Applied statistics and interpreting results Flashcards
Describe the types of data used/found in clinical trials
Binary = two possible responses (Yes/No)
Continuous = e.g. blood pressure, cholesterol
Time-to-event = time to particular outcome (death, pregnancy)
Describe what categorical data is
Consists of nominal (categories not ordered –> ethnic groups) and ordinal (Categories are ordered –> tumour stages)
Describe what nominal data is
Gives averages and means; consists of:
Discrete = only finite values possible (e.g. hospital admission)
Continuous = all values are theoretically possible (e.g. height)
Compare parametric testing and non-parametric testing
Parametric = assumes normal distribution, gives significant result more often
Non-parametric tests = comparison of rank order, not influenced by outliers much
What is a null hypothesis?
Population for which the sample was obtained, no difference exists in response to treatment A vs B
Nothing happens when experiment is completed
What is the alternate hypothesis?
Hypothesises that a difference exists in the response to treatment A compared to B
Describe a type I error (alpha)
Probability of falsely rejecting Ho (null) and detecting a stat sig difference when no difference exists in reality
False positive
Describe a type II error (beta)
Probability of falsely accepting the H0 (null) and not detecting stat sig diff when specified difference between grps exists
False negative
Describe what power (1-beta) is
Probability of correctly rejecting null (H0) and detecting a stat sig diff when specified difference between the groups in reality exists
The higher the power, the better
What do you use to measure the effect of binary data?
Risk difference (Absolute risk difference)
Relative Risk (RR)
Odds Ratio (OD)
Number Needed to Treat
Number Needed to Harm
Explain absolute risk (risk difference, risk inc, risk dec)
Absolute risk difference = diff between absolute risk of event in intervention and absolute risk of event in control group
Absolute risk reduction = treatment is effective and reduces risk of unwanted event
Absolute risk inc = treatment does not work, inc risk of event
What is the number needed to treat (NNT)?
Number of patients who needed to be treated to produce one additional successful outcome
Useful in comparing efficacy of treatments
What is the number needed to harm (NNH)?
Number of patients who need to be treated to produce one additional adverse event
Describe what relative risk is
Also known as risk ration, risk of an event (outcome) occurring
Where 1.0 is the middle number
If risk is death, you want the number to be low
If risk is cure, you want the number to be higher
What is the odds ration?
Probability of an event occurring over the risk of an event not occurring
What is selection bias?
Choosing participants in a way to underrepresent or overrepresent in a study
Does not represent the true population
What is performance bias?
Unequal care between experimental groups
What is attrition bias?
When participants who drop out of the study systematically differ from those who remain
What is detection bias?
Systematic difference between groups in how outcomes are determined
e.g. When the outcome of event is discovered using a different measure/method/criteria
What is the P value?
States the probability that the finding is due to chance or an actual difference.
Determines statistical significance
<0.05 = stat sig
> 0.05 = not stat sig
What is the confidence interval?
Gives estimate of precision of result; represents the range of values within which we are 95% confident that true population estimate lies
Says that, if trial was to be repeated, the findings will fall in that data range 95% of the time
What is a significant result in the context of CI?
CI is stat sig if it does not cross the null
1.0 for Relative Risk
0 for absolute risk reduction
What is the difference between statistical significance and clinical significance?
Stat sig = relates to size of effect and the 95% CI in relation to null hypothesis
Clinical sig = relates to size of the effect and the CI in relation to a minimum effect that would be considered clinically important
Describe what intention to treat is in RCT
It is a method for analysing results in a prospective randomised study where all participants who are randomised are included in the stat analysis
They are analysed according to the group that they were assigned
