CAPS written exam Flashcards
What is stratifying?
Stratifying means conducting separate analyses for each group
When is stratifying not useful?
When there are multiple confounders
What is the effect of stratifying on power?
It reduces power by reducing sample size
Randomization, matching and multivariate statistical adjustments are _
Alternatives to stratifying
What is an adjusted estimate?
The result of multivariable regression
What type of multivariate regression would you use for categorical (non-continuous) outcomes?
Multivariable logistic regression
What type of multivariate regression would you use for continuous outcomes?
Multiple linear regression
What type of multivariate regression would you use for time to event outcomes?
Multivariable Cox regression
In a y=mx+b model, the correlation coefficient (r) estimates _
how well outcome variable y can be predicted given variable x
Time to event data is best plotted using _
Kaplan Meier Survival curves
For a given survival curve, how do you calculate median survival time?
Find 50% on Y, extrapolate to X
What is censoring in survival data?
When the participant doesn’t participate for the duration of the experiment. (Reach outcome, lost to followup)
What is the log rank test?
A test to determine whether median survival is equivalent between groups (survival curve)
What is cox regression?
multivariable analysis of time-to-event data and is commonly seen in clinical trials and cohort studies.
A regression analysis that allows for statistical adjustment of hazard ration is _
Cox regression
The output of cox regression is _
Hazard ratio
What is sensitivity?
Measures how well a test identifies the presence of disease when it is present
Sensitivity can be calculated as _
Sensitivity=TP/(TP+FN)
To rule out a diagnosis you’d want a test with high _
Sensitivity
What is specificity?
Measures how well a test identifies the absence of disease when it is absent
Specificity can be calculated as _
Specificity=TN/(TN+FP)
To rule in a diagnosis you’d want high _
Specificity
Screening tests have high _ while confirmatory tests have more _
Sensitivity
Specificity
Positive predictive value is calculated as _
PPV=TP/(TP+FP)
True positives out of all positives
Negative predictive value is calculated as _
NPV=TN/(TN+FN)
True negative out of all negatives
How is PPV influenced by prevalence?
In general, the positive predictive value of a test is greater when the disease is more prevalent in the population.
How is NPV influenced by prevalence?
When prevalence increases, NPV decreases.
What is likelihood ratio?
LR relates pre-test and post-test probabilities.
LR+ is _
likelihood ratio for a positive test result
LR- is _
likelihood ratio for a negative test result
A high LR
increases the post-test probability relative to the pre-test probability
A low LR
decreases the post-test probability relative to the pre-test probability (means the disease is less likely).
How do you use a nomogram?
Connect the pretest probability and the LR (or post-test probability)
Convenient appraisal _
Taking everything at face value
Reading only the abstract and relying on its conclusions
Reading the entire paper and relying on its conclusions
Cynical appraisal _
Outright rejection of the paper due to one or two questions or concerns about the study, regardless of their importance
Ultimately, taking little or no value from a study
Critical appraisal _
Systematically considering positive and negative aspects of a study’s approach
Validity is a measure of accuracy. It may be reduced by _
Systematic error
Three steps to assessing internal validity are _
Bias
Confound
Chance
Bias is a systematic error. Selection bias affect _, while information bias affects _
Who ends up in the study
What end up in the study
Over estimation or underestimation of effect can result from _ (2)
Bias
Confounding
Suppose that 30 subjects in the placebo group and no subjects in the treatment group drop out due to illness. Because these dropouts are not counted in the calculations at the end of the study, this is a type of _
Selection bias
Pts over-reporting certain types of behaviors / or under-reporting is a form of _
Information bias
Ways to evaluate the effects of chance on data include _
Evaluation of p-value and confidence interval
External validity is also known as _. It should be conducted after _
generalizability.
Internal validity assessment
What are phase I, 2 and 3 trials?
first-in-man
small number of participants from the presumed target population
large groups of patients assigned to the new intervention or a control
In clinical trials, information bias can be reduced by _ (2)
Placebo
Blinding
What are single, double and triple blind studies?
Patients
Patients + clinicians
Patients, clinicians, data analysts
Confounding can be reduced in clinical trials by _
Randomization
In a clinical trial when participants in one group are more likely to drop out during the study period—say, due to the side effects of an intervention under study, this can violate _
Randomization
What is intention to treat analysis?
participants in RCTs should be analyzed the same way that they were randomized at the beginning of the study
NNT (number needed to treat) can be calculated by _
1/ARR (Absolute risk reduction)
Absolute risk reduction is _
[Incidence in exposed group - incidence in non-exposed group]
An NTT of 3.3 means _
four patients would need to be treated with the drug in order to prevent one case of dementia, on average
In cohort studies,
Participants in a cohort study are selected and sorted based on the presence or absence of an exposure of interest. They are then followed for the development of an outcome of interest
Confounding is more a problem is {RCT/cohort study]
Cohort study
In a case control study _
participants are selected based on the outcome—that is, whether they are a case or a control—and are evaluated with respect to a prior exposure.
Case control are [retrospective/prospective/both]
Retrospective. Out come already occured
The best type of study for a rare disease (hard to find participants) is _
Case control
The statistical outcome for a case control study is _
Odds ratio. Cannot calculate incidence rates since events have already occured
A major difference between RR and OR is _ (relative to incidence rate)
RR takes incidence rates into account
Recall bias is an especially difficult problem is what type of study design?
Case control
Case control is outcome, look back
Cohort study is exposure, look forward (or back)
Cross sectional looks at exposure and outcome at same time
Nothing to see
The major statistical outcome for cross sectional study is _
Odds ratio
You can use cross sectional studies to assign causality. True or false _
False
Two types of descriptive study designs are _
Case study
Case series
When multiple causes contribute to an event, this is referred to as _
Multifactorial cause
Competing causes _
Result in the same outcome via different mechanisms
Effect modification is where _
effect of an exposure differs with the presence of another variable. The effect modifier interacts with the exposure to change the effect of the exposure on the outcome.
