Stats Flashcards
Sensitivity?
TP/(TP+FN): Proportion with disease correctly identified
Specificity?
TN/(TN+FP): Proportion without disease correctly identified
Positive predictive value?
TP/(TP+FP)
The proportion with positive test results who have the disease
Negative predictive value?
TN/ TN+FN
The proportion with negative test results who don’t have the disease
How does prevalence change sensitivity and specificity?
Higher prevalence doesn’t change sensitivity or specificity but increases PPV and decreases NPV.
What are alpha and beta errors?
Reject the null hypothesis and it’s true —> alpha error
Not reject the null hypothesis and it’s false —> beta error
Power?
1-Beta error
FP rate?
FP/ FP+TP
FN rate?
FN/ FN+TN
How does sensitivity relate to FNR?
Sensitivity= 1-FNR
How does specificity relate to FPR?
SPP= 1- FPR
What is the odds ratio?
It tells you how the presence or absence of property X has an effect on the presence or absence of property Y.
B/D (odds a control exposed)
CxB (# times event didn’t)
Relative Risk?
Probability of exposed to the probability of not exposed
C/(C+D) (probability of not exposed getting dz)
Attributable Risk?
Excess risk of the outcome (e.g. disease) in the exposed group compared with the non-exposed group
AR = [a/(a+b)] – [c/(c+d)]
Likelihood ratio?
The probability of detecting cancer to the probability of not detecting cancer.
Positive LR = sensitivity / (100 – specificity)
Incident rate?
Number of cases in specific time / population of risk
RR or OR = 1.0 means?
No association of risk with disease
What does a regression analysis do?
Predicts or estimates a value of one variable corresponding to the given value of another variable.
Number Needed to Treat (NNT)
Number Needed to Treat (NNT): numbers needed for a screening test to detect one case to treat OR the number of vaccinations to prevent one case of disease. Ideal = 1
NNT=1/ Absolute risk reduction
E.g. drug reduces bad outcome from 50% to 40%
ARR= 0.5-0.4=0.1
NNT= 1/0.1 = 10
Student t-test
Measure difference between two groups
Paired t-test
measures difference in the mean of two paired tests
Mann-Whitney (Wilcoxon Rank Test)
Non-parametric analogue
ANOVA
more than 2
Chi Square
difference between two or more proportion variables or percentage of categorical outcome
Fischer Exact test
replaces chi square when numbers are small, only two responses (death/alive)
Log-rank
compares survival rate among groups (Kaplan-Meier Curve)
Cox regression
multivariate analyses control confounding variables
McNemmer test
used to measure sample size
Question: how to decrease type I error?
Increase sample size
ROC: Receiver Operator Characteristic?
ROC: Receiver Operator Characteristic True Positive against False positive Closer to the left border is more accurate Close to 45 degree is less accurate (1.0 excellent test, 0.5 worthless)
Best initial test?
Case control
Test for normally distributed data?
SD
Incidence vs. Prevalence
Incidence= New cases = # of cases/ Sum time period Prevalence= # of cases= # ppl with disease/Population.
Criteria for establishing a screening program:
1.Target disease burden the population
2.Known natural history
3.Treatment available
4.Early detection improves outcome
5.Screening method must be acceptable
6.Cost feasible
Two basic study designs:
1.Experimental: clinical trials
2.Observational
Advantages, disadvantages of cohort vs. case control
A. Cohort (prospective)
Advantage: cheap and feasible
Disadvantage: can’t blind and can’t find control
B. Case control (Retrospective)
Advantage: use when outcome is rare
Disadvantage: retrospective can’t account for variables
How to control a Sampling Bias?
Sampling a control in a population at risk
Mean vs Median vs. Mode
Mean: average
Median: value in the middle separates above and below
Mode: Most repeated number
SD?
Standard of deviation (SD): deviation from the mean Variance: SD is the square root of variance (V=(SD)2) SD= ⎇V
SE?
Standard of Error: SD divided by the square root of N
SE= SD/ ⎇N
Z-score?
Z-Score: number of SD that falls from the mean
Z-Score = = sample mean - population mean/ SD/ ⎇N (standard error)
CI?
CI: tells the most likely mean of a population, how much the average value fluctuate
95% CI: 95% chance to contain the true mean of the population
99% is wider. 100% is all the population
HR?
Hazard ratio: HR
the ratio of hazard rates corresponding to the two conditions described.
HR of 1: both groups experiencing the same events
HR 0.33: one group experiencing 1/3 of the events
HR >1 or <1: the survival was better in one group than the other
Kaplan-Meier?
Kaplan–Meier: is a non-parametric statistic used to estimate thesurvivalfunction from lifetime data.
Cox(Proportional Hazards)Regression?
Cox(Proportional Hazards)Regression. …Cox regression: is method for investigating the effect of several variables upon the time a specified event takes to happen.
Median survival?
Median survival: It is the time expressed in months or years when half (50%) the patients are expected to be alive.
How to conduct a RCT?
How to conduct a RCT
1.Hypothesis
2.Frame design, population, outcome what
Weaknesses of RCT?
Weaknesses of RCT:
Expensive
Time consuming
More patients