Biostats Flashcards
How to interpret AUC
Larger AUC= better diagnostic accuracy
What does AUC of 1.0 mea?
100% sensitivity and specificity
What does odds ratio of >1 vs <1 mean?
OR > 1 means exposure is associated with higher odds of an outcome
How to interpret CI of odds ratios
If it runs over 1.0, its not significant
What is non-response bias and how does it affect study results?
In questionaires, people who don’t respond may be very different from those that do respond. (i.e. people with more severe disease may not be able to respond to the questionaire, thus skewing your results)
How does stratifying help determine if a variable is confounding your results?
If there appears to be a significant difference between 2 groups, if you stratify the groups by one of their common a different variable and there is no signiicant difference, then the variable that you stratified is the confounding variable
“stroke is higher in obese than non obese, but when participants are separated into diabetes vs no diabetes, there isn’t a significant difference in rates of stroke in obese/non obese people”
Define Study power
The ability for a study to detect a true difference between two groups
2 ways to increase the power of a study
Increase the sample size
Increase the precision of measurement (more likely to detect a true difference)
What is the benefit of using Liklihood Ratios?
They are completely independent of a diseases prevalence
Susceptibility bias
The test and control groups have different prognosis based on some unforseen variable (i.e. confounding variables)
What is verification bias and when does it happen?
Happens in studies that use the gold standard treatment method as the control, and test against a new method. SO, to reduce this, you have to use the gold standard method on a small random group of people from the new group to VERIFY that the new method is better/worse than gold standard
Factorial study design
Study with >2 interventions and all the combinations of those interventions
Vit D alone
Calcium alone
Vit + calcium
Neither Vit or Cal
Crossover study
Participants are exposed to different treatments as the study progresses…essentially, they CROSSover to a different study arm mid way through to assess how the other intervention works for them
What is a pragmatic study
A study that aims to determine if an intervention works under real world conditions
i.e. NOT a randomized controlled trial, because in real life, people aren’t randomized and nothing is controlled
What is net clinical benefit, and how do you measure it?
Basically, weighing the risks and benefits of an intervention and determine if there is net benefit
NCB = (greatest possible benefit - greatest possible risk)
What is intention to treat analysis
Every participant is analyzed based on the group they were randomized to…regardless of whether they actually adhered to the treatment plan
What are the 3 pillars of designing a randomized controlled trial
Randomization
Blinding
Intention to treat analysis
How does power of a study affect confidence intervals?
The higher the power, the narrower the confidence interval
So, if a study as a really wide CI, it tells you the power of the study is poor (beause power means the study’s ability to detect a difference between 2 groups. If you have a wide CI, your ability to identify the difference isn’t very good since you have a huge range)
Positive Likelihood Ratio definition
How likely is a person WITH disease to test positive, as compared to liklihood of a person without disease testing positive
Negative Likelihood ratio definition
How likely is a person WITH disease to test negative, as opposed to likelihood of a person without disease testing negative
Positive LR equation
Sensitivity/ (1-Specificity)
NEgative LR equation
(1-sensitivity) / Specificity
Time to event analysis (when and what its used for)
Commonly used in survival analysis where the outcome is death…it looks at both groups in the study and analyzes how much time occured prior to their death to determine if there was a significant difference
Effect Modification
When an extraneous variable changes the strength or direction of of the effect that the Exposure has on the Outcome
Difference between effect mod and confounding variable
Effect mod - extraneous variable changes the strength or direction of the effect whereas confounding masks it
How to identify confounding vs effect mod
Stratification
Stratify confounding variable - no change between groups
Stratify effect modifying variable- Significant change in the relationship
Ecological Fallacy
A type of bias that occurs when you try and apply population level data to an individual
What does Kappa represent and how to interpret it
measures reliability between multiple observers
k = -1, means total disagreement k= 1 means total agreement
Interpreting Neg Likelihood ratio
Range 0 to infinity
If >1, liklihood of disease is increased
If <1, liklihood of disesase is decreased
Numerator and denominator for Incidenc
new cases / # of people at risk for getting disease
Exclude people who already have the disease or who have died etc.
Standardized Mortality Ratio
Observed deaths / Expected Deaths
Relationship of sample size to standard deviation
Inverse
More population = less variance in results
Ecological study design
Use population level data to make conclusions about the population as a whole
“Decrease in cigarette sales in USA is associated with decreaesed incidence of lung cancer”
3 main categories for bias
Selection Bias
Measurement Bias
Confounding bias
2 subclasses of measurement bias
Recall Bias
Observer Bias
What does selection bias mean?
Your participants are not representative of the study population
Berkson bias
Using sick individuals (hospitalized patients), and using the results to apply to the general population
Referral Bias
Comparing patients at high specialized hospitals and community hospitals…the onces at specialized centers are likely sicker and have more comorbidities
Prevalence (neyman) bias
When incidence of disease is estimated based on prevalence…thus, the selective survival is skewed in favor of the healthier people
Susceptiblity Bias
When the intervention chosen for each participant is based on how sick they are.
Ex- Sicker people with ACS get meds, while healthier people with ACS get cath. Results show cath is far superior to meds…is it just because the people that got the cath are healther in the first place?
Observer Bias aka
Ascertainment bias
Detection bias
If you see a funnel plot, think:
Publication bias
Hazard ratio >1 vs <1 interpretation
> 1 = the exposure has a detrimental effect
<1 means it has a protective effect
Relationship between Rate and confidence interval (interpretation)
Does not follow same rules as RR, OR etc…even if CI for a rate crosses 1, you can’t say it isn’t significant
Multiplicity
Occurs when you test mulitple secondary end points. By doing this, you increase your risk of having type I error (false positive)
Length time bias
Benefits of a screening tool are overestimated because they catch all the slow progressing, mild cases. However, they miss all the rapidly fatal causes of the same disease, so it makes it look like survival is way higher than it actually is
The LENGTH of the illness is what causes the issue
Lead time bias
Length of survival from time of diagnosis to death is falsely prolonged.
Ex- everyone dies in year 10. If you catch the disase in year 2 vs year 7, it will make it look like survival is increased. But, its the same, you just caught it earlier