EBM Day 4 Flashcards
Variance
Average of squared differences from the mean
SD
square root variance
SE +meaning
SD/sqrtN
The SD of the distribution of mean value
Central Limit Therom
When normal diet
More samples you take, more closer you get true mean
CI calc + meaning
mean+/-Z(usually 1.96)*SE
this is where we think sample mean lies
Point estimate
estimate of some value derived from study (mean, regression coeff etc)
Z=+1.96
Z=-1.96
what %
both in 97.5% or 2.5% chance of not being in
total = 5%
When to compare CI to 0 (2)or 1 (3)
0=correlation or dif
1-hazard ratio, relative risk, odds ratio (these all kinda same thing too)
ESSENTIALLY WHAT THE NULL IS
Smaller CI
Higher precision and more n
Marginally sig
A little bias either way can make sig or not sig
2x2 false pos and negative chart
draw
Misclassifcation
False positives and false negatives
TN, FN, TP, FP chart and sensitivity
draw
+PV and -PV and meaning
Predictive value
TP/TP+FP
TN/FN+TN
Probability that a patient with positive test result truly has disease
Probably that a patient with negative result does not have disease
Sensitivity and meaning and when best
TP/TP+FN
No false negatives
Negative result=rule out
SNOUT
Sennstive, negative rules out disease
GOOD FOR cheap and initial procedures
Specificity and meaning and when best
tN/FP+TN
No false positives
Positives-rule in
SPIN
Specific, positive rules in
GOOD FOR EXPENSIVE and dangerous PROCEDURES
Draw cut off point table with TP, TN, FP, FN
draw
Prevalence
TP+FN/TOTAL
LR+ and meaning
Likelihood ratio
TP/TP+FN/
FP/FP+TN
how likely to have disease when test is positive
how many times more likely a (+) test result is to be found in people with disease compared to those without
LR- and meaning
FN/TP+FN/
TN/TN+FP
how likely to have disease when test is negative
how many times more likely a (-) test result is to be found in people with disease compared to those without? i think
TP=
FP=
FN=
TN=
SE
1-SP (false positive rate)
1-SE
SP
1-SP
SEN
false postive rate
True positive rate
ROC Curve
Senstivity (TP) on Y
1-Specifity (FP rate) on X
Upper left is where you wanna be
Relationship between predictive value and prevalence
Lower prevalence, lower predictive value
Post test probability
Probability of a disease after test result is known
Natural Frequencies Approach
Choose large population
Use sensitivity to find TP
Use 1-spec to find FP
TP/TP+FP=+PV
Increase pretest probability
consider demograpics, presentation, clinical setting
Nomogram
use Pretest prob (prevalance) and then find LR and draw line to Post test prob (PV)
LR to change in disease prob % 10 5 2 .5 .1
45% 30% 15% -30% -45%
Parallel testing and sens/spec relationship
Test A or B, or C positive makes all positive
Sens up, spec down
Serial testing and sens/spec relationship
Test A AND B AND C postiv
Send down, Spec up
Independent tests
as related to serial/parallel testing
Assumed by parallel and serial testing
Index Test vs reference
new test when compared to gold standard
Basic approach to evaluating diagnostic tests
select patients test all patients with index diagnose in series then do gold standard Compare results of index test to reference index results Look for sources of error
5 challenges to establishing accuracy in diagnostic test
inappropraite ref standard spectrum bias verification bias observor bias chance
Spectrum bias and big problem
including only highly selected patients (such as those who would benefit)
- problem of including people who would benefit a lot, and very little and leaving out middle (where there would be a lot of false positives and false negatives - big change to specificity and sensitivity
verification bias
work up bias
not giving both tests
reference may be too invasive to give if respond to index
Observor bias (5 reasons why occur)
Ref standard is subjective
Index and reference standard not evaluated independently
Evaluators not blinded to results of other test
Evaluators not blidned to patient characteristics
Must design index test without out this value
Reference standard-leads to bias if don’t do this