biostat Flashcards
formula for RR
relative risk
EER/CER
formula for RRR
1 - RR
ARR formula
CER - EER
OR
EER - CER
[since this is the absolute value]
NNT formula for the duration of the study
1/ARR
NNT for 1 year
NNT x (# of years the study was conducted)
odds ratio (OR) formula
[EER/ENR] / [CER/CNR]
discrete data
generally refers to discrete, countable numbers with no decimals (cannot be divided)
ex: people
continuous data
in contract, is more of a continuum in which use of decimals allows for an infinite number of possible values
e: electrolyte levels
discrete data subtypes
nominal (categorical) data
ordinal (ranked) data
nominal (categorical) data
refers to assigning data to different CATEGORIES based on the occurrence of an outcome
ordinal (ranked) data
refers to data that come in a certain order or ranking but the intervals between the values are not necessarily equal
ex: NYHA classification
continuous data subtypes
interval data
ratio data
*can be expressed using decimals
interval data
refers to measurable data with equal intervals between values. there is no absolute zero
ex: temperature (a “0” temperature has meaning)
ratio data
type of data interval that has an absolute zero
ex: height, weight, hemoglobin level
continuous data tests
t - test
student’s t test
paired t test
analysis of variance (ANOVA) test
t-test
continuous data
means of two groups
ex: cholesterol levels OR HgbA1c
student’s t-test
continuous data
two groups are independent and separate
paired t-test
continuous data
test group acts as its own control (“PAIRED”)
ANOVA test
continuous data
groups of >/= 3 groups are being compared (similar to t-test but there are >/= 3 groups)
CAN BE USED FOR BOTH INDEPENDENT AND PAIRED GROUPS
nominal (categorical) data tests
DISCRETE DATA
chi-square test
McNemar’s test
Cochran’s Q test
chi square test
nominal // discrete
two or more independent groups
ex: the risk of GI bleed is compared in ASA group vs placebo group
McNemar’s test
nominal // discrete
two paired groups
Cochran’s Q test
nominal // discrete
> /= 3 paired groups
ordinal (ranked) data tests
DISCRETE DATA
wilcoxon rank sum test
wilcoxon signed rank test
kruskal-wallis test
friedman test
wilcoxon rank sum test
ordinal // discrete
two independent groups
wilcoxon signed rank test
ordinal // discrete
two paired groups
kruskal-wallis test
ordinal // discrete
> /= 3 independent groups
friedman test
ordinal // discrete
> /= 3 paired groups
randomized clinical trial
provides high quality statistical test
members of two groups are chosen at random
considered the gold standard
cohort studies
group (cohort) of people who share a common trait
observed over time
prospective study
outcome of interest is counted or measured
case-control studies
want to see if drug/effect is back –> looking backwards –> looking at previous cases
choose this when it is unethical to expose patients to a risk factor
CONTROL is different here. means individuals who have never had the outcome or event of interest regardless of their exposure status
GOING BACK IN TIME
cross-sectional studies
factors and status of group are studied at one specific time (one cross section of time)
looking at PREVALENCE
cross-over studies
study participants serve as their own control
treatment group will eventually become a control
control group will get the treatment
the two groups are “paired” or “matched”
ITT (intent to treat)
ALL pt data analyzed (all data that was INTENDED to be studied)
pro = closer to real life
PP (per protocol)
counts the data only from the pts who COMPLETED the study
pro = more accurate estimate of the drug
meta analysis
results of many relevant studies are reviewed objectively, quantitatively
want RCTs but need to avoid PUBLICATION BIAS meaning including results from not fancy studies // less exciting studies // studies not published
type I error
false rejection of the null hypothesis
there is no association but researchers THOUGHT there was an association
association was imagined
saw something that doesn’t exist
alpha = type 1 –> want <0.05 (5%)
means chance of type 1 error is less than 5/100
type II error
false acceptance of the null hypothesis
there is an association BUT IT WAS MISSED
AGREE WITH NULL BUT THIS IS INCORRECT~!
type II error = beta = 0.20
power = 1 - beta [power increases with increase in sample size AND/OR when the difference of interest between the two groups is more noticeable and/or when the sample size increases]
dont want chance of type II error to be more than 0.20
80% power is the typical MINIMUM power in order for a study to be considered independent
power = likelihood of NOT making a type II error
summary of statistical significance
RR
RRR
OR
a 95% CI range for ___ that does NOT cross ___ means the study IS statistically significant:
RR; 1
RRR; 0
OR; 1
incidence
the number of NEW individuals that develop an illness in a given time period (usually 1 year) divided by the total # of individuals at risk during that time
prevalence
the number of individuals in a population who HAVE an illness divided by the total population
relationship between incidence and prevalence
prevalence is equal to incidence multiplied by the length of the disease process
if the disease is long term, prevalence is higher than incidence. e.g., HTN
if the disease is short term, incidence is higher than prevalence e.g., meningitis
reliability
means reproducibility of the results if the test is repeated
validity
whether a test is assessing what it is supposed to be assessing. two components of validity are sensitivity and specificity
sensitivity
measures how well a test identifies truly ill people
with a highly sensitive test, a negative result is used to rule OUT the disease
specificity
measures how well a test identifies or rules out truly well people (without the disease)
a positive result in a highly specific test is used to CONFIRM the disease
