Clinical trials design 1 Flashcards
The ideal clinical trial:
Determines the safety and efficacy of a new drug in humans
• Clinically meaningful
– Represents a meaningful advance in healthcare/ patient outcomes (changes survival or symptoms not just biomarker)
• Reliable
– Results can be trusted, reproducible if trial is repeated
• Valid
– Internal – observed differences can be correctly attributed to the intervention - the extent to which the observed results represent the truth in the population we are studying and, thus, are not due to methodological errors
– External – trial results can be generalized to the real-world population
Clinical trial design
clinically meaningul
reliability
internal validity
external validity
Clinically-meaningful
Research question, primary outcome measure
Reliability
Study design, conduct, standard operating procedures, measurement techniques
Internal validity
– Bias/systematic errors
– Confounding
– Random error
External validity
Participant diversity – inclusion and exclusion criteria
– Intervention – feasibility (in routine practice), acceptability (to patients and practitioners
Superiority trials
– Test whether new drug is better than comparator e.g. placebo or current best treatment
Most clinical trials
– Treatment advance, comparison with placebo or existing best practice
null hypothesis - treatment is not better or same as comparator
Equivalence trials
Test whether a new drug is the same as an existing treatment
• Clinical equivalence – clinical outcomes
• Bioequivalence – PK parameters e.g. blood concentration or receptor occupancy
Treatment change, comparison with original e.g.
• Change in drug delivery or manufacture e.g. modified release, new delivery method
• Generic drugs or biosimilars
null hypothesis - treatment is not similar to comparator
• Non-inferiority trials
null hypothesis - treatment is worse than comparator
Test whether a new treatment is no worse than an existing treatment
compare new drugs to current effective treatments
one sides test
tests for the possibilty of a relationship in ONE direction
provides more power to detect an effect
risk of missing an effecr in other direction
Two sided test
tests for the possibility of a relationship in both directions
1 sided or 2 sided?
superiority and equivalence 2 sided test
non inferiority 1 sided
Bias
Systematic distortion of the results of a clinical study away from the truth
– Caused by inadequacies in design, conduct or analysis of the trial or publication of its results
– Reduces validity of study results
selection bias
systematic differce in way subjects are erolled/treatments allocated
can be controlled]by randomisation
or concealment of treatment allocation
bias in study management
treatment groups are not handled equally
controlled by standardisation of study procedures
observer bias
participants/investigators influences by knowledge of assigned drug
controlled by blinding
introduced by exclusionsa fter randomisation
missing data due to participant dropouts
controlled by analysis design
publication bias
positive trial outcomes more likely to be published
controlled by trial registration
types of bias
selection bias in study management observer bias introduced by exclusions after randomisation publicaiton bias
Intention to treat vs per protocol analysis
ITT: Preserves randomisation minimises bias deals with dropouts mirrors real life practical might miss a real effect
Per protocol
determines maximum potential effectiveness of treatment
dont have to worry about drop outs
missing data
missing data: patients miss visit or outcomes not measured
types of missing data
missing completely at random (maybe questionnaire got lost - nothing to do with person)
missing at random (to do with the person)
missing not at random (due to drop outs)
impact of missing data
loss of power
introduction of bias
Options fo handling missing data
analyse complete cases only
PRO - if data are missing at random analysis can be unbiased
CON - introduces bias if missing data are non-random
loss of power
analyse all available data
pros - more power can produce unbiased analysis
last obervsation carried forward
pro - simple and easy to implement
cons - assumption that measurement stay the same is flawed - severe bias
multiple imputation
missing data replaced by value predicted form statistical model
pro - ubiased if appropriate model used
Confounder
predicts the outcome and is associated with treatment but is NOT a consequence of the treatment
Minimising confounders
Randomisation
– Everyone has an equal chance of being allocated to each group
• Balances confounders between groups
• Stratification
– Randomisation is within subgroups of key confounders
• Increases balance of key confounders between groups
• Minimisation
– Dynamic method that assigns participants to treatment groups in an order that
minimises overall imbalance between groups for a number of selected
confounding factors
Random error
Even studies with ideal study design, conducted to minimise bias and confounding can still give a result due to random error rather than a treatment effect
– Type 1 error
• Rejecting the null hypothesis when it is true, false positive result
– Type 2 error
• Failing to reject the null hypothesis when it is false, false negative result
Sample
a set of observations from population
population
complete set of possible observations
Sample size
Number of people to be enrolled in the trial or study
• Optimal sample size for clinical studies:
– Big enough to:
• Minimise type 2 error
• Represent the population – so results are generalisable
– Small enough to:
• Be feasible and achievable
• Minimise risk to participants, cost and waste
• Needs to be specified in the clinical trial protocol BEFORE the study starts
– Reduces bias in analysis and interpreting results
• e.g. bias of stopping study when p<0.05 has been reached!!
Factors determining the sample size
Expected measure of the primary outcome and variance in control group
2. Smallest treatment effect or benefit we are trying to detect
3. The significance level at which we will reject the null hypothesis
4. The study power with which we will find a significant difference if it
exists
5. The design of the study
– Parallel group, cross-over etc
– Superiority, equivalence, non-inferiority
6. The expected dropout rate of participants in the study
primary outcome measure
- Most important study outcome, addresses main aim of the study
- Used to decide on overall result of study/support regulatory action
- Selection is a balance between what is most important and what is most likely to demonstrate a change/benefit
• Using a primary outcome measure reduces risk of error
– Outcome of study determined by single analysis
• Reduces the risk of type 1 error from multiple comparisons
– Primary outcome used to estimate sample size
• Reduces the risk of type 2 error from too small study
Type 1 error
rejecteing null hypothesis when it is correct
false positive
alpha set to 0.05
type 2 error
failing to reject null hypothesis when it is false set by b 20% (0.2) study power (ability to find a significant result is one exists) is 100-b so 80%
Sample size - other things to consider
Ethical implications • Resources e.g. – Patients – Investigators – Centres • Time • Costs
