Professional Documents
Culture Documents
➲ Retrospective study
Information bias
Misclassification bias
➲ Uses uniform criteria for normal WBCs in all
age groups as > 10 cells/mcL
➲ 1:500 rule used in study, but has been
discredited by several studies
➲ Could be Herpes meningoencephalitis or Lyme
disease
➲ Both missed patients were young infants – the
patient sub-group we are most interested in..!
➲ Not useful if patient was pre-treated with Abx
EBM – Clinical Prediction Rules
or
Attempts to quantify Intuition..!
Jegen Kandasamy
PGY-2
Maimonides Medical Centre
The Problem
➲ CSF Analysis
Protein/Glucose: 70/56
CSF red, turbid, 100 WBCs and 7500 RBCs per mcL
85% N, 15% L
No cells/organisms seen on gram stain
The Costs
Vs
????????
Clinical Dilemmas
➲ How likely is a disease ?
Which patient is most likely to have an ankle fracture..?
Which patient is most likely to have Strep Pharyngitis..?
Which patient is most likely to have a head injury that warrants CT..?
Which patient most likely has conjunctivitis that is bacterial..?
➲ Ambiguity
The language that physicians use to describe a patient’s
condition often is ambiguous
Probable, highly likely, etc.
➲ Probability
An alternative method of expressing uncertainty
% or odds
Probability estimates of various qualitative verbal
expressions
Certain
Likely
Possible
Probable
Low probability
Suggests
High probability
Unlikely
Moderate probability
Pathognomonic
Classic
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
The most important reason we order
diagnostic tests is to Refine Probability
In the words of Kant...
+ Experience
Priori Test result
Pretest Posttest
Probability Probability
Posteriori
The effect of test results on the
probability of disease
Pretest Posttest
Probability Probability
Perform Test
Probability of Disease
Pretest Probability Estimates
➲ Traditional
Clinical experience / judgment
➲ Objective:
Prevalence studies
Subgroup analyses
Clinical prediction rules
What is a CPR
➲ A tool that takes the experience of a group of
physicians with hundreds or even thousands of
patients and attempts to distill this experience
into a simple rule
➲ Quantifies contributions of
History
Physical Exam
Basic laboratory tests
Step 1. Derivation
Identification of factors with
predictive power
Step 2. Validation
Evidence of reproducible accuracy
Narrow Validation Broad Validation
Application of a rule Application of a rule
similar clinical setting in multiple clinical
setting and population settings with varying
Step 1 prevalence of disease
Step 3. Impact
Analysis
Evidence that rule changes
physician behavior and
improves patient outcomes
CPRs Make Pretest Probability
Estimates More Reliable
➲ Often useful in situations with
Complex decision making
High clinical stakes
Opportunity for cost savings w/o compromising
patient care
Methodologic standards -
Derivation
Methodologic Standards -
Derivation
➲ Spectrum
Accuracy is maintained in patient sample that,
on average, has more or less advanced disease
or different disease trajectory
Study Impact of Rule
➲ Time to use prediction rule must be “worth
it” and change clinician management or
behavior to
Improve patient care
Reduce costs with same quality of care
➲ Model is valuable if it divides patients into
meaningful groups
➲ Parsimony, not cumbersome
Summary
➲ Validated prediction rules should be used
when
Clinical decision is complex
Clinical stakes are high
➲ Understand
Patient population
Clinical setting
Purpose of the rule
➲ Look for rules that make a difference