Statistics Flashcards
Results are typically reported as:
Means
Or mean change (increase or decrease) from baseline
Proportion
Simple percentages
Risk ratio
Odds ratio
Time to an outcome
Survival curve
Hazard ratio
What is a mean?
Continuous outcomes (e.g., blood pressure) (variable that can take on any value within a certain range)
Compare the average of the outcome between groups
What is a proportion?
The fraction of the total that possesses the outcome
Compare the proportion that have the outcome between groups
What is the absolute risk reduction or absolute risk difference?
The absolute difference between the probability of the event in the control group and probability of the event in the intervention group
What does the ARD tell us?
The proportion of outcome in the intervention group is a % lower than the proportion of the outcome in the placebo group
Is the ARD always negative?
No - Needs to be interpreted as the outcome
What is relative risk?
Dichotomous outcomes (did an outcome occur – yes or no?)
Shows what the “risk” of the outcome in the intervention group is compared to the risk in the control group
Probability of event (intervention group) /Probability of event (control group)
What does the relative risk tell us?
RR = 1.0 No difference in risk between groups
RR < 1.0 Less risk of outcome in intervention group
RR > 1.0 Higher risk of outcome in intervention group
What is the relative risk reduction or increase?
The degree to which baseline risk is reduced (or increased) by the intervention
RRR = 1 – RR
What does the RRR tells us?
The intervention inreaces/decrease the risk of the outcome by ____%
What is the odds ratio?
Shows the odds of the outcome occurring in the intervention group compared to the control group
Odds = probability (risk) of having event /
probability (risk) of not having event
OR = odds (intervention group)/
odds (control group)
OR is usually close to the RR (if the outcome rate is low)
What is told to us by the odds ratio?
OR = 1.0 No difference in odds between groups
OR < 1.0 Less odds of outcome in intervention group
OR > 1.0 Higher odds of outcome in intervention group
What is the NNT?
Number of subjects who would have to be treated (receive the intervention) in order for one additional subject to “benefit” in comparison to the control
Must take into account the duration of the study
NNT = 100/ARD (%)
What does the NNT tell us?
We need to tx ___ # of subjects for (length of study) in order to prevent the outcome in one additional person compared to the control
What is the NNH?
Number of patients who would be treated before you see one additional subject with an adverse event compared to control
Find the ARD of an adverse effect between groups
NNH = 100
ARD (%)
What does the NNH tell us?
We need to treat x number of subjects with Drug X for …(length of study) in order for one additional adverse effect to occur compared to the control
What is a survival analysis?
Kaplan-Meier survival curves
Compares how long it takes subjects in each group to reach the outcome (have a stroke)
Often reported as “median survival time”
–> Time for half of the subjects to reach the outcome (the one we are evaluating)
–> p-value will tell you if there’s a statistically significant difference between the groups
Can flip the curve, just means the amount of people that have experience the outcome
What does the hazard ratio tell us?
HR = 1.0 No difference in hazard between groups
HR < 1.0 Less hazard of outcome in intervention group
HR > 1.0 Higher hazard of outcome in intervention group
What is a confidence interval?What does the size of the C.I. indicate?
Range where the true effect of the intervention (treatment) lies
The narrower the CI, the more precise the results
Increase sample size = increase precision
- Best case and worst-case scenario
What does the C.I> tell us?
The true value of risk reduction will lie between the best and worst case scenarion 95% of the time
Why are C.I. beneficial?
CI can show us the magnitude of the effect
Best-case…worst-case scenario
CI can show us if there really is a difference (statistical significance)
If the CI crosses the threshold for “no difference”, then the result is not statistically significant
For means or proportions: no difference = 0
For relative risk or odds ratio: no difference = 1
What is better: P-value or C.I.?
Reporting CI is becoming more common/required
CI show us the range in the actual data measurements (e.g., mmHg for blood pressure)
CI shows us the direction of the effect, p-value just indicates if a difference exists
CI can help distinguish between clinical and statistical significance
What is the difference in proving efficacy?
Superiority
Want to show an intervention is better than control (active/placebo)
Is there a statistically significant / clinically relevant difference between groups?
Non-inferiority
Is the new intervention not substantially worse than an established intervention?
Equivalence (rarely done)
Is the new intervention neither worse nor better than an established intervention?
E.g. Bioequivalence Study
What is a non-inferiority trial? Why do we want to do these?
Clinical trial to establish that an intervention is not clinically worse than a comparison by more than a pre-determined margin
When would we want to do this?
- Show that something is not substantially worse
- Cost could be benefit
– Different dosage forms (injection od vs 1w) –> New drug (sacrifice a bit of efficacy, but I have a benefit in no a/e/liver toxicity, safer/tolerability, convenience
How does one design a non-inferiority trial?
Same design considerations as superiority trials for inclusion/exclusion criteria, randomization, blinding, outcomes…
Non-inferiority margin determined a priori
–:> predetermined threshold or limit set before conducting a study to evaluate the non-inferiority
Should be decided based on a combination of statistical reasoning and clinical judgement
How can one interpret a non-ionferiority trial?
ITT and per-protocol analyses should be done and reported
If just ITT reported, it can bias towards showing non-inferiority if lots of participants are lost to follow-up or have protocol violations
ITT preferred
Confidence intervals are key
Interpretation of non-inferiority?
Study graph
Whats the difference between non-inferiorty and superiority?
Once non-inferiority is established, an analysis for potential superiority can be conducted (should be pre-specified in protocol)
Lack of superiority in a superiority trial design, does not mean non-inferiority, and cannot claim non-inferiority
What is a propensity score?
The probability that a subject would be in a particular treatment (study) group based on their observed baseline characteristics
How do we use a propensity score to sort subjects?
Use propensity score for:
Matching, Stratification, Covariate Adjustment, Weighting
What is the value of propensity scores?
Goal is to mimic RCTs by balancing observed characteristics (covariates) between study groups
Helps to reduce selection bias and confounding in observational studies
Doesn’t replace the value of randomization!
What is an adaptive clinical trial?
A clinical trial design that allows for prospectively planned modifications to one or more aspects of the design based on accumulating data from subjects in that trial
Describe the adaptive clinical trial design
GRAPH
Give an example of a pre-planned change and its benefit?
Adjusting sample size
-Prevent underpowered studies (waste of time/resources)
-Prevent unnecessary exposure to subjects
Stopping treatment arms/groups
-Prevent unnecessary exposure to useless/harmful treatment
-Prevent wasting resources and time
Adaptive clinical trial should have:
What is an issue with this?
Should have pre-planned all the interim analyses and potential modifications
Interim data analyses – potential for operational bias
Researchers, patients, scientific community behave differently if results known
Statistical analyses (design and results) more complex
