Section 1
Preview this deck
Factors leading to undersized trials
Front
| 5 | 0 | ||
| 4 | 0 | ||
| 3 | 0 | ||
| 2 | 0 | ||
| 1 | 0 |
Active users
0
All-time users
0
Favorites
0
Last updated
6 years ago
Date created
Mar 1, 2020
Cards (179)
Section 1
(50 cards)
Factors leading to undersized trials
Failure to make sample size calculation Sample size of convenience Avoidance of multicenter collaborations Inadequate finances Reward system of small trials Publish or perish mentality Lack of rigorous editorial policy of medical journals
What is outcome of interest in a trial 2 study
Efficacy
Adverse event
Any clinical event, sign, or symptom that goes in an unwanted direction. These are specifically defined in each protocol, including time period in which it is considered an adverse event.
Passive reporting of adverse events
When participant reports something themselves to site
Data Safety Monitoring Board (DSMB)
A board of members who are experts in the field but not a part of the study. May include ethicists, biostatisticians, epidemiologists
alpha in survival analysis
length of accrual period
If you are looking for a LARGE difference in BP, do you need a large or small population?
Small population
r in survival analysis
enrollment rate
What is type 2 error?
Null is not rejected when it is false. Wrongfully accepted.
Why might you use random selection when designing a clinical trial?
Ensure generalizability to the population of interest
Alpha
a P(Type 1 error) P(reject H0 | H0 is true) Often held constant at 0.05
Steps for sample size calculation in survival trial
Compare twp groups by detecting pre-specified log hazard ratio Beta in the cox model. 1. Calculate number of events needed 2. Calculate number of subjects needed
What is an example of Type 1 error?
If we concluded two drugs had different outcomes, when there was actually no difference
Active reporting of adverse events
Questionnaire based
Why do we need to meet a desired sample size?
Effect size
When should power and sample size be computed?
Before a study
Catchment area data
Recruiting in sector and will have broad demographic data. Should know who is in catchment area.
How do we estimate ß for survival analysis?
For rare events there are 2 approximates. 1. exp(ß) can be approximated by number of events in experimental arm /number of events in control arm 2. approximated by median survival time in control/ median survival time in experimental arm
d
Total number of events d=(Cα+Z γ)^2 / (pqß^2)
If you are looking for a SMALL difference in BP, do you need a small or large population?
Large population
________ trials lead to picking up more adverse events
Longer. Shorter studies are often too short or small to detect problems. Often dont have enough power to detect problems in subgroups
What can cause type 2 error?
small sample size
Do all adverse events need to be reported?
No. Should be monitored in study. Only serious adverse events need to be reported to funding agency.
Cα
Critical value. Cα= 1.96 when α=0.05
q
1-p
Composite endpoint
Combining endpoints with low event rates can increase a study's power, allowing enrollment of fewer patients or shorter follow-up. Individual outcomes within a composite endpoint should have similar value. For example, it would be inappropriate to combine death or severe morbidity with a more trivial outcome. It is important to examine results for the individual components of a composite. A statistically significant composite endpoint does not necessarily mean that results for all of the components also reached statistical significance. Improvement in a single component can be responsible for statistical significance of a composite endpoint. A negative outcome for one component can negate or dilute the positive outcome for another component.
Which is worse, Type 1 or Type 2 error?
Type 1
Null Hypotheses
µ=µ0 or µ1=µ2
Zγ
Z score associated with γ Zγ=1.28 when γ=0.90
Total number of subjects required in survival analysis
Specify 4 factors: 1. a 2.r 3.f 4. survival curve estimates for control group
Alternative hypothesis
µ1≠µ2 or µ≠µ0
p
proportion of subjects in experimental group
counterfactual
what the experimental group participants' responses would have been if they had not received the treatment
Why else might you want a large population?
Generalizability to other studies or populations. Heterogeneity.
Surrogate Endpoint Formal Definition
A surrogate endpoint of a clinical trial is a laboratory measurement or physical sign used as a substitute for a clinically meaningful endpoint .... Changes induced by a therapy on a surrogate endpoint are expected to reflect changes in a clinically meaningful endpoint.. a response variable for which a test of the null hypothesis of no relationship to the treatment groups under comparison is also a valid test of the corresponding null hypothesis based on the true endpoint"
Do all studies have the same sample size?
No. This depends on the desired outcome.
How would use of random selection enhance value of study results?
Make sure population does not influence bias bc being picked for certain characteristics. Gives more credibility. More external validity
Beta
P(Type 2 error)= P(do not reject null | alternative is true) Probability that null is accepted when it is false. Often range from 0.20 to 0.05 Usually around 0.10
exp(ß) approximation
#events in experimental arm/ # events in control arm
How does principle investigator handle adverse events?
determine if it is expected or unexpected. If it is related or unrelated
What is an example of type 2 error?
If it was concluded that two drugs had similar outcomes or no difference on average, when there was a difference in reality.
Power
1-beta P(reject null | alternative is true)
What are two reasons investigators might choose to give a placebo over no intervention
Avoid dropout from people who arent happy with allocation Double blind
When does Type 1 error occur?
When the null hypothesis is rejected when it is true. H0 is wrongly rejected
Mistakes to be avoided during sample size calculation
Unrealistic assumptions Failure to explore range of likely values Failure to compensate fro losses due to dropouts and noncompliance
What is sample size based off of?
Primary aim. Sometimes secondary aim as well.
γ
Gamma Study power 1-type 2 error
Why might you stratify randomization?
Eliminate residual confounding Distribution of outcomes similar among participants with same characteristics
f in survival analysis
additional follow up time between last enrollment and study closure
Sample size inflation
Should inflate sample size calculation by 20% because you will lose people during follow up
Section 2
(50 cards)
Survival curve formula
S(t)= P(T>t)
When do you measure outcome variables?
Interim results Final outcomes
Problems with GSD
Inflated sample size More difficult to show efficacy Access to interim analysis van bias future conduct of study
Why might participants not stay in a study?
Move Death Dont want to participate
What is penalty for taking interim looks
Larger sample size required Stricter p value requirements Controlled by which alpha spending function you choose
Cox proportional hazards model
multivariate survival analysis controlling for other factors. Allows calculation of hazard ratio. (and CI)
Alpha spending function controls for ____
Type 1 error introduced by multiple looks
What are issues with right censoring?
It is hard to estimate commonly used parameters such as the mean.
Deviations from planned interim analysis
Number and timing of looks should be prespecified Protocol should describe looks or requirements for generation
Left censoring
When we know that the event happened before a certain time but do not know the exact timing of the censoring. Recall questions, limits of detection
Observed data assumption
Only observe ~Ti=min(Ti,Ci) and ∆i=I(Ti<= Ci) Need to make independent censoring assumptoin T is independent of C conditioned on Z
Pitfalls in sample size calculations
Parameters are estimates- may differ from reality. Must have realistic assumptions - be conservative. Avoid samples of convenience
Flexible Group Sequential Design
Monitor accruing data at pre determined intervals and make important decisions concerning the future course of the study along the way
Null hypothesis in log rank test
Survival is equal in both groups
When does right censoring occur?
The study ends before event of interest happens People drop out of the study People die of other causes/competing risks
Why would a trial stop early?
Harm Benefit Lack of efficacy State of science changed
Typical patterns of harmful events
Usually slow Steady accumulation of harmful evidence
Control potential bias from interim analyses
External data monitoring committee External stat center preparing interim analysis results
Multiple looks
Looking at data multiple times through study. Smaller p value produced may lead to misleadingly significant results among trials stopped early for benefit.
Common alpha spending functions
Obrian flemming Pocock Haybittle peto boundary
Type 2 censoring
Subjects are followed for a period of the same length Censoring occurs when failure is observed on a prespecific proportion of participants Actual censoring time is unknown beforehand
Alternatives to post hoc power
Publishing confidence intervals with non significant results
What is the role of DSMB
Ensure participant safety Monitor adverse events Evaluate interim analyses Request information
Advantages of GSD
Early termination for efficacy (submit results early) Early intimation of inefficacy (stop early for futility, drop ineffective arm) Verification of design assumptions (variance, effect size, covariates, revise sample size to avoid underpowered study)
Uses of effect size
to make sure there is a relationship between variables when you have a large sample. Used in meta analyses.
ti
time in KM estimator
Type 3 censoring
Subjects enter study at different time Study ends at predetermined time Censoring occurs at end of study
di
number of events for KM estimator
What is used to compare survival of two or more groups
Use Log Rank Test
Multiple looks, premature termination, boosting sample size can increase _______
Type 1 and 2 errors
Yi
number in study at time ti for KM estimator
δ
Delta. Change in difference
Kaplan-Meier
A statistical technique used to analyze survival (life/death) data when there are censored observations (observations that are unknown because a subject has not been in the study long enough for the outcome to be observed). Also called the product limit estimator.
What regression model is used for time to event data
Cox Proportional Hazard Model
Type 1 Censoring
Every subject is followed for same length of time Censoring occurs at end of period Follow up length is predetermined
Survival curve
A curve that starts at 100% of the study population and shows the percentage of the population still surviving without event at successive times for as long as information is available. Common quantity of time to event outcomes
Interval censoring
When we know the timing happened in an interval but do not know exactly when the event happened
Effect size
A measure of the strength of relationship between two variables in a statistical population or a sample based estimate of that quantity. Magnitude of a relationship without any statement about whether this refelcts a true relationship in the population
Hazard rate
Instantaneous rate of experiencing event in a short interval after t
Event time
Amount of time from enrollment in a study until time of event of interest
alpha spending function
if clinical trial has more than one comparison, divide alpha by the number of comparisons
Right Censoring
When we do not know the exact time of the event but we know that it happened after a certain time (the censoring time)
Traditional Group Sequential Design
fix the sample size in advance and only perform one efficacy analysis after all subjects are enrolled and evaluated
Log-rank Test
test used to compare survival analysis curves between 2 or more groups.
What are problems with looking
If you take multiple looks at data, you are more likely to see spurious effects due to chance fluctuations in data
Ti
Event time for person i
Post hoc power
DO NOT DO Retrospective power or observed power. Usually test to see if tests were powerful enough after a study. This is redundant- if it were then your results would be significant and you wouldnt be testing
Criteria for evaluation of interim analyses
Determine whether interim data analyses across a predetermined boundary and become conclusive rather than suggestive to justify early stopping of study. When analyzed multiple times, must be determined as criteria for early stopping. Adjusted for multiple looks.
Ci
Censoring time for person i
2 types of group sequential design
traditional flexible
Section 3
(50 cards)
Tenets of recruiting
What works in one city might not work in another Maintain good relationships with PCPs Respect families Do not be overly aggressive Need multiple recruitment strategies
Hazard function
probability that a participant will die in the next instant given they are still alive now
What is the value of baseline measures for secondary study objectives
In placebo studies, can be used to conduct studies on natural history of diease for those without intervention
Problems in adequate accrual
Low budget MD not willing to refer patients Over estimation of condition prevalence Overly rigorous entry criteria
When is baseline measurement not valid
Hawthorne effect
What do you do when you need more participants
Recruit additional centers Loosen exclusion and inclusion criteria Increase time for recruitment Increase catchment area Accept lower sample size Change design Recycle participants (remeasure) Second chance at recruitment Stop study
How can using baseline data improve sensitivity of analysis
Usually less variation of measures of change than in absolute measures
What to consider in recruitment
Minorities Sample size Catchment area population What is different between people who do join and dont join
Fundamental point chapter 7
clinical trial should, ideally, have a double-blind design in order to limit potential problems of bias during data collection and assessment. In studies where such a design is impossible, other measures to reduce potential bias are advocated.
Recruitment methods
Mass media Dr Grand Rounds Partner with other centers Clinical trials databases Mail Phone Fliers Swag Social media ads
Manual of operations
MOP Details how hypotheses operationalized
Prevention of nonadherence
Make study short Make study simple Limit recruitment to compliers Run in period Dont take : addicts, those who live far or will move, uninformed
Recruitment monitoring graph
How many people we expect in study vs actually accruing It is cumulative Usually has a curve to show slow ramp up to accrual
Why is recruitment more difficult than planned?
Investigator overestimates ability to recruit Need to recruit minorities Need to honestly assess ability to get sample size
Assumptions of hazard, KM, Log rank, Cox
independent censoring
Log rank statistic
Used to compare the overall survival curve Reject null if |Z| > Z a/2
KM and Log rank assumptions
Do not make assumptions about shape of distribution (nonparametric)
Why are quality of life measures problematic
For certain diseases and conditions may exhibit floor or ceiling effects
When should baseline measurement occur?
Between randomization and intervention. Things may change over time
Interim monitoring
Allows us to estimate current information about δ from actual data of trial, reestimate sample size, and preserve power of the study
Baseline characteristics
Demographic, clinical, and other data collected for each participant at the beginning of the trial before the intervention is administered
Example of QOL measure
HUI
Cox regression model
a method that explores the effects of different covariates on hazards
Key data
Baseline characteristics Primary and secondary outcome measures
Two types of baseline measurement
1. Prior to consent (Not valid, cant use this data) 2. Post consent
Solutions for nonadherence
Supervision Client relationship Patient education Ethnocultural interventions
Main drawback of log rank test
It does not tell us which groups have a higher failure rate, only that some are different
Quality of life measures global
Assess overall quality of life of a participant. Used across studies and diseases
Quality of life disease specific measures
assess QOL associated with disease process specifically address issues of interest more sensitive to changes in that condition cannot be generalized
Considerations regarding PRO assessments in clinical trials
Population Design Frequency of assessment Length of questionnaire REcall Period Missing data Assessment method
What factors influence choice of instrument?
Is general or specific measure wanted Outcome of interest Method of administration Characteristics of population
Cox Regression Model assumptions
Assume about shape, but assumptions need to be checked covariates have same effect independent of time
Measuring indices of chronic disease
Which measure do you take out of multiple measures? Do you make them stop medication? Balance needs of study with risk of event
Why do we monitor accrual?
Make sure you are getting enough people needed to reach sample size target
Hawthorne effect
A change in a subject's behavior caused simply by the awareness of being studied
Who performs audits
An external group Checks records for validation of data entry. Ensure GCP followed.
Impact of nonadherence
differential leads to bias non differential leads to lack of power
Floor or ceiling effect
Are differences in health such that a global measure can detect them or do respondents consistently score low or high on the global measures Instruments are too sensitive or not sensitive enough to show a difference
Elements for a multi center trial
Planning committee Assessment of feasibility Coordinating centers Protocol Clear organizational structure Standards for data quality Monitoring Publication policies
QALY
Quality-adjusted life years; weigh each year of life by perceived quality from 0 to 1
Examples of quality of life measures
Short form 36 (SF36) Health Utilities index (HUI) EQ5d
What do you do if someone gives two different answers before and after consent?
Cant take them out of the study. Must do ITT analysis
Adherence
extent to which participant behavior corresponds with medical advice
What variables do you measure prior to consent
Inclusion and exclusion criteria
Why do people fail to adhere
Practicality side effects unwilling to change may not understand instructions culture change mind lack of support
PRO
Patient reported outcome
Independent censoring
the time participants are censored does not tell us anything about the failure time
What are baseline variables used for?
TO document comparability between groups with respect to known potential confounders. To adjust for differences in analysis should randomization fail to ensure comparability between arms with respect to known confounders. To compare study population to target populations
Good Clinical Practice (GCP)
A standard for the design, conduct, performance, monitoring, auditing, recording, analyses, and reporting of clinical trials that provides assurance that the data and reported results are credible and accurate, and that the rights, integrity, and confidentiality of trial subjects are protected.
regression toward the mean
the tendency for extreme or unusual scores to fall back (regress) toward their average.
Section 4
(29 cards)