Interpretations Flashcards
odds ratio interpretation
> assesses how a 1 unit increase in predictor variable affects likelihood (odds) of response
odds > 1 = likelihood increases as predictor increases
odds < 1 = likelihood decreases as predictor increases
Hazard ratio interpretation
> HR = 1: no difference in risk between groups
HR > 1: the event is x times more likely to occur to this group.
HR < 1: the event is x times less likely to occur in this group compared to….
per-protocol analysis definition
> only participants who fully adhered to study protocol are included in final analysis
dropouts are excluded
not censored; fully removed
intention to treat analysis definition
> includes all participants in analysis as per their initial group regardless of if they followed study protocol
maintains randomisation
reflects real-world treatment effects
data cleaning rules
> believe a prior question over a subsequent
believe and answered question over a missed one
Data checking steps
> check inclusion criteria
check variables for false/incorrect values
Difference between kruskal-wallis and mann-whitney tests for continuous data
> both for non-parametric
mann-whitney tends to be for differences between 2 groups (t-test)
kruskal-wallis is for differences in 3 or more groups (ANOVA)
Data type for logistic regression
binary outcome; to model with covariates
Data type for linear regression
continuous outcome; model with covariates
Data type for poisson regression
outcome is count data; model with covariates.
time-to-event analysis
> kaplan-meier
cox proportional hazards
logrank test
Analysis for categorical data
> chi - squared
fisher’s exact test
mann-whitney (if ordinal)
How does randomisation conclude causality?
because groups are balanced on any other differences and there is no subjective choice involved
Why are composite outcomes used?
Used to increase power when the event rates for the individual events are small.
Composite outcome meaning
there are two separate endpoints combined, both of which measure the same intervention effect.
randomisation ensures…
causality
hypotheses for proportion studies
proportions equal/proportions not the same
What is cox proportional hazards model testing?
null hypothesis that the hazard ratio is 1
what is logistic regression model testing?
null hypothesis that odds ratio = 1
Test to use for large sample, numerical data
Z test for difference of two means (same as t but for large samples)
