EBM Em Flashcards
sensitivity and specificity judge the _______ precision of a test
sensitivity and specificity judge the TECHNICAL precision of a test
sensitivity and specificity are ______ to changes in prevalence
sensitivity and specificity are INSENSITIVE to changes in prevalence
PPV and NPV judge the _______ of a test
PPV and NPV judge the CLINICAL PERFORMANCE of a test
PPV and NPV are ______ to changes in prevalence
PPV and NPV are SENSITIVE to changes in prevalence
sensitivity eqn
TP/(TP+FN)
specificity eqn
TN/(TN+FP)
sensitivity
% pts w/ disease who have + test
specificity
% pts w/o disease who have - test
positive likelihood ratio
compares the likelihood that someone with the disease in question has a positive test as compared with someone who doesn’t have the disease
negative likelihood ratio
compares the likelihood that someone without the disease in question will have a negative test as compared with someone who does have the disease
positive likelihood ratio eqn
sensitivity / (1-specificity)
negative likelihood ratio eqn
(1-sensitivity) / specificity
In practice, likelihood ratios are used in two ways:
- To calculate post-test odds of a disease, given pre-test odds and the LR
- To give a general interpretation of a test’s quality
likelihood ratio >10
good test to rule-in disease w/ positive result
likelihood ratio <0.1
good test to rule-out disease w/ a negative result
allocation concealment vs. blinding (masking)
Allocation concealment occurs DURING the process of SELECTING patients for a study and involves the person enrolling patients.
Blinding occurs DURING the CONDUCT OF THE STUDY ITSELF and usually involves the patient, the outcome assessor, and sometimes the investigator performing the intervention.
P-value of .05
5% risk that the difference is due to chance, or a 95% likelihood that the difference we see represents a real difference.
P-value
probability that difference between two averages, rates, etc. is due to chance
number needed to treat
how many people we need to treat instead of not treat, or the number that need to be treated with one therapy instead of another, for one ADDITIONAL person to benefit
calculated based only on results that are presented as rates (%’s)
NNT eqn
NNT=100/(% tx group - % cntrl group)
Relative risk
helps us understand the difference between two rates
- risk of harm with one drug as compared with another.
- it also can be the risk of benefit with one drug as compared with another.
relative risk = 1
no difference in risk of harm or benefit
relative risk = 1.3
likelihood of something happening is 30% higher in one group
relative risk = .7
likelihood of something happening is 30% lower in one group vs. another
OR > 1
the control is better than the intervention
OR < 1
the intervention is better than the control
OR = 1
outcome is the same in both groups
confidence interval
level of uncertainty around the measure of effects
if crosses 1, implies no difference between arms of the study
power
ability to find a difference between the groups if a difference truly exists
depends on the number of patients in the study but also the variability in the data and the magnitude of the effect
*only worry about if no difference between treatments
hierarchy of evidence
- controlled trials
- case-control studies
- case series
- expert consensus or opinion
- pathophysiologic reasoning
summary reviews
- traditional type
- cover full breadth of topic
- useful for background Qs
synthesis reviews
“systematic reviews”
- define 1 or 2 specific questions
- useful for foreground Qs
Statistical heterogeneity
occurs when the difference among study results is greater than predicted by chance alone
evidence of heterogeneity
- chi squared test (p=value <0.5)
2. degree of inconsistency (I^2)
95% confidence interval
we can be fairly (95%) certain that the true value will fall within this range
positive predictive value
probability that an individual who tests positive actually has the disease
(a/a+b)
negative predictive value
probability that an individual who tests negative actually does not have the disease
(d/d+c)
team-based learning approach main goal
use and apply course concepts
concealed allocation
preventing the ENROLLING investigator from knowing what tx the pt will recieve
blinding
preventing the TREATING investigator from knowing what tx the pt will recieve
preventing pts from knowing which tx they are receiving
intention-to-treat analysis
patients analyzed in the group to which they were initially assigned regardless of whether they actually received the treatment
degree of effect is reflected by a low NNT or a low P-value?
NNT
Publication bias
Likelihood that negative results, i.e., studies that do not show a difference or benefit of a treatment, are less likely to be published.
Since publication bias favors publication of research showing a benefit, a meta-analysis combining on published studies could inflate the real benefit of an intervention.
funnel plot
Compares the effect size in different studies with some measure of the variability of the data from each study.
A “funnel” formed by the data that is balanced on both sides of the mean shows there was no publication bias
basic criteria
Is the test sufficiently sensitive and specific? We have many tests with low sensitivity and specificity.
Minimally useful
The test changes diagnosis. As a result of a positive test, we now have a label to put on a patient. That doesn’t mean that we’ve done anything other than categorize their set of signs and symptoms.
More useful
The test changes treatment.
A test that results in changes in treatment is a good start. However, tests don’t always lead to changes.
Very useful
The test changes outcomes. Even tests that change diagnosis and change treatment decisions may not benefit patients.
Maximum benefit
The test is worthwhile to patients and/or society.
Exp. Screening newborns for congenital hypothyroidism, phenylketonuria, and other diseases (but not all) results in early treatment that permanently benefits affected children, which not only benefits them but benefits society.
usefulness of information
(relevance*validity)/work
Bayes’ theorem
pre-test probability
based on prevalence
pretest probability increases as we move from ______ through ______
pretest probability increases as we move from SCREENING through PROGNOSIS
testing leading to diagnosis is only helpful if
it leads to a change in treatment
