use Text word searching to identify all articles with the study title or abstract includes a certain
term
- Comprehensive search use both MeSH terms and Text words
- PubMed use the most appropriate MeSh terms and explode MesH terms
- Use Search hedges: Systematically tested search strategies that help identify methodologically
sound studies pertaining to questions of therapy diagnosis, prognosis or harm
- A complete listing of strategies is available
- Along with sensitivities and specificities for each different approach
- PubMed has a special section called Clinical queries
- Clinicians can use single best terms for finding higher quality studies, includes:
1) Clinical trial for treatment (Publication type)
2) Sensitivity (Text word) for diagnosis
3) Explode Cohort studies (MeSH term) for prognosis
4) Risk (text word) for harm
Answers to more general questions:
- Pre-filtered Evidence based medicine resources, e.g. Best Evidence, not helpful
- Refer to a textbook is likely to be faster
- Resources on Web, some may fail to meet criteria for evidence based guidelines e.g. MD Consult
- Web Site by reputable organization
Clinical Evidence: A growing compendium of evidence pertaining to treatments of specific
conditions
- Electronic resources that facilitate simultaneous searching of MEDLINE, Best Evidence, and the
Cochrane Library now available through services as OVID Technology’s Evidence Based Medicine
Reviews
Article: BMC Medical Informatics and Decision Making: Software askMEDLINE
- askMEDLINE evolved from the PICO (Patient, Intervention, Comparison, Outcome) search
interface
- User can simply enter a clinical question, then let the search engine retrieve relevant journal
articles
- User does not need to check the correctness of PICO elements
- User can manually enter search item
- Many go directly to other subscription only resources but they cover fewer journals than
MEDLINE
o Greater lag in published article
- Other accessed less evidence based resources
askMEDLINE: Search MEDLINE / PubMED using free text, national language query
- Links provided to journal, articles
Use multi-round search strategy:
1) First round ignores punctuation and deletes words from a “Stop word” list
2) If no journal are found in first round, the “All fiends” words are deleted from the query
a. Only MeSH terms remain
b. Or if 1st round retrieve > 50,000 , the all fields word are put back into the query
i. Contain all MeSH terms and All Fields words
3) In 2nd round, if retrieve 1 to 50,000, search stops
a. If search = 0, another list of terms is checked
4) If 4 articles found, search is done for related articles of the top 2 articles
British Medical Journal published a POEM (Patient Oriented Evidence that Matters)
Chapter 7: Digits, Decimals, Doctors
- In 1763, Sir Thomas Bayes devised a set of Theorems relate disease prevalence and disease
probability to the sensitivity, specificity, and predictive values of a test or clue
- Physicians use Bayes Theorems for:
1) The incidence of a disease in a population
2) The incidence of a specific clue in a disease
3) The incidence of this clue in person who do not have the disease
- D represents the presence of disease
- ^D = Absence of a disease
- C = presence of a clue (positive test)
- ^C = absence of a clue (negative test)
- P(D) = the prevalence of a disease or the probability that a disease exists in a particular patient
- Prevalence = # of cases per unit of population that exists at any one time
- Incidence = # of cases per unit of population over a given period of time
- Prevalence = 10 cases / 1000 people = 0.01
- Annual Incidence = 100 cases / (1000 people x year)
- P(D) in a patient = 0.2 = in a cohort of 100 identical patients, 20 patients will have the disease
- P(C/D) = 0.85 = 85% of persons with a disease will exhibit a given clue
- = The clue has a true positive (TP) rate of 0.85
- = The sensitivity of the clue = 0.85
- = Positivity in disease = 0.85
- P(^C/D) = 0.15 = The clue will be absent in 15% of cases
- = The false negative (FN) rate = 0.15
- = 1 – sensitivity
- Rule of ones = a clue is either present or absent
- P(^C/^D) = 0.95 = 95% of persons who do not have the disease do not have the clue
- = In 95% of the person without the disease, the clue is truly negative (TN)
- = The specificity of the clue = 0.95
- P(C/^D) = 0.05 = 5% of the people without the disease do have the clue
- = The clue is false positive (FP)
- = FP rate = 1 – specificity
- TP + FN = 1
- TN + FP = 1
- P(D/C) = Probability of a diagnosis given a clue
- = Positive Predictive Value (PV+)
- P(^D/^C) = Probability of the absence of a diagnosis given the absence of a clue
- = Negative Predictive Value (PV-)
2 x 2 table
D ^D
Positive Result + TP = P(C/D) FP = P(C/^D) PV+ = TP / (TP + FP)
Negative Result - FN = P(^C/D) TN = P(^C/^D) PV- = TN / (TN + FN)
As P(D) increases, PV+ increases, PV- decreases
- Incremental gain = Pretest P(D) – Posttest P(D)
- The operating characteristics of a clue are its 2 most important procedures: Sensitivity and
specificity
- Test efficiency = (Sensitivity + Specificity) / 2
- If a test raises the probability of a disease from 0.1 to 0.6, it is a good test
- If a test reduces the probability from 0.3 to 0.1, good test
- How to derive S and S
- Out of 600 patients with known disease, 540 have TP, 60 FN
- Sensitivity = 540 / (540 + 60) = 540 / 600 = 0.9
- Out of 400 patients without known disease, 320 have TN, 80 have FP
- Specificity = 320 / (320 + 80) = 320 / 400 = 0.8
- If original guess = P(D) = 0.1 and PV+ of a test is 0.33, a positive test would increase the
likelihood only a little
- But a negative test would rule out the disease with 99% certainty
- If original P(D) = 0.4, a positive test increases the P(D) to 0.75, a negative test reduce to 0.08
4 relationships between a clue and a disease
1) If a clue = 100% sensitive, then if the clue is absent, the disease is not there
2) If a clue = 100% specific, then if the clue is present, the disease IS there. But the clue is not
necessary present in all patients who have the disease
3) A clue that is 100% sensitive and 100% specific : If the clue is present, the patient has the
disease. If the clue is absence, no disease.
4) The clue is often, but not always present if the disease exists and is sometimes, but not
often present, if the disease does not exist (False positive)
- Ideally, a clue separate healthy from disease, no overlap
- Usually, there is an overlap between healthy and disease
- In deciding on a separator point, consider the consequences of false negative and false positive
results
- To minimize FP or FN?
- If initial P(D) = 0.5 and PV+ > 0.9, achieve diagnosis certainty
- Cannot be used because of high false positive
- When original P(D) is low, test with high FP is of no value
- The marginal benefit of a test = Posttest – Pretest
- The greatest value of a test is obtained when it is ordered with a diagnosis already in mind
- If a clue can mean 2 diseases and only these 2 diseases.
- P(D1) = 0.2 = common disease,
- P(D2) = 0.01 = rare disease. Population = 1000
- C = 10% sensitive in D1 = uncommon clue in disease 1
- C = 50% sensitive in D2 = common clue in disease 2
- If D1 exists, then there is no D2
- If D2 exists, then there is no D1
D1 = 200 ^D1 = D2 = 10
Positive Result + 200 x 0.1 = 20 0.5 x 10 = 5 PV+ = 20 / (20 + 5) = 0.8
Negative Result - 200 – 20 = 180 10 – 5 = 5 PV- = 5 / (180 + 5)
Probability of D1 given the clue C is 0.8
D2 = 10 ^D2 = D1 = 200
Positive Result + 0.5 x 10 = 5 200 x 0.1 = 20 PV+ = 5 / (20 + 5) = 0.2
Negative Result - 10 – 5 = 5 200 – 20 = 180 PV- = 180 / (180 + 5)
Probability of D2 given the clue C is 0.2
- If Clue C is present, D1 is 4X more likely than D2
- The uncommon manifestations (clue) of a common disease are more common than the
common manifestation (clue) of an uncommon disease
- Clue in 2 diseases:
- P(D1) = 75%
- P(D2) = 25%
- Population = 100
- Test is negative in 20% of patients with D1
- Test is negative in 80% of patients with D2
D1 = 75 ^D1 = D2 = 25
Positive Result + 75 – 15 = 60 25 – 20 = 5 PV+ = 60/ (5 + 60)
Negative Result - 0.2 x 75 = 15 0.8 x 25 = 20 PV- = 15/(15 + 20) =
Probability of D1 with no Clue C = 15/(15 + 20)
D2 = 25 ^D2 = D1 = 75
Positive Result + 25 – 20 = 5 75 – 15 = 60 PV+ = 5 / (5 + 60)
Negative Result - 0.8 x 25 = 20 0.2 x 75 = 15 PV- = 20/(15 + 20) =
Probability of D2 with no Clue C = 20/(15 + 20)
- P(D) changes as each new bit of evidence is added
- If test first reduce mortality by > 50%, it is worthwhile test
Treating a patient with 50% P(D) = Treating 100 patients with 50 have and 50 no
Chapter 8: How to use test
- Reasons for doing tests: To find out what is wrong with the patient
1) Diagnosis
2) Prognosis
3) Screening
4) Monitoring
5) Determination of baseline data
6) Decisions
7) Defensive medico-legal reasons
- Diagnosis is most important: A test or procedure helps to detect a disease
- Prognosis can be predicted by noting the degree of test abnormality
- Screening to exclude diseases I health persons
- Multi-phasic screening: Screen for many diseases at the same time
- Targeting screening is the ordering of a single test on a single person
Order of testing
1) From cheap to costly
2) From less to more risky
3) From simple to more complex
- Try to use test with the highest sensitivity, specificity, and predictive values
- If the need and benefit far exceed the cost and risk, the test should be done
- If the data derived are of questionable value, some risk, and high cost, do not do the test
- Know the monetary costs of every test
- Know the benefits
- Know the risks
- Know what normal is
- How abnormal is a test before concern
- At each increment of abnormality, the test has different operating characteristics, resulting in
different true positive and false positive rates
- The more abnormal the test results, the more likely it is to be truly positive and less likely to
be falsely positive
- Test to be perfect, it must be:
1) Accurate: give correct result
2) Precise: Reproducible
3) Discriminating – Distinguish between health and disease
4) Pain Free
5) Risk Free
6) Inexpensive
7) Useful
- Test that measure continuous variables do not clearly distinguish between health and disease
Reason for errors:
1) Prelab Error
a. Improper preparation of patient
b. Taking of drugs
c. Mislabeling of tubes
d. Obtaining poor specimens
2) Intralab Error
a. Misplacing of specimens
b. Error in recording
c. Illegible reports
d. Referent values and referent intervals
e. Placed in wrong chart
f. Incorrectly accessed by computer
3) Final Error
a. Physician interpret incorrectly
- Which tests to order?
1) Admission Studies
2) Problem oriented tests
a. If the initial batch of test does not confirm the primary clinical impression, tests may
be ordered for the second likelihood
b. Order tests for all leading likelihoods at once and want for all results to return
- Tests can become obsolete, or assume greater values
- Well done studies help us decide whether to use a test
- Criteria for a worthwhile test
1) Test performance
2) Criteria for abnormality
3) Efficient operating characteristics
4) Reasonable approximation of disease probability before doing the test
- Reasons for overuse:
1) Greater availability and variety of tests
2) More diseases are diagnosed and their treatment monitored by lab test
3) Ability to monitor drug level and do toxicologic studies
4) For medico-legal reasons
5) Order tests to be complete: Omission is greater sin than commission
6) Lack of Physician education in the proper use of tests, their operating characteristics, and
their predictive values
7) Patients do not directly pay their own bill
8) Different department biased in favor of its own procedure
9) Frequent need to follow up on a false positive result
- Multiple tests lead to unexpectedly positive test results:
1) A slightly abnormal test result may merely represent 2.5% on either end of the reference
interval
2) The degree of abnormality must be considered
- For an unexpected abnormal result, it must be added upon
1) Repeat test
2) If normal, forget about it
3) If still abnormal, consider any drug effect?
4) Consider a differential diagnostic for the abnormality
5) Track the problem down
- The most important test ordering pattern to be established is that there be NO pattern
- A separate catalog of test that list everything should be available
- Doctors must know where to retrieve selected subsets of available tests of to eliminate a
possible diagnosis
- Strategies for testing:
- From chief complaint, formulate a protocol for getting more data
- After history, PE are done, and initial hypotheses are formed
- Further tests and studies are often necessary
- Order tests in series or parallel
- In series after the first test, the second may not need
- The P(D) of first test becomes the new P(D) for the second test
- In parallel, they may contradict
- Which test to do?
- Do the test with the greatest sensitivity or the test with the greatest specificity
- Choose the highest sensitivity if cannot have a false positive
- Trade off between missing a patient with the disease or treating a patient without the disease
- When new tests become available, strategies change
- Use Printed Aids to help
Article: Prediction of Pulmonary Embolism in the Emergency Department - Diagnosis of a disease requires clinical probability assessment
- Implicit assessment is accurate but is not standardized
- Current prediction rules have shortcomings
- Construct a simple score based entirely on clinical variables and independent from physicians
implicit judgment
- Collect data
- The variables statistically significantly associated with the disease in uni-variate analysis were
included in a multivariate logistic regression model
- Points were assigned according to the regression coefficients
- The score was externally validated in an independent cohort
- Score based on 8 variables
- Entirely standardized and is based on clinical variables
- Revise Geneva Score:
- Results: The score comprised 8 variables (points): - age older than 65 years (1 point), - previous deep venous thrombosis or pulmonary embolism (3 points), - surgery or fracture within 1 month (2 points), - active malignant condition (2 points), - unilateral lower limb pain (3 points), - hemoptysis (2 points), - heart rate of 75 to 94 beats/min (3 points) or 95 beats/min or more (5 points), and - pain on lower-limb deep venous palpation and unilateral edema (4 points). - In the validation set, the prevalence of pulmonary embolism was - 8% in the low-probability category (0 to 3 points), - 28% in the intermediate-probability category (4 to 10 points), and - 74% in the high probability category (_11 points).
- Entirely standardized and is based on clinical variables
Article: Using Clinical Prediction Rule
- Different assessment of the pretest probability result in very different management strategies
- Potential sources to determine pretest probability:
1) Clinical expertise
2) Audit of practice
3) Primary Literature
- Clinical Prediction Rules (CPRs) : a tool that quantifies the individual contributions that various
components of the history, PE and basic lab test results make toward the diagnosis, prognosis or
probable response to a treatment of an individual patient
- Most useful when directed at frequent problems for which the stakes are high or cost saving is
possible
- CPR invoke complex algorithms to obtain a result
- Many computer based models have been developed
- Med.mssm.edu/ebm
- A small library can be found at Mount Sinai EBM web site
- 4 Levels:
1) Impact analysis performed
2) Validated
3) Validated
4) Derived
- Need to identify the level of evidence of a particular CPR or the population from which the rule
was derived
- Both information necessary to apply the rule safely and in the correct setting
Sensitivity and Specificity
Disease Present Disease Absent
Result + A B
Result - C D
A+C B+D
- Sensitivity = A / (A + C)
- SnNout = Sensitivity Negative Out
- = When a symptom has a high sensitivity, a Negative result rules Out the disease
- Specificity = D / (B + D)
- SpPin = Specificity Positive In
- = When a symptom has a high specificity, a positive result rules in the disease
Algorithm for the prediction of disease
1) Revised Cardiac Risk Index (RCRI)
- To determine cardiac risk in patient undergoing non-cardiac surgery
- Underestimated risk
- VSG-CRI: New index using multivariate analysis
- For predicting re operative risk in vascular surgery patients
2) Centor for strepthroat
3) McIassc Rule for Strepthroat
4) Revised Geneva Score for the Prediction of Pulmonary Embolism
Chapter 5: Data Resolution Skills
- Data Resolution skills include knowing:
1) How to put things together
2) Signs and symptoms of disease
3) Additional high yield data to solve knotty problems
4) Clues that exclude or confirm a diagnosis
5) Subsets of data to be investigated for many common complaints
- Must quickly decide what to do with information and where to categorize it. Methods:
1) The selection of data subsets for 8 mini-cases
2) The performance of relevance exercises involving 3 common medical presentations
3) The Decoding of 3 problem sets
- Start with age, gender, and chief complaint:
1) Make a short list of the diagnostic contenders
2) Decide which questions might elicit the most valuable information
3) Select the part of the PE that will help further narrow the field
4) Choose the tests that are most likely to clinch the diagnosis
- Request early data that will be very predictive and do not ask for useless information
- Need to be very selective in search for clues and know exactly what to look for an how to find it
1) List the differential diagnosis
2) Ask 3 to 7 questions
3) Examine 3 to 7 parts
4) Decide which 1 to 4 tests will help the most
- The list can be reduced if the answers to a question of the results of an examination leads to one
direction
- The algorithm principles by which we build on pieces of positive data to direct our further line of
inquiry is a method commonly used by the problem solver
- Diagnosis must be supported by harder evidence
- Evidence for serious illness must be robust
- Need to decide which item is:
1) Relevant
2) Possibly Relevant
3) Pertinent Negative
4) Irrelevant
5) A Separate Problem
- Usually we deal with entire clusters of data wherein individual clues tend to reinforce each
other
- As each new clue is added, you are building a weight of evidence for the diagnosis
- With the appearance of each positive clue, you increase the likelihood of D:^D
- If only a few clues are positive, if there is a considerable overlap of clues between 2 diagnosis, or
if some clues are unexpectedly positive or negative, further investigation is needed
- By obtaining more data from history and PE or test
- The same set of clues occurring in dissimilar population subsets can have vastly different
diagnostic and therapeutic implications
- In each instance, you must analyze the presentation, see if the age, gender, occupation,
geography, or other demographic features influence your impression
- Occasionally, an additionally stated clue will help you draw a conclusion, and you must seek
more data to confirm or reject your hypothesis
Chapter 6: More Clinical Tools
- The successful doctor is able to interlace the patient base data with his own knowledge base in
order to arrive at a sensible problem list
- Traditionally:
1) Gather a complete data base
2) Diagnose
3) Act
- But usually provisional diagnosis based on the first few bits of information, and evidence for and
against this diagnosis is then sought via a branching technique
- The problem solver must always consider the patient’s age, gender, race, habitat, and the
natural history of disorder when weighing the various contender for the diagnosis
- Many disease have courses that fit specific patterns
- The pattern, progression, and intermittency of the disease process must be documented in the
history of the present illness
- Match the patient’s history of present illness to the known clinical courses of suspected
diagnosis
- 3 Maxims:
1) Common diseases occur commonly
2) Uncommon manifestations of common disease are more common than common
manifestations of uncommon disease
3) No disease is rare to the patient who has it
- Think Common, but Remember Rare
- Solve by reverse:
- Search the index of a standard disease oriented textbook, to look for disease associated with a
symptom, signs, and lab abnormality
- Depth of study:
1) Short Cases: Solved in a few min
2) Revisits: Chronic visits: 10 to 15 min
3) Puzzling Cases: 10 Days
4) Hospital Cases: Should have complete work up
5) Midlong Cases
6) The Complete Workup
- Short Cuts
1) Diagnosis at a Glance
2) Simple Inspection during history taking and the beginning of PE
3) Zeroing in:
a. Piece Together 3 to 4 Clues,
b. formulate 1 to 2 hypotheses,
c. seek out an additional bit of information that confirms one hypotheses and
eliminate the others
- Most commonly used method to solve problems
4) Be Free wheeling: Shift to a more selective search of a patient’s database
5) Symptoms and signs together: Ask questions at the same time as you examine each systems
- Communications can be enhanced if likelihoods are expressed in % rather than in words
- Always have the patient quantify the symptoms
- Be Precise with units for measurements
- A physician usually works with probability, likelihood, and weight of evidence
- Diagnostic certainty consider:
1) Seriousness of the disease
2) Age and general conditions of patient
3) Risk of further study
4) The already existing disease probability
5) Costs to the patient and society
- The degree of diagnostic certainty can be revised upward by:
1) The addition of positive clues
2) The elimination of contending hypothesis
3) Confirmatory positive tests
- The degree revised downward by:
1) No additional positive clues
2) A contending hypothesis increases its own likelihood
3) Findings of Supposedly confirmatory tests are normal or negative
- Before making a diagnosis, you may sometimes have to undiagnosed a patient
- When wrong labels are applied, patients are treated for illness that are more serious than the
one they already have
- 2 persons with the same disease can present vastly different clinical pictures
- Many symptoms, physical signs, and lab test for a patient disease have different sensitivity
- There is a % of patients problems that we are unable to solve without patience
Article: Likelihood ratios: Getting Diagnostic testing into perspective - The Likelihood ratio (LR) is a semi quantitative measure of the performance of diagnostic tests
which indicates how much a diagnostic procedure modifies the probability of a disease, and is
calculated from the sensitivity and specificity of the test
- Frequent belief : test are more definitive than medical history and PE
- However, testing exposes patients to risks, compromise their well-beings, increase costs, and
may impede diagnosis when false positive results
- New tests provided conclusive information in 30% of cases and misleading in 10%
- History and PE established the diagnosis in more than 60% and misleading in less than 2%
- Doctors have neglected objective assessment of bedside evaluation as a diagnostic tool
- The Likelihood ratio (LR) for a test result is the ratio between the chance of observing that
results in patients with the disease in question, and the chance of the result in subjects without
the disease.
- LR for a positive test: D+ / ^D+
- = Sensitivity / (1 – Specificity) = TP / FP
- LR for a negative test: D- / ^D-
- = (1 – Sensitivity ) / Specificity = FN / TN
- The product of the LR and the pretest odds determines post test odds
- LR range from 0 to infinity
- A value of 1 means that the test provides no additional information
- Ratios above 1 increase the likelihood of disease
- Ratios below 1 decrease the likelihood of disease
- Odds = Probability / (1 – Probability)
- Probability = Odds / (1 + Odds)
- By using a normogram, these calculations may be avoided
- When several tests are performed, their joined LR is equal to the product of the LRs of the
individual tests
- Tests cannot be compared linearly by their LR
- The power of a test with a LR of 100 is NOT 10X greater than that of a test with a LR of 10
- LR may be inaccurate if the studies defining test performance are of poor quality
- Differences between the population of a test and the applied
Integrate Clinical Evaluation + Ancillary tests
- Case 1:
- DVT Sensitivity = 97%, Specificity = 97%
- LR+ = D+ / ^D+ = Sensitivity / (1 – Specificity)
- = 0.97 / (1 – 0.97) = 32
- LR- = D- / ^D- = (1 – Sensitivity) / Specificity
- = (1 – 0.97) / 0.97 = 0.03
- Pretest Probability of DVT = 85%
- Pretest Odds = Pretest Probability / (1 – Pretest Probability)
- = 0.85 / (1 – 0.85)
- = 0.85 / 0.15 = 5.667
- Post Test odds = Pre Test Odds X LR
- If the test is positive, positive post test odds = pretest odds x LR+
- = 5.667 x 32 = 181.333
- For negative result, negative post tests odds = pretest odds x LR-
- = 5.667 x 0.03 = 0.17
- Post test probability = Odds / (1 + Odds)
- Positive Result = Positive Post Test Probability
- = Positive Post Test Odds / (1 + Positive Post Test Odds)
- = 181.333 / (1 + 181.333) = 99.4%
- Negative Result = Negative Post test Probability
- = Negative Post Test Odd / ( 1+ Negative Post Test Odd)
- = 0.17 / (1 + 0.17) = 14.5%
- An increase in the probability of disease from 25% to 85% (by a refined clinical assessment) is an
increase in odds from 25/75 = 0.333 to 85/15 = 5.667
- Likelihood ratio = 5.667 / 0.333 = 17
- Steps:
1) Conversion of pretest odds to post test odds
2) Conversion of post test odds to post test probability
3) Conversion of an increase in the probability of disease to a LR
- Start:
- Given: Pretest Probability
- LR of Positive test
1) Pretest Probability = a
2) Pretest Odds = a / (1 – a)
3) Post Test Odds = Pretest odds x LR+
4) Post test probability + = Post test odds+/(1 + Post test odds+)
By refined assessment( e.g. test), increases the chance of LVT from 25% to 85%
- Initial Odds = Initial P(D) / ( 1 – Initial P(D)) = 25 / (1 – 25) =25 / 75
- Final Odds = Final P(D) / (1 – Final P(D) = 85 / ( 1 – 85) = 85 / 15
- Likelihood Ratio = Final Odds / Initial Odds
- The reliability of a test depends on experience
- Clinical evaluation is also operator dependent
- How to determine the pretest probability?
- In many disease, pretest probability is estimated by combining epidemiological data (about
population prevalence) with clinical impression based on prior education (Similarly of the
patients picture to descriptions of disease in textbooks), personal experience, or pattern
recognition
- The impact of refined assessment is greater than the information from a test
- Clinical evaluation is not necessarily more precise than ancillary tests, but its accuracy is of
marked consequence to the correct interpretation of test results
- In some diseases, the reliability of clinical data has been quantified or specific criteria have been
described to help the formulation of a pre-test probability
- Should attempt to quantify clinical performance
- Diagnostic tests are most valuable as complementary information to clinical assessment, when
the pretest probability of disease is intermediate
- Doctors should order tests to corroborate or challenge a clinical hypotheses
- Judgment from clinical findings may be a predictor of disease as powerful as any of the tests
available to confirm the diagnosis in question
- The Combination of clinical assessment and ancillary testing, allows multiplication of their LR
with a sharp increase in diagnostic accuracy
- A disease with prevalence of 5%,
- A test with LR of 10
- A refined clinical evaluation of 10
- Each alone will yield a post test probability of 34%
- The combination of the 2 modalities = 10 x 10 = 100, raises the post test probability to 84%
- The powerful sequence of validation of a clinical concept by testing is one of the most important
steps in the diagnostic process
- LR put the value of testing in proper perspective, showing the increment in diagnostic certainty
expectable from a test or from a refined clinical evaluation and the remarkable synergisms from
the combined modalities
- Before ordering a test, one must consider potential benefit and risks, know its performance to
define how the results may help reaching or rejecting diagnosis and whether the result may
alter judgment
- Test should be used to probe the hypotheses and increase confidence, but not to obtain
certainty in diagnosis
Article: Failure Modes and Effects Analysis (FMEA)
- FMEA is a systematic , proactive method for evaluating a process to identify where and how it
might fail and to assess the relative impact of different failures in order to identify the parts of
the process that are most in need of change
- FMEA review:
1) Steps in the process
2) What could go wrong
3) Why would the failure happen
4) What would be the consequences of each failure
- Use to evaluate a new process prior to implementation and existing process
- Steps:
1) Select a process to evaluate with FMEA
a. Do FMEA on sub-processes or variants
b. Consider individual analysis of the medication ordering, dispersing, etc
2) Recruit a multidisciplinary team
a. Include everyone who is involved in any point in the process
3) Have the team meet together to list all of the steps in the process
a. Number every step of the process and be as specific as possible
b. Obtain consensus from the group
4) Have the team list failure modes and causes
a. For each step, list all possible “failure modes”
b. For each failure modes, identify all possible causes
5) Failure effects: What would be the consequences of failure
6) For each failure mode, have the team assign a numeric value (Risk Priority Number RPN) for
likelihood of occurrence, likelihood of detection, and severity
a. Likelihood of occurrence: How likely is that this failure mode will occur?
b. Likelihood of detection: If this failure occurs, how likely is it that the failure will be
detected?
c. Severity: How likely is it that harm will occur?
7) To calculate the Risk Priority number, multiply the 3 scores
a. Identify the failure mode with the top 10 highest RPN
b. Add up all of the individual RPN for each failure mode
8) Use RPN to plan improvement efforts
a. Failure modes with the highest RPN to focus on improvement efforts
Use FMEA For:
1) Use FMEA to plan and reduce harm from failure modes
a. If Failure mode is likely to occur:
i. Evaluate the causes and see if any of them can be eliminated
ii. Consider adding a forcing function: A physical constraint that makes
committing an error impossible
iii. Add a verification step
iv. Modify other processes that contribute to cause
b. If Failure is unlikely to be detected
i. Identify other events that may occur prior to the failure mode that may
serve as “flag”
ii. Add a step to the process that intervenes at the earlier event to prevent
the failure mode
iii. Consider technological alerts
c. If a failure is likely to cause harm:
i. Identify early warning signs that a failure mode has occurred and train
staff to recognize them for early intervention
ii. Provide information and resources at point of care for events that may
require immediate action
iii. Provide Info
3) Use FMEA to evaluate the potential impact of changes under consideration
- Calculate the change in RPN if the change were implemented
- Verbally simulate the change and evaluate its impact in a safe environment
- Some ideas could increase RPN
4) Use FMEA to monitor and track improvement over time
- Set a goal of improvement
www.rapid-diagnostics.org/accuracy.htm
- Accuracy of diagnostic tests
- Confidence intervals can be calculated to reflect the statistical significance of each accuracy
measure
Disease No Disease
Test Result + TP FP
Test Result - FN TN
- Sensitivity = TP / (TP + FN)
- Specificity = TN / (TN + FP)
- Positive Predictive Value = TP / (TP + FP)
- Negative Predictive Value = TN / (TN + FN)
- Positive Diagnostics Likelihood Ratios (DLRs+)
- = D+ / ^D+ = Sensitivity / (1 – Specificity)
- = TP / (TP + FN) / (1 – TN / (TN + FP))
- = TP / (TP + FN) / (FP / (TN + FP)
- Useful tests have large DLRs+
- Negative Diagnostics Likelihood Ratios (DLRs-)
- = D- / ^D- = (1 – Sensitivity) / Specificity
- = (1 – TP / (TP + FN)) / (TN / (TN + FP))
- = (FN / (TP + FN)) / (TN / (TN + FP))
Med Rec
- What med they have before?
- What they will be given?
- One need to constantly know what a patient is taking to know if any contraindication is going to
have
- What reason to provide medicine
- Appropriate doses of medium needed
- Medicine Reconciliation
Article: The Epidemiology of Prescribing Errors: The potential impact of computerized
Prescriber Order Entry (CPOE)
- Adverse drug events (ADEs): Injuries resulting from medical interventions related to the
administration of a drug, are the most common cause of injury to hospitalized patients and are
often preventable
- Computerized prescriber order entry (CPOE): Prescribers write orders online, has been shown to
decrease medication errors by 55% to 80%
- Medication error most commonly occur at the ordering stage
- Resources to identify potential prescribing errors
1) The Pharmacy computer system, with drug allergy, drug interaction checking capabilities
2) Computer access to current lab values
3) Lineal guidelines published in the staff manual
4) Reference texts
5) Internet Access to Clinical Pharmacists Tools
- Causes
1) Medication knowledge deficiency
2) Patient knowledge Deficiency
3) Non-adherence to policy and procedures
4) Slips or memory lapse
5) Nomenclature related errors
6) Transcription errors
7) Calculation unit expression errors
8) Faulty patient identity checking
9) Illegible Handwriting
- 3 likelihoods prevented with CPOE
1) Likely
2) Possibly
3) Unlikely
- Most errors occurred at the time of admission to the hospital
- Most errors unlikely to cause harm
- 1/3 errors involved anti-infective agents
- For the more serious error, incorrect dose was the most common error
- Medication knowledge deficiency: The most frequently cause of clinically significant prescribing
errors
- Benefits of CPOE
1) Create legible orders
2) Decrease the need for transcription
3) Aid the medication ordering process
- CPOE with an advanced level of clinical decision support is needed to prevent many of the
prescribing errors with the greatest potential to lead to patient harm
- Anti infective agents, cardiovascular agents, and opiod analgeses = 57% error
- Nomenclature issues contributed to these errors
- Dosing mistakes are the most common preventable medication error
- A large portion can be attributed to medical knowledge deficiency
- Patients admission, discharge, or transfer are stages prone to error
- Due to poor patient history CPOE to conduct drug histories
- CPOE should use rule based algorithms to mitigate prescribing errors
- CPOE with advanced computer based decision support deliver specific recommendation by
matching individual patient characteristics to computerized knowledge base
- To reduce error:
1) Pharmacist involvement in the multidisciplinary team
2) Prospective data collection: assigning potential medication errors: a CPOE design that will
have the greatest impact on patient outcomes
- CPOE not implemented because:
1) CPOE packages need to be modified or adapted to each hospital
2) The number of clinical rules built into a system needs to be weighed against the increased
time it will take prescribers to write orders
3) Lack of specificity of alerts: Warning too often: Prescribers ignore alerts
4) Order entry process constraints
5) Capacity limitations: The more clinical rules, the slower the speed. Therefore system
limited to 25 rules
- CPOE need to take the patient’s pathophysiologic state and medical conditions into account to
present the physician with warning and recommendations of what to prescribe
- CPOE problem: If CPOE does not allow incomplete or partial order, the prescriber may forget
about the medication entirely
- Not trigger to the other clinicians that the patient needs to be treated with that medication
- CPOE does not obtain accurate medication histories
Article: Clinical Decision Support on Electronic Prescribing: Recommendation and an
Action Plan
- Clinical decision support (CDS): providing clinicians or patients with clinical knowledge and
patient-related information, intelligently filtered and presented at appropriate times can
improve the safety, quality, efficiency, and cost effectiveness of care when applied to electronic
prescribing systems (eRx)
- To realize the full positive impact of CDS:
1) Advances in the capabilities, usability, and customizability of CDS systems
2) New mechanisms to provide access to current knowledge
3) Accelerated implementation of standards and coding systems
4) Appropriate incentives for use
Article: Electronic Prescribing: Toward Maximum Value and Rapid Adoption
- Advances in CDS capabilities: 4 areas:
1) State of knowledge base (The set of rules, content, and workflow opportunities for
intervention
2) Necessary database elements to support CDS
3) Operational features to promote usability and to measure performance
4) Organizational structures to help manage and govern current and new CDS interventions
- CDS Features: (eRx)
1) Reactive alerts and reminders
2) Structured order forms for correct entries
3) Pick lists and patient specific dose checking
4) Proactive guideline support to prevent errors of omission
5) Medication reference information for prescribers and patients
6) Other knowledge driven intervention that can promote safety, education, workflow
improvement, communication, and improved quality of care
- CDS Health care objectives:
1) Reduced medication errors and adverse medical events
2) Improved management of specific acute and chronic conditions
3) Improved personalization of care of individual patients
4) Best clinical practices consistent with available medical evidence
5) Cost effective and appropriate prescription medication use
6) Effective professional and consumer education about medication use
7) Effective communication and collaboration about medications across providers
8) Efficient and convenient clinical practice and self care
9) Better reporting and followup of adverse events
10) Compliance with accreditation and regulatory requirements
11) Improved dissemination of expert knowledge from government and professional bodies to
clinicians and patients
- Barriers impeding the optimal adoption and effectiveness of CDS intervention
1) Functionality: limited CDS feature / function: Usability problem
2) Data: Lack of integration to EHR
3) Knowledge: Uneven availability, standards, and management of best-practice CDS
knowledge
4) Costs: For implementation and ongoing use, as well as liability concerns
- 3 areas necessary to bring CDS to the desired state:
1) Determine and encourage core CDS functionality in all products:
a. Knowledge
b. Database elements
c. Functionality and Usability feature
d. Organizational matters
2) Enhance the knowledge management infrastructure for eRx-related CDS, making it possible
for more providers to have access to references, rules, and guidelines, that are
comprehensive, high quality, usable, actionable, and configurable
- Enhanced standards and vocabularies for CDS related eRX operations
3) Providing incentives, financial, regulatory, and legal, for implementation and use of CDS
enabled eRx
- Core Features to support CDS
1) Knowledge base: the types of rules, content, and interventions that are available in the
system
2) Database
3) Functionality and usability
4) Organizational governance, communication, policy, and management structures and
processes
- Other crucial elements:
1) Standards and terminologies
- Different vendor to exchange data
- Information transfer among providers
- Reconciliation of conflicting prescription standards from different states
2) Structures and Methods for exchanging CDS content
- Concept of knowledge clearing houses
- The publishing of multiple knowledge sets or clearing houses by different agencies, using a
common structure
- Local clinicians select and configure specific interventions for their situation
- Incentives:
1) Protection from increased liability for providers who use suitably strong CDS systems
2) Malpractice benefits for providers who use CDS systems
3) Incentives funding for use of system meeting appropriate certification criteria
Article: Health Information Technology: Fallacies and sober realities
- HIT adoption rates are low
- HIT may not reliably improve care quality or reduce costs
1) HIT not Risk Free
- We cannot get design and deploy complex software systems that are on time, within budget,
meet the specified requirements, satisfy their users, reliable, maintainable, and safe
- They are often opaque to users and system managers to understand how a particular failure
occurred
- Hard to mitigate failure in advance
- Magnifying property: One exchanges a large number of small failures for a small # of large
failures for using HIT vs traditional methods
- Healthcare has been solve to embrace safety critical computing
- HIT software has commonly been identified as being among the least reliable
2) HIT is a Medical device
- Order entry with decision support
- Pharmacy checking and dispensing systems
- Robotic medication delivery and dispensing systems
- Bedside medication management system
- No regulatory effort
- Need a pre market requirement for a rigorous independent safety assessment
3) HIT is NOT a learned intermediary
- That human alone ultimately makes the decision
- People will accept worse solution from an external aid
- People affected by
o Placements (Info availability)
o Font Size
o Info Similarity
o Perceived Credentials
4) Many HC problems are NOT due primarily to human shortcomings
- Bad outcome are the result of the interactions among systems components
- HIT outcomes resulted from interactions of multiple factors
- People, tools, technologies, physical environment, work place, culture, organizational state,
federal policies
5) Don’t equate HIT use with design success
- “Meaningful use” may lead to undesirable consequences if such use is not contextually
grounded
6) Clinical work does not need to be rationalized into something that is neat and linear
- Mismatch between the reality of clinical work and how it is rationalized by HIT
- Systems are disruptive and inefficient
- HIT need to accommodate the non-linearity of HC delivery
7) Most HIT has been designed to meet the needs of people who do not interact with the
primary data
- Mismatch between who benefits and who pays leads to incomplete or inaccurate data entry
- Grudin Law: When those who benefit from a technology are not those who do the work, then
the technology is likely to fail
8) If you provide HIT to clinicians, they may still not use it
- Not always user error
- HIT must support and extend the work of users, not try to replace human intelligent
- Cognitive support that offers clinicians and patients assistance for thinking about and solving
problems related to specific instance of HC
9) One Size Does not Fit all
- HIT should be designed to
o Facilitate the necessary collaboration between Health professionals, patients, and
failures
o Recognize that each member of the team has different mental models and information
needs
o Support both individual and team care needs across multiple diverse care environments
and context
10) Paper continued to be used extensively
- Paper are sophisticated cognitive artifacts that support memory, forecasting and planning,
communication, coordination, and education
11) HC can be understood by people outside of the domain
- People other than clinicians can understand and solve complex HIT issues
- Problems are framed too narrowly
- Need to understand what would help people in their complex work
- What clinicians want and what will actually improve their work may be quite different
- It is health that people desire and health technology utilization is merely the means to achieve it
- The needs of users and complexities of clinical work must be analyzed first followed by the
evaluation of the entire scope of potential solutions
- Appropriate metrics for HIT success should not be adoption or usage, but rather impact on
population health
- Recommendations:
1) Collaborating substantively with those who can contribute unique and important expertise
in cognitive , social, and physical performance and safety to improve safety
2) Need research on how clinical work is actually done and should be done by
a. Cognitive field analysis
i. Cognitive work analysis
ii. Cognitive task analysis
b. Work flow and task analyses
i. Hierarchical task analysis
ii. Sequence diagrams
c. Human centered design evaluation
i. Usability testing
3) Focus meaningful measures of design success
a. Clinician and patient ease of learning
b. Time to find information
c. Time to resolve relevant clinical problems
d. Accuracy of found info
e. Changes in task and information flow, workflow, situation awareness
f. Communication and coordination effectiveness
g. Patient and clinician satisfaction
Absolute Risk Reduction
- = Y placebo – Y treatment
- = 1/3000 – 1/6000
- = 1/6000
# of patients needed to treat
- = 1 / absolute risk reduction
- = 1/ (1/ 6000) = 6000
Absolute risk reduction
- = 5/1000 – 4/1000 = 1/1000
- = 0.001
- = 0.1%
Relative risk reduction
- = (5 – 4) / 5 = 20%
Article: Collective Statistically Illiteracy
- The inability of many physicians, patients, to understand what health statistics mean
- Recommendations:
1) Train medical students in risk literacy with transparent ways to communicate health
statistics efficiently to patients
2) Incentives for complete and transparent reporting of health statistics in journals
- Usually, most report benefits in big numbers (relative risk reductions)
- Harm in small number (absolute risk reduction)
3) Change school curriculum
Article: Fumbled handoffs: One dropped ball after another
- Current systems in HC do not reliably ensure that test results are received and acted upon by
ordering physicians
- System problems:
1) Poor continuity (multiple provider involvement)
2) Lack of communication of test results and other clinical information
3) Several handoffs
- Strategies for improvements
1) Explicit criteria for communication of abnormal results
2) Test tracking systems for ordering providers
3) Use of IT
- The failure of an ordering physician to track and react to test results in a timely manner
- Failure to followup on abnormal diagnostic test results
- Reasons:
1) Large volume of data to be reviewed
2) Test results become available much later – Physicians forget to look for their results
3) Paper based test reporting systems are subject to delivery delay and misfiling
4) Test specialist do not have adequate clinical information as to why a test was ordered and
which results require a phone call
5) No established communication mode
- ED miss medical history and lab results from outpatient caregivers
- A key feature of a fail-safe system is a backup system
- Planned redundancy
Human Error
- Use digital imaging to sort the correct date
- Physician could be overloaded
- Lower work hours lead to more handoffs
- Implement and evaluate a process to improve the timeliness of reporting a critical test results
1) Identify gap in the process
2) Develop failsafe mechanisms of results communication and explicit criteria to identify when
results must be immediately communicated
a. Procedures for reporting panic values
3) Explicit communication strategies and documentation of this communication and clear
escalation strategies should be devised
Automatic notification problems
1) Results dictated as text
2) Result followup by several providers (who should contact?)
3) Issue of Re-Read
- Systems detect changes to evaluation and allow for notification when the report was altered
- A result management tool integrated with EMR, may improve the process of test result review
- Use log books to track tests results
- Hand off problems: Clear lines of responsibility for followup must be established to prevent
misunderstanding
- Involving patients in their care and ensuring they understand what tests are ordered and when
they should receive results
- No more “No news is good news”
Article: Assessing the value of a diagnostic test
- Diagnostic imaging increased a lot because it is a combination of medical culture to:
1) Eliminate uncertainty
2) Scientific and technical advanced leading to new and improved non-invasive tests
3) Substantial barriers to evaluate the value of each test
4) Patient preference
- As the uncertainty in prognosis and diagnosis increases, so does the number of tests ordered
- The more expensive imaging technology is increased at a much faster rate than ultrasound or
radiography
- To determine appropriate use, must know the cost per benefits to the patient and society
- The real benefit is improved outcome for the patient
- A test may be of substantial value depending on what is being diagnosed, the test price, and the
alternative testing
- Patient outcome must be estimated (e.g. with decision analysis) using diagnostic yield and test
accuracy, with expected outcome and cost for each test result
- The Grade of recommendation, assessment, development, and education (GRADE), to evaluate
diagnostic tests
- Stresses the use of outcomes important to patient (Survival, quality of life)
- Stresses the data accuracy (S&S) often provide poor quality evidence for making clinical
decisions
- Recommendation based on the balance of outcomes (positive and negative) as the result of true
and false positive and true and false negative test results and test complications
- Also the quality of evidence
- Uncertainty of data
- Impact on resource use
- For most patients, view that more testing is always better
- Unintended adverse effects of testing: Additional testing from incidental findings
- Tests with benefits are of value to patients if they do not pay the cost
- Fear of malpractice
- The lack of accepted guidelines for evaluation
- Need:
1) More trials to inform clinical practice guidelines
2) An outcome based approach (as opposed to a mechanistic approach)
a. Cost of trials of diagnostic testing strategies is a major barrier
3) Prior authorization to limit test use
4) Use expert judgment or appropriate use of technology
Article: The Wrong Patient
- The process of obtaining informed consent is often deeply flawed
- Involving the patient in the decision making process is not a top priority
- “Organizational Accidents”: They happen to complex, modern organizations, not to individuals
- The error of many individuals converge and interact with system weaknesses
- Environmental factors are not readily changeable
- Latent conditions are system faults that can be remedied and act within individual hospitals
- Environmental factors, e.g. increase sub specialization in medicine, reduce medical staff, reduce
hospital stay
- Latent conditions e.g. Failures of communications, teamwork, and identity verification
- Effective teams:
1) Communicate well
2) Allocate role responsibilities clearly
3) Train to back up team members
4) Monitor member’s performance
5) Resolve conflict efficiently
6) Use well designed protocols and procedures to assure that complex tasks are executed
flawlessly
- A culture of low expectations developed
- Need standardize of protocol to verify patient identity
- A patchwork of home grown Information systems, few interact with each other
- Increase invasive procedures = less documentation in EMR
- How to avoid medical errors:
1) It could happen
2) Open decision
3) Develop and adhere to routine, standardized procedures for verifying patient identity
- Communication and teamwork CRM: Grow resource management
- Acknowledge to adverse effect of fatigue
- Junior members to question decisions of senior members
- Protocol + Communication + Teamwork
Article: Triage of patients with Acute chest paint and possible cardiac Ischemia: The
elusive search for diagnostic perfection
- Determine whether any combination of initial symptoms, signs, lab studies, or ECG findings has
enough information to reduce the likelihood of mis-diagnosing an acute coronary syndrome
- ECG is the most important piece of information
- Description of presenting symptoms
- Adopt clear guidelines for triaging patients
- The effect of fatigue on the cognitive performance of physicians
- Strategies to reduce fatigue to avoid adverse collateral effects
- Triage systems should recommend immediate in person evaluation of all patients with chest
pain
- Rely on the patient’s willingness to be reassured rather than insisting the patient come to ER
and rely on the clean bill of health from ER after the visit
- Modern Approach: emphasizes “system thinking” rather than individual cognitive mistakes
- Create processes and solutions to prevent human errors
o Diagnostic protocol and pathways
o Decision aids
o Novel approaches to staffing
o Other systems changes
- Physicians with higher level of training had a higher sensitivity for detecting MI, but at the
expense of decreased specificity
- Goldman intensive care triage algorithms
- 2 approaches:
1) Incorporating validated predictive instruments into the ECG reading
2) Using validated predictive models to develop and implement local consensus guidelines
- There is no fail safe way to exclude MI at the time of a patient’s initial presentation
- Can use a short period of monitoring and measuring serial biomarker levels
- A range of triage options to match not only the diversity of risks but also the differences in the
level of care and monitoring
- Honesty is a trigger for change
- Mistakes must be opportunities for progress, not punishments
Article: Yield of Diagnostic tests in Evaluating Syncopal Episodes in Older Patients
- Many unnecessary test are obtained to evaluate syncope
- Selecting tests based on history and examinations and prioritizing by expensive and higher yield
test would ensure a more informed and cost effective approach to evaluate patients
- San Francisco Syncope rule (SFSR) was developed to improve prediction of the likelihood of
serious outcome in patients presenting with syncope
- Cost to change ratio
- Cardiac and neurologic tests were commonly obtained in the evaluation of syncope in older
patients despite a minimal effect on diagnosis
- Postural BP recordings had the highest yield but were performed in only about 1/3 of admission
- Application of SFSR improved yields and lowered costs without compromises
- Lowest likelihood of useful test results : Highest cost per yield: was incurred by EEG, CT, and
cardiac enzymes tests
- Unhelpful test continue to be performed despite evidence
- Easy availability of lower risk testing = overuse of resources
- Unnecessary testing is a substantial contribution to high care costs
- SFSR criteria helpful to identify patients likely to benefit from cardiac testing
- Instituting evidence based diagnostic guidelines
- Base tests in the results of initial history and examinations and prioritizing higher yield test
Article: Computerization can create safety hazards: A bar coding near miss
- Bar code technology guarantees only that the information recorded on the wristband is
transmitted to the computer faithfully
- It does nothing to ensure that the information on the wrist band is correct in the first place
- 2 kinds of error:
1) Errors at registration time
2) Placing a wristband on the wrong patient
- Redundancy is the best defense
- Ask “What is your name?”
- Use 2 identifiers rule
- Barcode read what they can see
- Use RFID to minimize patient disturbance and eliminate the extra time
- Talking labels
- Concerns about interference between wireless monitors and the radio signals from these RFID
chips
- Implant RFID device
- Privacy concerns
- Use Fingerprints technology
- System based Redundancy
- Redundancy must be balanced against eh increment safety they provide
- Patient misidentification is an endemic problem
- Mis Entry Error
- The physician who knows his patient will remain an important defense against medical error
- Physicians check new results for consistently against what they already knew about the patient
- Bayesian Reasoning
- Hospital managers accept new system even when they have their doubts
- New system increases their staff load and takes more time away from the patient care
- Barcode too slow to respond in ER
- System refused to accept medication because the dosing deadline had passed
- Problem of tight computer control over complicated medical processes
- Fail to consider all of the realities of clinical care setting
- Use CPOE: Computerizing the prescribing process
- CPOE improves care process
- But no evidence it improves patient outcome
- CPOE Advantages
1) Institutional efficiency and communication
2) Speedier order completion and treatment delivery
3) Opportunities to inform physician about the benefit, changes and costs of orders
- Simple human processes and innovation provide large opportunities for improvements
- When harmonized with robust technological solutions
- Physicians who knows his patient well is the best defense against system errors
Article: The Cognitive Psychology of Missed Diagnoses
- Cognitive Psychology: The science that examines how people reason, formulate judgments, and
make decisions
- Mis Diagnosis can occur and can be connected with followup
- Scantly PE recorded
- Difficult to learn from mistakes
- Use Bayesian reasoning to learn from mistakes by numerically expressing uncertainty as a
probability for each decision point
- Recognition of misleading intuitions
- Studying cognitive psychology provide aware ness that can help avoid predictive pitfalls
- Heuristics = Short cuts in reasoning
- Availability heuristics: Judge likelihood by the ease with which examples spring to mind
- Anchoring Heuristics: Stick with initial impression once they are solidify informed
o Need to check for disconfirming evidence
- Framing effect: Decision made on how the question is formed: e.g. 10% of dying vs 90% of
surviving
o Skills required to collect clinical findings and frame them correctly
- Overreliance on diagnostic technology results (and under appreciation of technology’s
limitation) = Blind obedience of technology
- Anchoring Bias: Premature closure
o Reluctance to pursue alternative possibilities once a commitment is made
o When 1 alternative is available , check
o When many alternatives are available: do nothing
- Use followup to overcome cognitive fallibilities
o To reconsider the entire picture from an alternative perspective
- Need Appropriate Timing for followup
- Add safeguards to minimize reflexive decisions
- Scientific evidence would eliminate the uncertainty that leads to missed diagnoses
Article: Randomized comparison of Guaiac-Based vs Immunochemical FOBT
- Imm FOBT = Better sensitivity but lower specificity for detecting cancer led to similar positive
predictive values vs Guaiac-Based FOBT
- Wider use of Imm FOBT would result in more colonoscopies
- Lead to detection of more cancer
Article: Prescription Errors and outcomes related to inconsistent information
transmitted through computerized order entry
- Order Entry errors related to inconsistent information within the same prescription
- Clarifications or non standard specifications may still be entered through a free text comment
field in CPOE
- Allow flexibility for providers to enter clarifications regarding timing, dosage, or route of
administration
- Led to discrepancies between the elements selected through the structured template and the
provider’s free text comments
- Free text window should be maintained in CPOE but with improvements to avoid inconsistencies
1) Providers select an automated dosage default rather than typing the desired dosage
2) Comments were transferred to the new prescription when modifying an existing
prescription
3) Insufficient knowledge of ordering mechanism
a. Simple training on the use of complex orders for certain prescription to reduce
errors
4) Standardizations through CPOE was not adequately integrated with the workflow, needs
and preferences of the providers
- Need to improve the usability of the CPOE interface and integrating it with workflow
- Identify breakdown points
- Maximize the effectiveness, efficiency, and satisfaction of electronic communication through
CPOE
Article: Health Care Technology: A Cloud Around the Silver Lining?
- CPOE and EHR will lead to safer and more effective care and decision making by providing
guidance and critical information in these complex situations
- Free text contradicts the selected medication order details
- Use free text because of:
1) Lack of CPOE system flexibility
2) Inappropriate defaults
3) Inadequate options
- Use Free Text but:
1) Lack of Physician knowledge of the correct way to enter text orders
2) Impatience during the ordering process
3) Lack of checked guidance by the system
- Negative unintended consequences related to CPOE:
1) Additional work for clinicians
2) Unfavorable workflow change
3) Never ending demand for system changes
4) Problems related to persistence of paper records
5) Changes in communication practices with false assumption
6) Negative emotions generated from changing established practices
7) Generation of new types of errors (“e-iatrogenesis”)
8) Loss of ordering autonomy to accommodate CPOE goals and system limitations
9) Over dependence on the new technology
- Other types of CPOE Errors:
1) Misapplication of bolus and drip orders
2) Ambiguous orders for complex or changing dosing regimes
3) Automating inappropriate dosing intervals
4) Inadvertent ordering of a juxtaposed order by a single click
5) Inflexibility to document administration of a subsequently cancelled medication order
6) Inability to select the correct location or encounter for a patient : Missed orders
7) Failure to provide clear views of preexisting medications when entering new drug orders
8) Inappropriate thresholds for electronic alerts : Most alerts ignored
- For EHR
1) Documentation to record a patient’s current clinical state and related decision making: Lost
in the move to electronic
2) Cut and pasted patient’s story and test result lead to incorrect decisions
3) Independent histories and exams may be missed
4) “Copy Forward”: Overly long notes: Inability to quickly page through the medical chart for
critical information
- Decision Support system to:
1) Anticipate and facilitate appropriate next steps
2) Require application of safety logic to the patient medical chart
3) Identification and tracking of new and old clinical loose ends to assure appropriate followup
4) Suggest next steps based on newly available results
5) Adverse trend recognition
6) Identification of overdue orders
7) Tracking of missed procedures
8) Non intrusive automated surveillance of loose ends + Anticipatory guidance
Dr Ash Presentation:
- Patient Centered Medical Homes:
1) Patient Oriented Care
2) Whole Person orientation
3) Physician directed team care
4) Coordinated and integrated care
5) Quality and safety
6) Enhanced access
7) Self management support
8) Payment reform
9) Information management enabled
- How doctors get paid
1) Procedures: CPT-4
2) Visits: E&M Codes
3) Pay for Usage: eRx, MIPPA
4) Pay for performance: PQRI
- Natural Language Processing (NLP)
- A subfield of AI and computational linguistics. It studies the problems of automated generation
and understanding of natural human languages
1) Movement to electronic notes, containing narrative text
2) Financial reimbursement increasingly tied to documentation
a. Reimbursement tied to CMS algorithm
b. Facility fees increasing based on diagnoses
3) Growing interest in measuring quality
4) Research
- Computer Assisted Coding
- Computerized tools to automate the assignment of codes to clinical documents that are
traditionally assigned by coding of HIM professional
- A set of tools to process narrative text documents to tag phrases with SNOMED CT codes, then
apply algorithms to assign E&M code supported by the document
- Handles qualifiers, negation, and applies rules to increase tagging precision
-
Index# of patients needed to treat, 51 6 recommendations for the government, 6 Absolute Risk Reduction, 2, 51 Access Medicine, 26 ADEs, 45 Advances in CDS capabilities, 47 Adverse drug events, 45 Agenda, 1, 3 Algorithms, 16 Anchoring Bias, 57 Anchoring Heuristics, 57 Annual Incidence, 29 ARRA, 19 askMEDLINE, 1, 28, 29 Availability heuristics, 57 Bar code, 56 Barriers impeding the optimal adoption and
effectiveness of CDS intervention, 47 Bayes, 21 Bayes Theorems, 29 Bayesian, 14 Benefits can include, 7 Best Evidence, 27, 28 Better IT systems, 7 Biopsies, 22 Blind obedience of technology, 57 CCTR, 27 CDS, 46, 47 CDS Features, 47 CDS Health care objectives, 47 CDSR, 27 Centor, 36 certified, 20 Chief Symptom, 3 Clinical decision support, 46 Clinical Evidence, 26, 28 Clinical Prediction Rules, 36 Clinical queries, 28 clue does not fit, 17 Clusters, 14 Cochrane Controlled Trials Registry, 27 Cochrane Database of Systematic Reviews, 27 Cochrane Library, 26, 27, 28 Cognitive Psychology, 57 Collection of patient data, 9
Color Doppler, 22 Complete Interview, 3 Computed Tomography, 22 Computerized prescriber order entry, 45 Computers, 16 Core Features to support CDS, 48 CPOE, 45, 58 CPOE Advantages, 56 CPOE Errors, 59 CPOE not implemented because, 46 CPR, 36 CPRs, 36 Criteria for a worthwhile test, 34 CT, 22 DARE, 27 Data Paths, 16 Data processing, 11 Data Resolution skills, 37 Database of Reviews of Effectiveness, 27 Decision Support system, 59 Decision Trees, 16 Decisive clues, 17 Depth of study, 38 Diagnosis at a Glance, 39 Diagnosis Ways:, 14 Diagnostic certainty, 39 differential diagnosis, 15 Digital Subtraction Angiography, 22 Dissection of Symptoms, 8 Early Hypothesis Generation, 14 Environmental factors, 54 False clues, 17 false negative, 29 false positive, 30 Final Error, 33 Flowcharts, 16 FMEA, 42 FN, 29 FP, 30 Framing effect, 57 Goldman intensive care triage algorithms, 55 Google Scholar, 1, 24, 25 GRADE, 53 Grudin Law, 49 Guaiac, 57
HC organizations should, 7 HC Persistent Problem, 4 Heuristics, 57 HITECH, 20 ideal clue, 18 Immunochemical FOBT, 57 Incidence, 29 Incremental gain, 30 Info Retriever, 26 Initial Problem list, 12 Intralab Error, 33 IOM vision for 21th century HC, 4 IOM: HC quality, 4 Key clues, 17 Latent conditions, 54 Likelihood of detection, 43 Likelihood of occurrence, 43 Likelihood ratio, 39, 40, 41 LR, 39 LR for a negative test, 40 LR for a positive test, 40 Magnetic Resonance Angiograph, 22 Magnetic Resonance Imaging, 22 marginal benefit, 31 McIassc, 36 MDConsult, 26 meaningful use, 20 Med Rec, 44 Medical Subject Heading, 27 MEDLINE, 25, 26, 27, 28, 29 MeSH, 25, 27, 28, 29 meta cognition, 21 MRA, 22 MRI, 22 Negative clues, 17 Negative Diagnostics Likelihood Ratios, 44 Negative Post test Probability, 41 negative post tests odds, 41 Negative Predictive Value, 30, 44 normogram, 40 Nuclear imaging, 22 Odds, 40 ONCHIT, 20 operating characteristics of a clue, 30 OVID, 28 P(^C/^D), 30 P(^C/D), 29 P(^D/^C), 30
P(C/^D), 30 P(C/D), 29 P(D), 29 P(D/C), 30 Pattern recognition, 14 pay for performance, 8 PE, 11 PICO, 25, 28 pivotal clue, 15 POEM, 29 Positive Clues, 17 Positive Diagnostics Likelihood Ratios, 44 positive post test odds, 40 Positive Post Test Probability, 41 Positive Predictive Value, 30, 44 Positivity in disease, 29 Positron Emission Tomography, 23 Post Test odds, 40 Post test probability, 41 Prelab Error, 33 Pretest Odds, 40, 41 Prevalence, 29 Primary clues, 17 Principles for Evolutionary Change, 5 Principles for radical change, 5 Probability, 40 Problem list, 11 Problem List, 11 PubMed, 25 PubMED, 29 PV-, 30 PV+, 30 Questions asked by ER, 8 RCRI, 36 Reason for errors, 33 Reasons for doing tests, 32 Reasons for overuse, 34 Relative Importance of a clue, 18 Relative risk reduction, 51 Revise Geneva Score, 35 Revised Cardiac Risk Index, 36 Revised Geneva Score, 36 RFID, 56 Risk Priority number, 43 Risk Priority Number, 43 RPN, 43 San Francisco Syncope rule, 55 Scientific American Medicine, 26
Scirus, 25 Secondary clues, 17 sensitivity, 29 Sensitivity, 17, 36, 44 Severity, 43 SFSR, 55 Short Cuts, 39 SnNout, 36 Solve by reverse, 38 specificity, 30 Specificity, 18, 36, 44 SpPin, 36 STAT!Ref, 26 Students must acquire many skills, 13 syncope, 55 Syndromes, 14 Tactics with clues, 15 Technology’s Evidence Based Medicine
Reviews, 28
Test efficiency, 30 Test to be perfect, 33 The Key Clue, 15 Theorem of Bayes, 21 TN, 30 To cross the HC IT chasm, 5 TRIP, 26 true positive, 29 truly negative, 30 Ultrasound, 22 UptoDate, 27 UpToDate, 26 Use FMEA For, 43 Use free text because, 58 Use Free Text but, 58 Vision of patient centered support cognitive
support, 6 VSG-CRI, 36 Written Record:, 24