Flashcards in Danny Liew Lectures Deck (94):
What is level one level of evidence?
systematic review of RCTs
What is PICOT?
What do you call the extent to which the results of a study are valid for the sample studied?
What does internal validity depend on? 3 things
Why randomize? 2 reasons
Why would you stratify randomization? eg. country/smoking status
make composition of groups more similar with regarding key confounders and reduce confounding
what's selection bias?
investigators assigning to particular intervention
reduce information bias
2. double blind
3. triple blind
2. subjects + investigators
3. subjects + investigators + outcome assessors
when outcomes are determined according to strict standardized, objective criteria, this is called:
Objective Outcome Ascertainment
How would you achieve Objective Outcome Ascertainment in a multi-centre study?
have a centralised process
What's intention to treat?
Keeping the subjects in their randomized group regardless of actual dropouts, crossovers, losses
What's the point of intention-to-treat?
reduce selection bias
What does ITT analysis do to treatment effect?
always underestimates because assumes:
less in intervention group
more in control group
What's the p-value?
probability that the result arose from chance
What's the cut off for p-value?
0.05 not stat significant
What's a 95% confidence interval?
interval where you're 95% sure the value is within
What's it mean if there's no null value?
the result is stat significant
What is a null value?
if there's no difference between groups compared:
1.0 for ratios
0 for absolute risk differences
How are precision and Confidence Interval (CI) related?
The narrower the CI window, the more precise
If you have bigger sample size vs. smaller sample size, how does that affect CI width?
bigger = narrower CI
smaller - wider CI
What is a type 1 alpha error?
study shows effect, but in reality no effect
what is a type 2 beta error?
study shows no effect, reality there is an effect. (false negative)
What would lack of power indicate?
non stat significant results possibly due to small sample size
Number needed to treat is what?
how many people need to be treated to prevent outcome in one person
how to calculate NTT?
1 / absolute risk or rate reduction
What's NNT affected by? 2 things:
underlying likelihood of outcome
External validity depends on?
Does a systematic review only focus on a single question?
How is the criteria for the systematic review?
what kind of data does a systematic review looks at?
4 key words for systematic review process:
What are the 4 purposes of a meta-analysis?
answer other questions
3 important data sources for systematic review
Medline, Embase, CINAHL etc
inline with question
aware of bias eg. language
How many people for selecting studies?
2 people independently
Quality of studies included important why?
garbage in = garbage out
EXAM question re: forest plots: make sure you know it!
what should be done about heterogeneity?
what determines weighting of ind. studies?
2 measures of outcome important in systematic review?
relative: RR, OR
absolute: mean difference
validity of a meta-analysis relies on what?
studies that are similar enough to be pooled
What two kinds of similarity determines metaanalysis potential?
statistical - effect sizes/variances
non-statistical - PICOT
Can you objectively assess non-statistical heterogeneity?
When do you order a diagnostic test?
if clinical suspicion of disease
Are the results mostly definitive or preliminary in a diagnostic test?
Whats the purpose of a diagnostic test?
to confirm disease
Who do you apply a screening test to?
no clinical suspicion of disease
Are the results mostly definitive or preliminary in a screening test?
preliminary - need confirmation
How do you calculate sensitivity?
True Pos/True Pos+False Neg
% positive test who actually have it
How do you calculate Specificity?
True Neg/True Neg+False Pos
% negative test who actually don't have it
How do you calculate Positive predictive value?
% positive tests that are truly positive
How do you calculate Negative predictive value?
% negative tests that are truly positive
Are sensitivity and specificity constant?
Yes, they are inherent to a test
What are PPV and NPV dependent on? 2 things:
Prevalance of disease
Utility of diagnositc/screeing test is highly dependent on what?
prevalence of disease
Should you just screen EVERYONE?
need to be targeted to high risk groups: prostate exam to men over 50
T/F? To screen a disease you should use a diagnostic test?
FEV1 and COPD uses a continuous scale, what denotes disease versus non-disease?
If you have a lower threshold, how does that affect sensitivity/specificity?
If you have a higher threshold, how does that affect sensitivity/specificity?
What is a Receiver Operator Characteristic curve (ROC)?
plot of 1-specificity vs. sensitivity
represents trade-off between sensitivity/specificity.
Ideal test on an ROC curve is where?
upper left, 100% sensitivity with 0% 1-specificity
What is a worthless test on ROC curve?
diagnonal line, with a perfect correlation of axis, that means 50-50 chance it's there or not there.
on an ROC curve, where do you find the discriminating ability of a test?
area under the curve, above the diagonal
What is the rationale for screening?
early detection -->better outcomes
what is primary prevention?
identifying risk factors
what is secondary prevention?
identifying early disease
what population is screening undertaken on?
largely healthy people
What is an important criteria for screening that may be overlooked?
limitations of screening tests include 4 things:
3 biases in screening
What is selection bias in screening?
healthy people more likely to be screened
What is lead-time bias in screening?
early detection, not prolonged survival
What is length-time bias in screening?
detection of non-aggressive diseases
Whats the difference between prevalence and incidence?
Prevalence: number of existing cases at a single point in time (% or proportion)
Incidence, is number of new cases in a time interval (rate)
new cases/population at risk
new cases/follow-up person-time
What's so great about person-time?
reflects a more accurate picture of the rate
T/F Risk is better than rate in representation?
Nope. Rate is better cause it uses person-years
What's the difference between risk and hazard?
Risk is a single point in time
Hazard: continuously updated rate, applicable throughout the entire time period
2 kinds of associations are:
Cause - effect
Difference between Absolute risk/rate and relative risk?
Absolute: isolated number, no indication of causes
Relative risk: provides association
T/F? Relative Risk, risk ratio mean the same thing?
How do you calculete RR?
What is attributable risk?
absolute magnatude of change in risk/rate of outcome with associated exposure (exam)
How to calculate AR?
How to calculate AR %?
What does AR% mean?
proportion of incident disease among exposed people that is DUE TO exposure
How to calculate Population Attributable risk? (PAR?)
PAR = Rt-Ru
Rt-risk/rate in whole population
Ru-risk/rate in unexposed
How to calculate PAR%?
Preventable fraction is a synonym for what?
Population attributable risk percentage
What does preventable fraction mean?
If you remove the risk factor, you help the PAR% number of people