Topic 7: Analysis and Analysis Populations Flashcards
What tests/analysis methods do we use for comparing continuous data
T-test, linear regression
What tests/analysis methods do we use when comparing categorical data
chi square test, logistic regression, ordinal logistic regression, multinomial logistic regression
What tests/analysis methods do we use when comparing time to event data.
Kaplan Meir survival curve, log rank test, cox regression
What are two reasons you can have censoring
The patient has withdrawn from the trial or reached the end of the trial without witnessing the event.
What can you do if the assumptions of the analysis model are not met
Transformations, non-parametric methods
What do most basic methods of analysis assume
The observations are independent.
How is analysis done on paired data
Look at the difference between observations for each participant and the variability of differences, then do analysis on a single sample.
How does the randomisation method affect the analysis
Any factors used in randomisation should be adjusted for in analysis.
What happens if we don’t adjust for factors used in randomisation
We are ignoring the correlation between study groups that comes from balancing treatment groups using stratification. When the correlation is ignored, the standard errors of the treatment effect will be biased upwards, and we would say there is a bigger treatment effect than there actually is.
How is adjusted analysis conducted
Through regression where we have covariates for things we want to adjust for.
True/False: We should carry out statistical tests for differences between groups in terms of baseline characteristics
False - its not appropriate to do so
In cluster randomised trials, what is the issue in terms of comparability
Comparability is not guaranteed when randomising small numbers of clusters.
When should the analysis population be defined
Before analysis
Give two ways to handle missing data that can arise through an intention to treat analysis population
Complete case population (only use participants for which we have no missing data) imputation of missing data (replacing with reasonable substitute values).
What else do we need to clarify when saying we’re using per - protocol
What is counted as an adherence from protocol
What is the safety analysis population usually
All randomised participants to any treatment. Usually defined as those who have received at least one dose of the treatment.
What is a safety analysis population
Population of patients used to assess the safety of that treatment.
Why is the safety analysis population usually done according to the treatment actually received
To give a true reflection of the safety of that treatment.
What is the risk with using multiple subgroup analyses
Risk of false positive findings.
What is the purpose of sensitivity analysis
Assess the robustness of findings and conclusions. Robust findings are those that are consistent with those from the primary analysis, even when the trial design has been changed slightly.
What does sensitivity analysis assess the impact of
Impact or influence of key assumptions or variations on the overall conclusions.
What kind of things might be changed in a sensitivity analysis
Methods of analysis, definitions of outcomes, protocol deviations, missing data, outliers.
