Evidence 1 & 2 Flashcards
likelihood ratios
probability that you’d see some evidence if your hypothesis were true/probability that you’d see the exact same evidence if your hypothesis were false
A number representing the diagnostic usefulness of a test
likelihood ratio
LR = 1
is useless
doesn’t help you differentiate
LR > 1
increases probability (the higher the better)
LR < 1
decreases probability (the lower the better)
LR +
means the finding was present, not that it necessarily increases the probability
LR -
means that the finding was absent, not that it’s a negative number and not necessarily that it decreases the probability
Does an LR- decrease the probability?
NO
beware of double-counting evidence
can inflate LRs or deflate
when choosing between multiple LR options, which one do you choose?
the one that provides the best evidence
in patients who have the disease, the probability that the test will be positive
sensitivity
in patients who don’t have the disease, the
the probability that your test will be negative
specificity
the probability that you would see certain evidence if your hypothesis were true
sensitivity
the complement of the probability that you would see the same evidence if your hypothesis were false
specificity
Calculating LR
sensitivity/specificity
An LR- (finding is absent) can be calculated similarly:
1 – sensitivity/specificity
is the change of the application of LR linear?
No, an LR of 10 is better than an LR of 5, but it doesn’t increase the probability by twice as much
the more extreme your initial probability (the closer to 0 or 100), the _______ it should be to change your mind
harder
pre-test probability can be based on the ________ of a disease in a population
prevalence
Is it reasonable to assume that there’s a low probability that your patient has a rare disease?
yes
50% certainty is as uncertain as you can get; any lower and you’re more certain the condition is…
not present
no amount of evidence can take you to which percent?
0 or 100
which of the following would have the greatest increase in probability?
a. an LR of 5 starting with a probability of 5%
b. an LR of 5 starting with a probability of 50%
c. an LR of 5 starting with a probability of 95%
B
the more extreme the initial probability, the less it will change in light of evidence
why is it useful clinically to be familiar with LR’s
clinician is able to prioritize information-gathering and help clinicians behave more expertly
spPIN/snNOUT
a specific test, if positive, helps rule the condition IN, and a sensitive test, if negative, helps rule the condition OUT
it takes more/better evidence to go from, 1%-5% than from 51% to 55% certainty. Why?
1% is an extreme
when you are around 50% then it should be easily swayed one way or the other
- 50% is the most uncertain you can get
how can you roughly calculate the post-test probability for low pretest and moderate LRs?
you can multiply the probability by the LR and you will get a reasonably good estimate
example: a pre-test probability of 2% with an LR of 5 = post-test probability of about 10%
the probability that disease is present given that a test was positive
positive predictive value
the probability that disease is absent given that a test was negative
negative predictive value
what is the assumption when using PPV and NPVs
your patient’s pretest probability is the same as those in the study that determined the PPV/NPV
It’s almost as if it has already done the LR calculation for you but assumed a pretest probability
I.e. simpler but not as ACCURATE (depending on pretest probability) as applying LRs
what should you think of when you hear “pathognomonic finding”
HIGH LR
- pathos “disease“, gnomon “indicator”
Findings that, if present, strongly increase the probability of a condition
is a pathognomonic finding particularly sensitive or highly specific?
Not necessarily particularly sensitive but highly specific
what should you think of when you hear “Sine qua non-findings”
LOW LR - CLOSE TO 0
“without which it could not be”
Findings that, if absent, strongly decrease the probability of a condition
Evidence: The Process In Practice
- Recognize the need for more evidence
- Choose a test (question, physical exam, lab test)
- Perform the test correctly
- For labs, this may be the responsibility of the technician. For physical exams, we have the Clinical Skills Practicum - Interpret the results correctly
- Repeat steps 1-4 until you cross a threshold (see thresholds lecture)
Evidence: Errors In Practice
- Fail to recognize the need for more evidence (premature closure)
- Poorly chosen or missed tests (question, physical exam, lab, etc.)
- Incorrectly performed test (esp. physical exam)
- Incorrect interpretation (lack of knowledge, bias)
premature closure
fail to recognize the need for more evidence
what are examples of incorrect interpretation?
lack of knowledge, bias
as someone goes from novice to expert, the amount of data gathered goes up, but their diagnostic accuracy improves.. why?
The quality of the data collected
an expert gathers more information - data is more information RICH
ask questions where the results provide MORE EVIDENCE - the questions are associated with higher LRs
