FA - Behavioral science - statistics Flashcards
cross-sectional study
what’s happening? (observational)
assess prevalence
case control study
what happened?
assess risk factors (prior exposures)
assess Odds Ratio
cohort study
What will happen? Does exposure increase likelihood of disease
assess Relative Risk
case series
observational - smaller # of ppl w/ disease (ø controls) to generate profile of disease (characterize a new type of disease)
sensitivity
proportion of ppl w/ disease who test (+)
specificity
proportion of ppl w/o disease who test (-)
high sensitivity =
when negative - rule out dz!
SN-N-OUT
high specificity =
when positive - rule out in!
SP-P-IN
PPV
proportion of + tests that are true +
varies with dz prevalence and pre-test probability
(PP = + +)
NPV
proportion of - tests that are true -
varies inversely with dz prevalence and pre-test probability
(NP = - +)
when is OR used?
case control studies
when is RR used?
cohort studies
RRR
relative risk reduction - proportion of risk reduction attributable to an intervention compared to a control
= 1 - (% dz w. intervention / %dz w.o intervention)
ARR
absolute risk reduction - difference in risk attributable to an intervention compared to a control
= % dz w.o intervention - %dz w. intervention)
AR
attributable risk - ∆ risk between exposed and unexposed (proportion of occurrences attributable to exposure
= % exposed w. disease - % unexposed w. disease
needed to treat for one person to benefit
= 1/ ARR
needed to harm (# of patients who need to be exposed to risk factor to be harmed)
= 1/ AR
cohort study where 3 different grps are studied over time (ie smokers vs non-smokers vs former smokers)
type of bias?
selection bias
study looking only at in patients
type of bias?
selection bias - berkson bias
study looking at a disease with an early mortality or loss to follow-up (esp of a particular type of population)
type of bias?
attrition bias (type of selection bias)
note that this type of bias does not occur when the losses happen equally and randomly between the exposed and unexposed groups.
studying populations that are generally healthier than the general population
type of bias?
selection bias - healthy workers + volunteer bias
patients w. disease remember exposure after learning of similar cases
type of bias?
recall bias
groups who know they’re being studied behave differently than they would otherwise
type of bias?
measurement bias - hawthorne effect
(think of a hawk watching a bird on a thorne - the bird is likely to act differently if it knew it was being watched by a predator)
patients in treatment group spends more time in highly specialized hospital units
type of bias?
procedure bias (subjects in different groups are not treated the same)
observer expects treatment group to show signs of recovery, and is likely to document more + outcomes
type of bias?
how to prevent?
observer-expectancy bias
prevent by performing a double blind study in which neither subjects nor investigators are aware of treatment assignments
pulmonary dz is more common in coal workers than the general population - however, people who work in coal mines are also smoke more frequently than the general population
type of bias?
confounding bias (factor related to both the exposure + outcome and distorts the effect of the exposure on outcome)
early detection = increased survival
type of bias?
lead time bias - early detection shows increased survival, even though the natural history of the disease has not changed
SD
how much variability exists from the mean in a set of values “spread”
SEM
measure of how accurate the means is relative to the real mean (measure of CONFIDENCE in the sample mean)
= SD/√n
1 SD =
contains 68% of values
2 SD =
contains 95% of values
3 SD =
contains 99.7% of values
Type I error (α)
what is it?
what is it also known as?
stating that there IS a difference/effect when none exists
(null hypothesis IS incorrectly rejected)
aka fαlse + error
α = you sαw a difference that did not exist
Type II error (ß)
what is it?
what is it also known as?
stating that there is NOT an difference/effect when one exists (null hypothesis is NOT rejected when it is in fact false)
aka false - error
ß = ßlind to difference that exist (setting a guilty man free)
meta analysis
pools data together from several similar studies to reach an overall conclusion
confidence interval
range of values in which a specified probability of the means of repeated samples would be expected to fall
CI = range from (mean - ZSEM) to( mean + ZSEM)
95% CI = Z = 1.96
99% CI = Z = 2.58
if 95% CI for a mean difference btwn 2 variables includes 0, then there is no sig. difference = null is NOT rejected
if 95% CI for OR or RR includes 1, then there is no sig. difference = null is NOT rejected
if CI btwn 2 groups do not overlap = significant difference exists
If CI btwn 2 groups overlap, usually no significant difference exists
if 95% CI for a MEAN difference btwn 2 variables include 0, then
there is no sig. difference = null is NOT rejected
if 95% CI for OR or RR includes 1, then
there is no sig. difference = null is NOT rejected
if CI btwn 2 groups do not overlap
significant difference exists
If CI btwn 2 groups overlap
usually no significant difference exists
of groups tested in t-test
means of 2 grps
T is meant for 2
grps tested in ANOVA
3 (or more) groups
anova = analysis of variance = 3 words
chi square
checks difference between 2 or more %s or proportions of categorical outcomes
Chi-tegorical outcomes (ex: % of members of 3 different ethnic groups who have essential HTN)
disease prevention 1˚ 2˚ 3˚ 4˚
1˚ = Primary = Prevent 2˚ = Secondary = Screen 3˚ = Tertiary = Treatment to reduce 4˚ = quaternary = risk/harm of unnecessary treatment
effect modification
occurs when the effect of a main exposure of an outcome is modified by another variable
Note: this is NOT a bias (it is NOT due to flaws in the design or analysis phases of the study)
Ex1 - smokers taking the Rx have an increased risk of developing DVT while non-smokers taking the Rx do not
Ex2 - likelihood of asbestos exposure will result in lung cancer (a phenomenon significantly impacted by smoking)
latent period
time elapsed from:
initial exposure to clinically apparent disease
(ie infectious dz)
or
exposure to risk modifiers (ie antioxidants, better diet) to when the effects of the exposure is clinically evident
berkson’s bias
a type of selection bias created by selecting hospitalized patients as the control group
pygmalion effect
researcher’s beliefs in the efficacy of treatment that can affect the outcome (ie the greater the expectation placed on people, the better they will perform; form of self-fulfilling prophecy)
lead time bias
apparent prolongation of survival after applying a screening test w/o any real effect on prognosis
recall bias effect
inaccurate recall of past exposures by patient (esp if they’re asked to recall something from long ago)
hawthorne effect
tendency of a study population to affect an outcome due to knowledge of being studied
case fatality rate
of fatal cases / (# fatal + # non-fatal)
if RR is given and the 95% CI does not include 1, what p value is expected?
p<0.5 (significant)
if RR is given and the 95% CI does include 1, what p value is expected?
p>0.5 (not significant)
if 95% CI for a mean difference btwn 2 variables includes 0, what p value is expected?
p>0.5 (not significant)
if 95% CI for a mean difference btwn 2 variables does not include 0, what p value is expected?
p<0.5 (significant)
referral (admission rate) bias
when the case and control population differ due to admission or referral practices
(ie study involving cancer risk factors performed at cancer center may enroll patients from all over the country, whereas hospitalized control subjects may come from only the local area
detection bias
risk factor itself may lead to extensive diagnostic investigation and increase the probability that a disease is identified.
allocation bias
may result from the way that treatment and control groups are assembled (if it’s not assigned in a non-random fashion)
sampling bias
non-random sampling of a population. can lead to a study population that has characteristics that differ from the target population.
severely ill patients are the most likely to enroll in cancer trials, leading to results that are not applicable to patients w/ less advanced cancers. type of bias?
sampling bias
patients who smoke may undergo increased surveillance due to their smoking status, which would detect more cases of cancer in general. type of bias?
detection bias
in a study comparing oral NSAIDs and intraarticular steroid injections for treatment of OA, obese patients may preferentially be assigned to the steroid group. type of bias?
allocation bias
attributable risk percent (ARP)
measure of the impact of a risk factor (or excess risk in a population that can be explained by exposure to a particular risk factor)
ARP = (RR - 1)/ RR
probability/chance of getting one (+) test result of x number of tests using a serologic test that has 95% specificity
probability (all negative) = 0.95^x
probability (at least 1 positive) = 1 - 0.95^x
Power of a study - what it is and how do you calculate it?
indicates probability of seeing a difference when one truly exists; reciprocally related to a type II error (ß) ie stating that there is no difference between groups when one truly exists.
Power = 1-ß
to measure validity of a new screening test, what must you do?
results must be compared to those obtained with the “gold standard” test on the same individual.
validity (accuracy)
how close is the test to the true value?
reliability
how reproducible is a test? does it give similar or very close results on repeat measures or are they far apart?
draw a 2x2 and calculate OR
when do you use OR?
OR = (a/c) / (b/d)
used for case control
draw a 2x2 and calculate RR
when do you use RR?
RR = ( a / (a+b) ) / ( c / (c+d) )
used in cohort studies
Relative risk reduction (RRR) equation
= 1-RR
= (unexposed - exposed )/ unexposed
Attributable risk (AR) equation
Attributable risk percentage (ARP) equation
AR = exposed - unexposed
= ( a / (a+b) ) - ( c / (c+d) )
ARP = (RR - 1 )/ RR
needed to harm 1
1/ AR
Absolute Risk Reduction (ARR) equation
= unexposed - exposed
= ( c / (c+d) ) - ( a / (a+b) )
needed for one person to benefit equation
1 / ARR
incidence calculation
new cases per year / total population at risk (does not include those already affected)
late look bias
problem with gathering information about severe diseases, since the most severe cases will be dead or inaccessible before their information can be gathered.
calculating SEM?
SD/√n
of these, which one will change BOTH incidence and prevalence?
new treatment new vaccine increased death from disease decreased risk factors increased recovery increased survival
new vaccine
decreased risk factors (1˚ prevention programs)
What is a (+) skew distribution and how will it affect:
mean, median, mode with respect to one another?
where the bulk of the population lies?
where the longer tail is?
mean > median > mode
bulk of population lies towards the L
longer tail: R
What is a (-) skew distribution and how will it affect:
mean, median, mode with respect to one another?
where the bulk of the population lies?
where the longer tail is?
mean < median < mode
bulk of population lies towards the R
longer tail: L
If there is a guassian distribution (100% bell shaped), how are these values affected?
mean, median, mode with respect to one another?
where the bulk of the population lies?
where the longer tail is?
mean = median = mode
bulk of population lies in the center
longer tail: equal on both ends
How do you determine the median if there are even numbers?
median is then the average of the middle two numbers
duhhhh
what can be done to control potential confounding data (ie age, race)
MATCHING