Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
104 Cards in this Set
- Front
- Back
Population- definition
|
all persons with a characteristic in the
universe |
|
sample definition
|
a portion of that population (e.g. representative sample of type 2 diabetics)
|
|
parameter definition
|
measurement that describes the population
(average age, average blood glucose) |
|
statistics definition
|
measurement that describes a sample
|
|
If you could measure the whole population you would not need.....because...
|
“statistics” you have parameters already
|
|
Why are stats needed (and what part of stats helps you to achieve whatever, for 2 of them) (4)
|
Describe study population- descriptive statistics
Tell likelihood experimental results occurred due to chance- test of statistical significance or analytic statistics Draw conclusions Increase experimental efficiency |
|
what determines the proper
descriptive statistics and statistical tests that can be used? |
Type of data
|
|
4 levels of measurement and whether they are discrete or continuous
|
Nominal (discrete)
– Ordinal (discrete) – Interval (continuous) – Ratio (continuous) |
|
nominal data properties (4)
what's it mean? |
basically means...are you this or that?
Weakest level of measurement Assigns numbers or symbols for identity purposes Mutually exclusive and internally equivalent subgroups Only relationship between numbers subgroup is equivalence; can't do any math- no avgs or anything, no order |
|
ordinal data properties (5)
property of ordinal data vs. nominal strength of measurement data measured how? distance b/t subgroups subgroups property example what DOESN'T matter for ordinal variables |
Stronger level of measurement
Same characteristics as nominal, however, there is a definite order Data measured qualitatively, not quantitatively Distance between subgroups does not need to be equal Subgroups have transitivity ex) pain scale – (if a>b & b>c, then a>c) A ordinal variable, is one where the order matters but not the difference between values. For example, you might ask patients to express the amount of pain they are feeling on a scale of 1 to 10. A score of 7 means more pain that a score of 5, and that is more than a score of 3. But the difference between the 7 and the 5 may not be the same as that between 5 and 3. The values simply express an order. Another example would be movie ratings, from * to ***** |
|
interval vs. ratio measurements
|
interval and ratio data are treated the same for all intents and purposes
|
|
interval mesaurements- level of strength for measurement
same characteristics as what type of data except for...(3) |
Stronger level of measurement
Same characteristics as ordinal, however... – true quantitative characteristics – constant distance between numbers – numbers have a defined unit of measurement and represent real numbers |
|
example of interval data
|
ex) blood pressure or age- continuous scale
-can do averages now |
|
ratio measurement- level of measurement
what is it? 3 properties |
Highest level of measurement
interval scale absolute zero- like zero actually has meaning (kelvin vs. F/C) Absolute zero allows the use of all arithmetic operations Numbers represent actual quantities Normally treated like interval data statistically in medical articles |
|
2 types of interval scales
|
– numerically discrete scale
– numerically continuous scale |
|
descriptive statistics (5 kinds)
|
mean, median, mode
frequency distribution st dev st derr 95% confidence interval |
|
measures of the central tendency of the data (3)
|
mean median mode
|
|
Frequency distribution is useful for what type of data
|
discrete (nominal or ordinal) data
|
|
st dev definition
|
a measure of the variability in
response from one subject to another in the sample |
|
st error what is it? equation
|
SD/ square root of sample size- a
measure of how close the sample mean is to the true population mean |
|
what is a 95% confidence level?
|
the estimated value from the
study + twice the SE (standard error) |
|
mean- useful in describing interval/ratio data?
is it used to describe ordinal data? nominal data? affected by outliers? |
yes
no no yes |
|
median- useful in describing interval/ratio data?
is it used to describe ordinal data? nominal data? affected by outliers? |
yes
yes no no |
|
mode- useful in describing interval/ratio data?
is it used to describe ordinal data? nominal data? affected by outliers? |
yes
yes yes no |
|
interquartile range definition
|
measure of statistical dispersion, being equal to the difference between the third and first quartiles. IQR = Q3 − Q1
|
|
RANGE:
useful to describe interval or ratio data? used to describe ordinal data? descriptive of sample variability? assists in statistical inference? used to calc confidence intervals? |
yes
yes yes no no |
|
INTERQUARTILE RANGE
useful to describe interval or ratio data? used to describe ordinal data? descriptive of sample variability? assists in statistical inference? used to calc confidence intervals? |
yes
yes yes no no |
|
STANDARD DEV
useful to describe interval or ratio data? used to describe ordinal data? descriptive of sample variability? assists in statistical inference? used to calc confidence intervals? |
YES
NO yes yes no (but involved indirectly through calc of SE...) |
|
standard error (SEM...)
useful to describe interval or ratio data? used to describe ordinal data? descriptive of sample variability? assists in statistical inference? used to calc confidence intervals? |
yes
no no yes yes |
|
mode
|
most frequent response
|
|
st dev vs st error DIFFERENCE
tell what each does |
-st dev is representation within the sample (variability)
-st error is statistical value to estimate how well value represents larger population |
|
Application of measure of variability (2)
|
interquartile range- middle half of sample (25-75th percentile)??
i have no idea why i wrote this in my notes and st dev... |
|
how dichotomous (branching into 2 categories) are commonly presented (8)
|
-incidence- how often it happens in a pop
-prevalence- how often, right now (not over time) -add hazard ratio to slide- similar to RR (interpreted same way) Relative risk Odds ratio Relative risk reduction Absolute risk reduction Number needed to treat/ Number needed to harm |
|
Hazard ratio, when compared to relative risk
|
takes into consideration that incidence of events are occurring all at end of study (incidence is all at the end of study
|
|
incidence definition
conveys what info? |
the number of new cases (or events)
that occur per unit of time (often calculated per year but be sure you notice the specific interval) conveys information about the risk of contracting the disease |
|
calculation of incidence
|
Number of individuals who develop the
disease (or event) over a period of time divided by: Total number of individuals at risk during that time |
|
prevalence definition
|
the probability of having a disease (or
characteristic) at one point in time |
|
prevalence calculation (how)
|
Number of individuals who have the
disease (or characteristic) at one point in time divided by: total number of individuals in the population at that time |
|
risk definition
|
Synonymous with cumulative
incidence- total number of events over a period of time |
|
risk calculation
|
The number of individuals who
develop the disease during the time period divided by: The number of disease -free persons at the beginning of the time period |
|
definition of odds
|
a proportion that compares the number
of times that an event occurs to the number of times that it does not occur |
|
odds calculation
|
Number of times the event occurs divided by Number of times the event does not occur
|
|
risk vs. odds- when can odds be a good approximation for risk? when is it not (and why)?
|
Odds are slightly different than risk but when the event is rare the odds are a good approximation of risk
frequent occurrence means that odds will not approximate risk as well (denominator difference) example: Risk of drawing an Ace from a deck of cards is 4/52 = 1/13 Odds of drawing an Ace from a deck of cards is 4/48 = 1/12 |
|
what type of study do you usually see odds calculation approximating risk? why?
|
case control study- retrospective so things have already occurred
can't calculate an incidence- so use odds to approximate risk |
|
relative risk definition
|
A ratio of the probability of developing the outcome over a specified period of time if a risk factor is present
divided by the probability of developing the outcome in the same time period if the risk factor is absent |
|
calculating RR
|
see slide
rate of disease in people with factor/rate of disease in people without factor |
|
relative risk- what do the numbers mean if > 1
what if < 1? talk about what it means in percentages too |
If the number is > 1 the risk of that event (or disease) is
greater if the intervention (or risk factor) is present than if it is absent. e.g. if the relative risk of stroke is 3 if you take PPA, you have three times the risk of a stroke if you take it compared to the risk if you do not If the number is < 1 the risk of that event is less if the intervention is present. e.g. if the relative risk of death is 0.8 if you take lisinopril compared to no treatment you have reduced your risk by 20% if you take lisinopril |
|
what if RR is = 1?
|
means that risk of dev disease for exposed group is same as unexposed...so no association.
|
|
confidence intervals- combines info about what 2 things?
|
Combines information about the strength of an association (effect size) and the statistical significance (effects of chance) of the result
|
|
so what is a 95% confidence interval saying? (say in layman's terms)
|
95% of all samples would give a result within the ranged specific by the CI
|
|
For RR or OR- value of 1 represents what?
if CI includes 1 then what does this mean? |
For RR or OR the value of 1 represents no difference between groups (no effect of treatment)
i.e. If the 95% CI of the OR or RR includes 1, the result is not statistically significant |
|
confidence interval size relation to sample size
|
Bigger sample size = smaller confidence interval
|
|
how to calculate NNT
|
1/ARR
|
|
if RRR is the same in 2 instances, does this mean NNT will be the same?
|
NO. depends on ARR
|
|
CER
|
CER = control event rate (comes from study? event rate in control group?)
|
|
PEER- stands for?
what is it? |
PEER = patient expected event rate
refers to the rate of events we'd expect in a patient who received no treatment or conventional treatment. |
|
Adjusting Estimate of Benefit (NNT) for an
Individual Patient- why? |
I guess...if there are factors that increase risk for an event...your NNT will vary depending on risk factors in specific pt groups
|
|
how to adjust estimate of benefit (NNT) for individual patients
2 equations |
use equations like
CER/PEER (i guess this...controls...for it) x NNT or 1/(PEER x RRR) |
|
steps for constructing a statistical significance test (6)
|
state hypothesis: there is a difference between groups
formulate null hypothesis (the one where no difference exists) decide significance level (usually 5%) collect data apply significance test. this determines probability of obtaining the observed data if null is true reject or fail to reject null (reject = effect) |
|
type I vs. II error
|
Type I: Reject the null when it is true (made a conclusion there was a difference but there wasn't)
Type II: Accept the null when it is false (say there is no difference when there actually is) |
|
st dev vs st error DIFFERENCE
tell what each does |
-st dev is representation within the sample (variability)
-st error is statistical value to estimate how well value represents larger population |
|
Application of measure of variability (2)
|
interquartile range- middle half of sample (25-75th percentile)??
i have no idea why i wrote this in my notes and st dev... |
|
how dichotomous (branching into 2 categories) are commonly (8)
|
-incidence- how often it happens in a pop
-prevalence- how often, right now (not over time) -add hazard ratio to slide- similar to RR (interpreted same way) Relative risk Odds ratio Relative risk reduction Absolute risk reduction Number needed to treat/ Number needed to harm |
|
Hazard ratio, when compared to relative risk
|
takes into consideration that incidence of events are occurring all at end of study (incidence is all at the end of study
|
|
incidence definition
|
the number of new cases (or events)
that occur per unit of time (often calculated per year but be sure you notice the specific interval) |
|
calculation of incidence
|
Number of individuals who develop the
disease (or event) over a period of time divided by: Total number of individuals at risk during that time |
|
prevalence definition
|
the probability of having a disease (or
characteristic) at one point in time |
|
prevalence calculation (how)
|
Number of individuals who have the
disease (or characteristic) at one point in time divided by: total number of individuals in the population at that time |
|
risk definition
|
Synonymous with cumulative
incidence- total number of events over a period of time |
|
risk calculation
|
The number of individuals who
develop the disease during the time period divided by: The number of disease -free persons at the beginning of the time period |
|
For RR or OR- value of 1 represents what?
if CI includes 1 then what does this mean? |
For RR or OR the value of 1 represents no difference between groups (no effect of treatment)
i.e. If the 95% CI of the OR or RR includes 1, the result is not statistically significant |
|
confidence interval size relation to sample size
|
Bigger sample size = smaller confidence interval
|
|
how to calculate NNT
|
1/ARR
|
|
if RRR is the same in 2 instances, does this mean NNT will be the same?
|
NO. depends on ARR
|
|
CER
|
CER = control event rate (comes from study? event rate in control group?)
|
|
PEER- stands for?
what is it? |
PEER = patient expected event rate
refers to the rate of events we'd expect in a patient who received no treatment or conventional treatment. |
|
Adjusting Estimate of Benefit (NNT) for an
Individual Patient- why? |
I guess...if there are factors that increase risk for an event...your NNT will vary depending on risk factors in specific pt groups
|
|
how to adjust estimate of benefit (NNT) for individual patients
2 equations |
use equations like
CER/PEER (i guess this...controls...for it) x NNT or 1/(PEER x RRR) |
|
steps for constructing a statistical significance test (6)
|
state hypothesis: there is a difference between groups
formulate null hypothesis (the one where no difference exists) decide significance level (usually 5%) collect data apply significance test. this determines probability of obtaining the observed data if null is true reject or fail to reject null (reject = effect) |
|
type I vs. II error
|
Type I: Reject the null when it is true (made a conclusion there was a difference but there wasn't)
Type II: Accept the null when it is false (say there is no difference when there actually is) |
|
Type I error parameter and accepted value
|
accepted value of this (alpha) is 0.05
|
|
Type II error parameter and accepted value
|
accepted value (beta) not as standard, typical is between 0.10 and 0.20 but many studies fail
|
|
what is power?
|
1 - beta
With a beta of 0.10 you have a 90% chance to detect a difference (of a specified size) if it exists, or 90 % power |
|
consequences of Repeated Tests of Significance such as interim analyses (at 2, 5, 10, 20 tests what happens)
look this up more i don't get why |
if alpha = 0.05
– 2 tests: 8% probability of type I error – 5 tests: 14% probability of type I error – 10 tests: 20% probability of type I error – 20 tests: 35% probability of type I error |
|
If the null hypothesis is tested once, at
significance level of 5% (alpha), this means what? |
the probability of a type I error is 5% by definition
|
|
Sample Size and Power relationship
|
greater sample size = greater power
|
|
Sample size is usually calculated based on...(5)
|
a, b, d , sigma
– Accepted alpha (usually 0.05) – Accepted beta (0.1 to 0.2) – Size of treatment difference you want to detect (referred to as delta) – Measure of variability (SD) for continuous data or Rate of events in control patients for dichotomous data (- if continuous fourth factor is SD, if dichotomous it's rate of events |
|
positive vs negative study
|
positive study finds that there is a difference (definition); negative study fails to find that difference
|
|
regression to the mean: this usually shows up like...(2)
|
Extreme measures move closer to the mean when they are repeated
Patients selected for “abnormalcy” will tend to become more normal when reassessed |
|
for a negative study, how can we tell if it's legit
|
for negative study- look for power to see if conclusion is legit
|
|
_____ drivers power
why? |
-events drives power- if nobody dies/gets hospitalized/gets sick you can't prove that your treatment worked)
|
|
Regression to the mean: things that affect the size of regression (2)
|
The less reliable the form of measurement for the outcome... the greater the expected improvement when remeasured
The more abnormal the initial result of the measure the greater the expected improvement when remeasured |
|
how to control for regression to the mean? (4)
|
use most reliable measures, use
averages of more than one measurement, take measures at more than one time, use a run-in period before trial |
|
Subgroup analysis- how to approach
what 2 questions should you ask yourself and why? subgroup analysis power |
be cautious...in interpreting.
was the subgroup and specific analysis planned before the trial began? Was the sample size adequate to find a difference in the subgroup if one exists (power) study is rarely powered to find treatment difference in subgroups- typically only powered for overall sample; so if fail to find diff, doesn't mean it doesn't exist |
|
subgroup analysis- what must be adjusted for? why?
|
f they don't adjust for multiple testing of subgroups (multiple subgroups?), run risk of type I error (increased)
|
|
subgroup analysis- many times more useful for what?
|
Many times this is more useful for hypothesis generation rather than proof of a hypothesis
|
|
what is an interim analysis?
|
Predefined time points to examine the results during the trial to determine if proof of benefit or of harm has been established
|
|
who typically analyzes interim studies?
|
There is a separate monitoring board designated to do this analysis independent of the primary
study investigators |
|
interim analysis- what must be adjusted?
|
p values must be adjusted for multiple testing
|
|
per protocol analysis
|
only those who actually
complete the protocol are included in the analysis i.e. drop outs are basically ignored |
|
intent to treat analysis
|
each subject will be included
in the analysis for the group to which they were randomized; i.e. their results are counted (usually as a failure) even if they did not complete the full treatment |
|
intent to treat vs. per protocol- which is more rigorous and what does this mean
|
more rigorous analysis (more "real world") and if the results are still positive by this method the conclusions are more reliable
|
|
what must we keep in mind about p-values (2)
|
Statistical significance by itself means little
Weigh against practical concerns |
|
for p-values- what practical concerns must we weigh "statistical significance" against? (5)
|
– what is the size of benefit
– how does it compare to alternatives – what is the CI or other measure of precision – what are the relative toxicities – cost |