Alfred Saturday Statistics Flashcards
Nominal vs ordinal data
Both categorical data but ordinal can be ordered/ranked (still have to use non-parametric analysis methods)
What is the explanatory variable
The independent variable/exposure (usually the x-axis)
Aim of multivariable analysis
To determine the unique contribution of various factors to a single event/outcome
Parametric vs non parametric data
Parametric data
- Based on the following assumptions
- Continuous data
- Normal distribution in the population
- Population and sample have the same variance and standard deviation
- Better powered than non-parametric
Non-parametric data
- Distribution free (less assumption), does not require numerical parameters
What factors affect power?
- Difference in outcome rates
- Level of significance (p value)
- SD of population (smaller SD higher power)
- Sample size
How to calculate the probability of a type I and type II error?
Type I
- Alpha (designated p value)
- ALso false positive rate = 1-specificity
Type II
- 1-Beta (power)
- Also false negative = 1-sensitivity
How to calculate odds
rate of occurance / rate of non-occurance
(c.f. probability where is the rate of occurance / sample number)
What is the median ratio in survival analysis?
Better explains clinical effect
Ratio of duration of survival at 50% event mark
Self controlled case series
- When is it used
- What is the aim
- When you can’t suitable controls
- Assessing time of event compared to time of exposure
- Aims to estimate the relative incidence within periods of hypothesised excess risk due to exposure versus incidence of adverse effects during all other times
Often used with vaccines and assessing event rate during high risk time after vaccine (e.g. stroke and MI rates transient increased risk after exposure to herpes zoster)
What is a matched block design?
It is where subjects are put into pairs with closely matching characteristics. It can also be done as a “repeated measure design” where the 2 arms are repeat in 1 subject
The aim is to reduce confounders
How to reduce information bias/selection bias
Blinding
How to reduce the effect of confounders?
Randomisation
What bias occur in cross-over studies when the likelihood of crossing over is different in the 2 groups?
Selection bias - When the sick patients are less likely to cross over into the control group
What is ascertainment bias
Sampling bias
E.g. patients who have Alzheimer’s and people who are on statins are generally associated with the frequency of visiting the doctor
What does selection/ascertainment bias affect?
Generalisability of the study
What is length time bias
What is lead time bias
Length time bias
- Mostly applying to screening tests - tumours that are slower growing has a longer phase for them to be detected in screening
Lead time bias
- That the perceived survival is longer because it is detected earlier through screening
What is the aim of intention to treat analysis and what effect does it have on the calculated treatment effect?
The aim is to minimise selection bias from crossover
Always underestimates any treatment effect
Reliability vs validity
Reliability is how consistent the results are and validity is how true it is
How does the mean compare to the median when a graph is negatively skewed?
mean < median
Rejecting the null hypothesis when it is true - what type of error is this?
What type of error is increased by post-hoc analysis if the p-value is not corrected?
Type I error
How to calculate relative risk?
RR = risk of bad outcome in intervention group / risk of bad outcome in control group
How to calculate relative risk reduction?
1 - relative risk
How to calculate absolute risk reduction?
Risk of bad outcome in control group - risk of bad outcome in intervention group
Does disease severity change relative risk or absolute risk?
Absolute risk
What type of observational study begins with the outcome?
Case control
What kind of bias occurs when a patient’s prognosis or degree of illness influences which group he/she is put into a study
Channeling bias
What does selection bias occur?
Generalisability of a study
What bias occurs when a lot of people leave a study?
Attrition bias
How to calculate post test probability from likelihood ratio?
Likelihood ratio can be used to calculate post test ODDS
Post-test odds = pre-test odds x LR+
How to detect homogenity in forrest box plots?
If the whiskers of the individual studies overlap
or I^2 < 25%
