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;
120 Cards in this Set
- Front
- Back
3 "effects" that an RCT prospectively can measure:
|
1. Relative risk (dichotomous variables)
2. Mean differences (continuous variables) 3. Survival (time to event) |
|
2 measures of statistical significance in an RCT:
|
-P-value
-95% confidence interval |
|
Pvalue:
|
the probability that a result is due to chance
|
|
CI:
|
range of likely values
|
|
3 Uses of Survival Analysis:
|
-For time-to-event data
-To follow treatment of terminal illnesses -When subjects are enrolled for different lengths of time. |
|
Curve used for survival analysis:
|
Kaplan-Meier Survival curve
|
|
The Kaplan-Meier survival curve is based on:
|
the fact that probability of multiple independent events is the product of the probabilities of each event.
|
|
What test is used to determine whether 2 Kaplan-Meier survival curves are statistically significant? What will it generate?
|
A log-rank test - generates a P value
|
|
What are Cohort and Case-control studies used for?
|
Demonstrating associations between variables to show Etiology or Prognosis
|
|
Why would you want to use cohort or case-control studies vs RCT? (3 reasons)
|
-Less expensive
-More feasible -Fewer ethical issues |
|
A cohort study begins with:
|
A study population that is FREE OF DISEASE.
|
|
Who is excluded from a Cohort study population?
|
Any subjects that already have the outcome.
|
|
What is done in a Cohort study once the study population has been chosen?
|
Measurement and classification of the exposed and unexposed groups
|
|
Final step in a cohort study:
|
Measure the # of diseased vs undiseased in each group (exposed vs unexposed).
|
|
How is it decided who gets exposed and who doesn't in a cohort study?
|
Nature decides it.
|
|
Is there randomization, allocation concealment, or intention-to-treat in a cohort study?
|
NO - it is purely observational.
|
|
What is the temporality of a cohort study like?
|
Prospective or retrospective
|
|
What are 4 questions that give level 2 evidence and can be answered by a cohort study?
|
1. Questions of treatment
2. Questions of prevention 3. Questions of etiology 4. Questions of harm |
|
What are 2 questions that give level 1 evidence and can be answered by a cohort study?
|
1. Questions of prognosis/ natural history
2. Questions about diagnosis |
|
ways that you can define the outcome of a cohort study:
|
1. Incidence
2. Relative risk 3. Attributable risk (ARD) 4. Relative risk difference RRD 5. NNT |
|
Definition of incidence:
|
The # of new cases of a disease that occur during a period of time in a population at risk for developing the disease.
|
|
2 Things required in order to be able to express incidence:
|
1. Numerator - the number of new cases
2. Denominator - the total population at risk for disease |
|
EER:
|
Experimental event rate - the incidence in exposed.
|
|
CER:
|
Control event rate - the incidence in the controls
|
|
Relative Risk:
|
EER/CER
|
|
ARD
|
Absolute risk difference
EER - CER |
|
RRD
|
Relative Risk Difference
EER-CER/CER |
|
NNT
|
Number needed to treat
1/ARD |
|
PECO
|
Population
Exposure Comparison Outcome |
|
What is the significance of the PEcO format?
|
It should be included in the hypothesis that begins a cohort study.
|
|
How can you tell if a study's results are valid internally?
|
They should be free from bias and error.
|
|
What is Bias?
|
Systematic error that can give false findings.
|
|
What is Random Error?
|
Noise that reduces precision of measurements
|
|
What is a Confounder?
|
Another variable that influences the dependent variable and may actually be responsible for association or effect.
|
|
What is External Validity?
|
Generalizability
|
|
What factors in a study design contribute to generalizability?
|
Inclusion and exclusion criteria for sample selection.
|
|
Definition of Bias:
|
Any systematic error in a study design, conduct, or analysis, resulting in a mistaken estimate of an exposure's effect on the risk of disease.
|
|
What causes Selection bias?
|
Selecting or classifying exposed and unexposed subjects in such a way that an APPARENT association is seen, if if it's not true.
|
|
Common cause of selection bias:
|
Non-responders - people who decline to participate in a study.
|
|
What is a side-effect of non-responders other than selection bias?
|
Influence on generalizability - the responders may not represent the population of interest.
|
|
2 types of Misclassification of exposure:
|
-Random/nondifferential
-Systematic/Differential |
|
What is Random/Nondifferential misclassification of exposure?
|
When errors in classification of exposed vs nonexposed to one study variable are independent of the level of other variables.
|
|
Exmple of Random/nondifferential misclassification of exposure:
|
-Data entry
|
|
What direction does bias tend toward in a random misclassific. of exposure?
|
Toward the null
|
|
What is Systematic/differential misclassification of exposure?
|
When errors in classification of one study variable depend on the status of other variables.
|
|
In what direction does bias tend in systematic/differential misclassification of exposure?
|
Bias is unpredictable - could result in a type I error.
|
|
How should the exposed/nonexpsd groups in a cohort study compare?
|
They should be similar in every way other than exposure.
|
|
How can you help to ensure that cohort groups are similar?
|
You can't intervene much - it's just an observational study so you can't randomize like in RCT.
|
|
3 ways to control confounding through study design:
|
1. Randomization
2. Selection 3. Matching |
|
How does randomization limit confounding?
|
By distributing confounders evenly between groups
|
|
Can you randomize in a cohort or case control study?
|
No
|
|
How does Selection limit confounding?
|
By letting you study only those with or without the confounder present.
|
|
What is a disadvantage of Selection?
|
It limits generalizability
|
|
How is Matching done?
|
By selecting a control for each individual subject on the basis of confounders.
|
|
3 ways to control confounding through ANALYSIS:
|
1. Stratification
2. Adjustment 3. Multivariate analysis |
|
What is stratification?
|
Seperation of study subjects into groups based on values of confounding variables.
|
|
What is adjustment?
|
Using statistics to control for a confounding variable's effect.
|
|
What does Multivariate analysis do?
|
Adjusts for or determines the influence of multiple variables at once.
|
|
Example of multivariate analysis:
|
Multiple regression
|
|
Major difference in how we deal with confounders in cohort studies versus RCT:
|
RCT - eliminates confounders
Cohort - control confounders |
|
3 Techniques that identify independent variables that may be associated with the outcome in question:
|
1. Linear regression
2. Logistic regression 3. Proportional hazard regression |
|
What is linear regression used for?
|
Continuous outcomes
|
|
What is Logistic regression used for?
|
Binary outcomes
|
|
What is a proportional hazard regression used for?
|
Time-to-event
|
|
2 types of linear regression:
|
-Linear regression for one continuous risk fx and outcome
-Multiple linear regression |
|
What is Multiple linear regression used for?
|
Estimating the contributions of several continuous dependent variables on the outcome or to adjust for confounders.
|
|
What can Logistic Regression be used for?
|
-Multiple independent risk factors, which can be continuous or dichotomous
-One single dichotomous dependent variable |
|
How are results from logistic regression expressed?
|
As individual odds ratios for each variable, holding the others constant.
|
|
What is another name for proportional hazards regression?
|
Cox regression
|
|
What is Cox/Ppl hazards regression related to?
|
Survival analysis - time to event data.
|
|
What is the Cox regression used for?
|
To adjust for contribution of multiple risk factors for time to event analyses.
|
|
So what is achieved by using cox regression?
|
Limit/control confounders in a time-to-event study.
|
|
3 types of Outcome Measurement Validity:
|
1. Face validity
2. Criterion-based validity 3. Predictive validity |
|
What is face validity?
|
How much common-sense is in the measurement.
|
|
What is criterion-based validity?
|
How well a measurement agrees with other measurements of the same phenomenon (ie selfreport of phys activity vs pedometer)
|
|
What is predictive validity?
|
How well a measurement predicts an event.
|
|
What is POEM?
|
Patient Oriented Evidence that Matters
|
|
What is DOE?
|
Disease-Oriented Evidence
|
|
2 types of error:
|
-Random
-Systematic |
|
What does Random error affect?
|
Precision or reliability
|
|
What does Systematic error affect?
|
Validity
|
|
How is random error commonly described?
|
-By standard deviation
|
|
How can random erro be minimized? (4 ways)
|
1. Create standards
2. Provide training 3. Refining instruments 4. Repeating measurements |
|
3 types of systematic error:
|
1. Instrument bias
2. Investigator bias 3. Subject bias |
|
Cause of instrument bias:
|
miscalibrated scale
|
|
What is investigator bias?
|
Differential interpretation of results based on knowledge of exposure
|
|
3 forms of Subject Bias:
|
1. Social desirability bias
2. Attribution bias 3. Hawthorne effect |
|
What is the Hawthorne effect?
|
The change in behavior that may occur when subjects realize they are being observed.
|
|
How is Bias identified?
|
By comparison with a reference standard.
|
|
How is Systematic bias/error minimized? (3 ways)
|
1. Calibration
2. Blinding 3. Unobtrustive measurement |
|
Major difference between Cohort and Case-control studies:
|
Cohort = prospective (mainly)
Case-control = retrospective |
|
How does a Case-control study compare to others in terms of strength?
|
Not as strong as RCT or cohort
|
|
What level 3 evidence can be provided by a case-control study? (4 questions)
|
-Questions of treatment
-Questions of prevention -Questions of etiology -Questions of harm |
|
What level 4 evidence can be provided by a case-control study? (1 question)
|
-Questions about diagnosis
|
|
What unique problem arises in case-control studies compared to cohort or RCT?
|
Selection bias because you're selecting the cases and controls that already have or don't have the outcome.
|
|
What is important about where you get your cases/controls from?
|
They should arise from the same population and be similar in every way except for outcome.
|
|
What does it do when you have >1 control per case?
|
Increases the power of the study.
|
|
What new source of measurement error arises in case-control studies?
|
Differential recall - because the subjects have to recall details of exposure.
|
|
How do you minimize confounders in a case-control study?
|
By matching cases and controls.
|
|
What analysis technique in a case-control study can minimize confounders?
|
Regression techniques
|
|
Why can't you calculate incidence in a case-control study?
|
Because you lack the denominator - the overall population at risk.
|
|
What is the outcome measure for a case-control study?
|
the Odds Ratio
|
|
What is an Odds Ratio?
|
the ratio of "likelihood of having been exposed" in diseased versus nondiseased.
|
|
What is Probability?
|
A # representing the likelihood that a specific event will occur
|
|
How is probability expressed?
|
As the ratio of the number of actual occurrences to the number of possible occurrences.
|
|
Examples of probability (2):
|
-Incidence
-Prevalence |
|
What is Risk as defined and used properly?
|
A probability
|
|
What does it mean to say a woman's lifetime risk of breast cancer is 12%?
|
Out of 100 women ~12 will get it
|
|
What is Odds?
|
The probability of event divided by the probability of NO event
|
|
If a women's risk of breast cancer is 12% what are her odds of getting it?
|
12:88 (.13:1)
|
|
If there's an 80% risk of rain, what are the odds of rain?
|
80:20 = 4
|
|
Formula for Odds Ratio:
|
Odds of Exposure in cases
------------------------- Odds of Exposure in controls |
|
How does an Odds ratio compare to relative risk?
|
It approximates RR if the disease is rare.
|
|
How is an odds ratio interpreted?
|
Similar to relative risk
|
|
What does an OR=1 mean?
|
There is no association between the disease and risk factor.
|
|
What does an OR>1 mean?
|
There is a relationship of the risk factor to the disease - causative.
|
|
What does an OR<1 mean?
|
The risk factor is protective of disease.
|
|
What do you use if you want to look at several risk factors in a case control study?
|
Logistic regression
|
|
What does logistic regression do?
|
Controls for confounders, but does not eliminate them.
|
|
How are results from logistic regression expressed?
|
As an odds ratio
|