Precision in estimates of treatment effects Flashcards
Compare the 3 study designs of a trial
Do you think it is important to pre-specify primary and subsidiary outcomes?
Yes
- Failure to do so decreases the validity of the trial and avoids the impression that the researchers are “fishing” for a positive result.- Studies that fail to do so are also less influential when evidence is assessed to determine clinical guidelines
Issues with studying the whole population:
•Very difficult to do! •Possible if routine sources of health data are available... still limited •Not necessary •Things change over time •Need updated information
Why do we use samples?
We use samplesto provide estimatesof population parameters
Sampling Bias
Sample is selected in such a way that the individuals chosen are not representative of
the whole study population
Example:- estimating SBP in middle aged adults in England, but only those that are in
employment will underestimate true answer
How do we avoid bias?
Take random sample if possible of the population of interest
Difference between accuracy and precision
Accuracy is to do with getting
the right answer – avoiding bias
Precision is to do with inherent
error associated with any
estimate
Define Reference Range
A measure of the spread of the continuous numerical data only
What is confidence interval
confidence interval is a measure of statistical uncertainty
or “precision”
Confidence intervals are
used to quantify the
precision (or uncertainty)
around estimates
A measure of the precision of a sample estimate
How would you interpret confidence interval
- Reference range: In population of similar 45y old men, mean SBP is ~133mmHg, expect 95% of men to have SBP values between ~104-162mmHg;
- Confidence interval: 95% sure that average SBP in similar population of men would be from 132-133mmHg
Formula for reference range
Formula for confidence interval
Formula: Mean +/- 2 Standard Deviations
Formula: Mean +/- 2 Standard Errors
When is the confidence interval statistically significant?
It is statistically significant when it doesn’t include a null value (0)
Risk ratio
Risk ratio (RR) = risk of outcome in drug ÷ risk of outcome in placebo (control)
Calculate relative risk
Value of no effect = “1” Unitless ratio (units cancel out)
Risk ratio = 0.79 for simvastatin vs placebo
How would you interpret this?
Risk of vascular event 21% lower for simvastatin vs placebo
Tip! You can derive this interpretation by subtracting 1 from the relative risk
- 79 minus 1 = -0.21 => 21% decrease
- 79 minus 1 = +0.79 => 79% increase
The 95%CI for the relative risk for any vascular event was 0.72 to 0.81 between simvastatin & placebo group
What can you conclude here? There is …
Strong evidence to suggest that simvastatin reduces overall risk
We are 95% confident that the interval contains the “true” treatment effect
- Since the 95% CI for the relative risk does not contain 1 (which would represent no reduction in risk) and both ends of the interval are <1 suggesting a reduction
Absolute Risk Difference = -5.4% for simvastatin vs placebo
How would you interpret this?
The difference in the overall risk of a vascular event in the simvastatin group is 5.4 percent lower compared to placebo
In a group of 1,000 people, 54 fewer people with have a vascular event if they all switch from placebo to Simvastatin
1 less person will have a vascular event for approximately every 19 people that switch from placebo to simvastatin
Mean LDL Difference = -0.9 mmol/L, 95%CI = -1.06 to -0.74 mmol/L
What would you conclude based on these findings? There is …
Evidence to suggest simvastatin decreases LDL
Since both ends of the 95%CI are in the same direction. i.e. a reduction in LDL and zero (no benefit) is not included in the 95%CI, we have evidence to conclude that the results are consistent with an overall benefit of simvastatin in lowering LDL cholesterol levels after 3 years of treatment.
Calculate the absolute difference?
Value of no effect (no difference) = “0”
Differences have “UNITS” e.g.