Review Flashcards
CER (Control Event Rate)
of people that had the event/# of people in the control group
EER (Experimental Event Rate
of people that had the event/# of people in the experimental group
ARR (Absolute Risk Reduction)
CER - EER (control event rate - experimental event rate)
if ARR = 0.074, control group = coumadin, experimental group = ceprotex, then you can interpret it as:
“ If 100 people were treated with Ceprotex and 100 people were treated with coumadin, 7.4 fewer people in the Ceprotex group would suffer a stroke over a 5 year period”
RRR (Relative Risk Reduction)
RRR = (CER - EER)/CER
if RRR = 0.50, control group = coumadin, experimental group = ceprotex,
then you can interpret it as:
“Compared to coumadin, Ceprotex is associated with a 50% relative risk reduction in the 5-year incidence of stroke.”
NNT (Number Needed to Treat)
NNT = 1/ARR = 1/(CER - EER)
If NNT = 1/0.074 = 13.5:
“Approximately 14 people need to be treated with Ceprotex compared to coumadin to prevent one stroke”
ARI (Absolute Risk Increase)
EER - CER (experimental event rate - control event rate)
NNH (Number Needed to Harm)
NNH = 1/ARI
if we calculate NNH = 1/0.015 = 67:
“If 67 people are treated with Ceprotex instead of coumadin, 1 extra person will suffer gastrointestinal bleeding”
deductive reasoning
○ hypothesis testing
○ comparing current case against a pattern
○ does the patient have pulmonary embolism?
inductive reasoning
○ differential diagnosis
○ find a pattern
○ what does the patient have?
When assessing a new diagnostic test:
○ use new test among the patient who would receive test normally in clinical practice
○ compare to appropriate reproducible gold standard, which should also be performed on all patients that would normally receive the test in clinical practice
2x2 table
Disease + Disease -
Test + true positive false positive
Test - false negative true negative
Sensitivity
true positive rate
the number of people that test positive and have the disease out of all the people that have the disease
You know patient has the disease. What is the chance that the patient will test positive?
A/(A+C)
Specificity
true negative rate
the number of people that test negative and do not have the disease out of all the people that do not have the disease
You know the patient does not have the disease. What is the chance that the patient will test negative?
D/(D+B)
positive predictive value (PPV)
The number of people that actually have the disease among the people that test positive for the disease
You know the patient has tested positive. What is the chance that the patient has the disease?
A/(A+B)
negative predictive value (NPV)
the number of people that don’t have the disease among the people that test negative for the disease
You know the patient has tested negative. What is the chance that the patient does not have the disease?
D/(C+D)
Effect of Prevalence
○ no effect on sensitivity or specificity
○ higher prevalence: increases PPV and decreases NPV
○ lower prevalence: decreases PPV and increases NPV
likelihood ratio
people with a given result among people with disease / people the the same result among people without the disease
likelihood ratio (positive)
sensitivity / (1 - specificity)
A/A+C / B/B+D
likelihood ratio (negative)
(1 - sensitivity) / specificity
C/A+C / D/B+D
Comparing likelihood ratios and predictive values
○ both look at what it means to the patient to receive a given test
○ predictive values depends on the prevalence
○ likelihood ratios are derived from sensitivity and specificity which are not affected by prevalence
receiver operator curve
○ can be used to determine the optimal cutoff for the distinction between a positive and a negative test result
○ plot of sensitivity against 1 - specificity or true positive rate versus false positive rate
○ built by plotting the sensitivity and 1-specificity values that were calculated when using varied cutoffs
○ slope of a tangent to ROC at any given point is the ratio of sensitivity/(1-specificity) or LR
○ goal: optimize sensitivity and specificity
○ optimal cutoff is the one closest to the upper left hand corner of the graph
Bayes Theorem: underlying concept
it is a method for evaluating new information in conjunction with prior information
most helpful when pretest probability is intermediate
Bayes Theorem: general form
Prior odds hypothesis x Bayes Factor = Final (Posterior) odds of hypothesis
Bayes Theorem: form for diagnostic tests
Pre-test odds of disease x likelihood ratio = Post-test odds of disease
Odds
if p = 0.5
if p = 0.33
Odds = probability/(1 – probability)
then odds = 0.5/ (1-0.5) = 1:1
then odds = 0.33/ (1-0.33) = 1:2
Probability
if odds = 3:1
if odds = 1:4
Probability = odds in favor/total odds = odds in favor/(odds in favor + odds against)
then p = 3/ (3+1) = 0.75
then p = 1/(1+4) = 0.2
Pretest odds of disease is determined by
clinical risk factors for the disease, “subjective impression”
The pre-test odds are multiplied by _ to get post test odds
likelihood ratio
if pretest probability = .33
then pretest odds = _ (probability = _)
if LR = 10 then posttest odds = _
covert posttest odds to probability: _
1:2, 0.5
0.5 x 10 = 5 (5:1)
5/(5+1) = .8
population vs sample
population: all the members of a particular group
sample: use a small subset of individuals to draw conclusions about the larger population
● relevant because may not be feasible to measure entire population
● infer information about population from sample results
● random selection to best reflect composition of population
mean
sum(x) / n
median
middle number when ordered from smallest to largest (or vice versa)
mode
most often represented number
standard deviation
square root of: sum(x - mean)^2 / (n-1)
square root of variance
variance
find distance between each value and mean
add of the squares of the distances (to prevent canceling of positives by negatives)
divide the sum of squares by the number of values
range
largest value - smallest value
interquartile range
values corresponding to the 25th and 75th percentile
frequency
how often a value occurs in a data set
standard error of the mean
standard deviation / square root of sample size