Lecture 4 Woensdag
ICH – E9
-Statistical principles, not procedures or methods
-Targeted audience: Trial statistician
-Primary aim Late phase, mainly confirmatory trials of efficacy
-Also for safety PD and PK variables
-Also for integrated data across trials
-Also relevant for earlier phase, exploratory trials
Direction
-Minimising bias
If you have 10 sites, 1 site has reported way less AE;
-Are they biased?
-Do they have healthier population
-> you can’t just tell the PI you underreport comparing to other sites, you can create bias
with this
In CT bias should be avoided, we need to avoid the perception of bias
Do this by:
-Maximising precision
-Evaluate robustness of results and conclusion
-Guidance refers to using frequentist methods
-Bayesian and other approaches may be considered
Statistical bias: The systematic tendency of any factors associated with the design, conduct,
analysis and evaluation of the results of a clinical trial to make the estimate of a treatment
effect deviate from its true value
Operational bias: Bias introduced through deviations in conduct is referred to as operational
bias.
Sources of bias
-Design of the trial; assignment of treatment such that subjects at lower risks are
systematically assigned to one treatment
-Arising during the conduct and analysis; protocol violations or exclusion of subjects from
analysis based upon knowledge of subject outcomes
The moment a patient becomes unblended before DBL you have to eliminate data, it is not
used for the analysis. However, the patient was treated and may or not have had an AE,
therefore the safety data will be kept
Robustness: the sensitivity of the overall conclusions to various limitations of the data,
assumptions, and analytic approaches to data analysis. Treatment effect and primary
conclusions not substantially affected when analyses are carried out based on alternative
assumptions or analytic approaches
1
, Bayesian or frequentist
Bayesian approach
Frequentist method
-Most used in CT
-With CI and p-values
-When rejecting a hypothesis there is a 10-20% chance the hypothesis was actually true
Design factors
-Randomisation
-Blinding
-Trial design
-Control group
Randomisation
=Process of assigning clinical trial participants to treatment groups with the element of
chance
-Avoids selection bias
-Similar treatment groups
-Try to ensure only one factor is different between two or more treatment groups
-Not possible
-Any difference in outcome of randomisation is due to chance
Random allocation
-Known chance of receiving a treatment
-Cannot predict the treatment to be given
-Codes are prepared before trial
-When blinded codes not broken before end of the trial
-Only in case of emergency
-Only if outcome is needed for medical decisions (which treatment to give depending
on what medicine they received; unblind to eliminate immediate hazard)
When you suspect that in one group the risk is higher a DSMB group is hired to do analysis
before DBL. It is unethical to continue when two groups are to different. When the active
treatment is the worse group it is reason to discontinue the trial.
When the active treatment is way better -> stop trial as it is going now. Discontinue the
control group. Continue with active treatment, give those from control group the chance to
switch to active treatment
These results will come forward out of the analysis of the DSMB
DSMB = Data and safety monitoring board
Types of randomisation
-Simple randomisation
-Blocked randomisation
-Unequal randomisation
2
ICH – E9
-Statistical principles, not procedures or methods
-Targeted audience: Trial statistician
-Primary aim Late phase, mainly confirmatory trials of efficacy
-Also for safety PD and PK variables
-Also for integrated data across trials
-Also relevant for earlier phase, exploratory trials
Direction
-Minimising bias
If you have 10 sites, 1 site has reported way less AE;
-Are they biased?
-Do they have healthier population
-> you can’t just tell the PI you underreport comparing to other sites, you can create bias
with this
In CT bias should be avoided, we need to avoid the perception of bias
Do this by:
-Maximising precision
-Evaluate robustness of results and conclusion
-Guidance refers to using frequentist methods
-Bayesian and other approaches may be considered
Statistical bias: The systematic tendency of any factors associated with the design, conduct,
analysis and evaluation of the results of a clinical trial to make the estimate of a treatment
effect deviate from its true value
Operational bias: Bias introduced through deviations in conduct is referred to as operational
bias.
Sources of bias
-Design of the trial; assignment of treatment such that subjects at lower risks are
systematically assigned to one treatment
-Arising during the conduct and analysis; protocol violations or exclusion of subjects from
analysis based upon knowledge of subject outcomes
The moment a patient becomes unblended before DBL you have to eliminate data, it is not
used for the analysis. However, the patient was treated and may or not have had an AE,
therefore the safety data will be kept
Robustness: the sensitivity of the overall conclusions to various limitations of the data,
assumptions, and analytic approaches to data analysis. Treatment effect and primary
conclusions not substantially affected when analyses are carried out based on alternative
assumptions or analytic approaches
1
, Bayesian or frequentist
Bayesian approach
Frequentist method
-Most used in CT
-With CI and p-values
-When rejecting a hypothesis there is a 10-20% chance the hypothesis was actually true
Design factors
-Randomisation
-Blinding
-Trial design
-Control group
Randomisation
=Process of assigning clinical trial participants to treatment groups with the element of
chance
-Avoids selection bias
-Similar treatment groups
-Try to ensure only one factor is different between two or more treatment groups
-Not possible
-Any difference in outcome of randomisation is due to chance
Random allocation
-Known chance of receiving a treatment
-Cannot predict the treatment to be given
-Codes are prepared before trial
-When blinded codes not broken before end of the trial
-Only in case of emergency
-Only if outcome is needed for medical decisions (which treatment to give depending
on what medicine they received; unblind to eliminate immediate hazard)
When you suspect that in one group the risk is higher a DSMB group is hired to do analysis
before DBL. It is unethical to continue when two groups are to different. When the active
treatment is the worse group it is reason to discontinue the trial.
When the active treatment is way better -> stop trial as it is going now. Discontinue the
control group. Continue with active treatment, give those from control group the chance to
switch to active treatment
These results will come forward out of the analysis of the DSMB
DSMB = Data and safety monitoring board
Types of randomisation
-Simple randomisation
-Blocked randomisation
-Unequal randomisation
2
