Stats Flashcards
Likelihood ratio
Express a change in odds
Positive = increase in probability of disease if test is positive. Larger the +LR, more informative the test.
Negative = decrease in odds of having disease with negative results.
Smaller -LR, more informative the test
Positive predictive value
Ratio of patients truly diagnosed as positive to all those who had a positive test result:
TP = TP/(TP+FP)
Negative predictive value
Ratio of patients truly diagnosed as negative who are negative for the test
NPV = TN/(TN+FN)
Likelihood ratio calculate
LR+ = DSens/1-DSpec
LR- = 1-DSens/DSpec
Standard error of mean
Standard deviation ÷ Square root of n
Power
Measure of a studies ability to detect the smallest difference between 2 groups, considered clinically worthwhile.
Probability that a type 2 error (incorrectly accepting null hypothesis) will not be made
Power increase
Large sample size
Increase p-value
Decrease in variability of outcome measure (standard deviation)
Variance
Dispersion of values around the mean
Square of SD=variance
Sum of (difference between all values and the mean) squared÷ n-1
Standard deviation
A statistical measure of the degree of data spread about the mean
SD = square root of variance
Standard deviation in normal distribution
1 SD from the mean, 68% of observations
2 SD from the mean, 95% of observations
3 SD from the mean, 99.7% of observations
Standard error of the mean
Standard deviation ÷ Square root of n
Confidence level
1-alpha
Ability to reject alternate hypothesis correctly as there is no effect
Type 1 error
Alpha
Failure to reject null hypothesis, e.g. saying there is a difference when there isn’t
False positive
Caused by multiple testing
Power
1-beta
To find true effect
Beta
Type 2 error
Incorrectly accepting null hypothesis, saying no difference when there is
False negative
NNT
1/ARR
RRR, relative benefit increase
CER-EER/CER
Prevalance equation
Prevalance = incidence x duration
Concurrent validity
Agreement of a new instrument to a gold standard instrument
Content validity
Are the contents e.g. subscales in line with what test is meant to measure
Convergent validity
The extent to which the same trait is measured by different methods
Confidence Interval
Mean +/- 1.96 x standard error
