This preview shows page 1. Sign up to view the full content.
Unformatted text preview: CHL 5225 H The Language of Clinical Trials USEFUL TERMINOLOGY
Treatment Arm (Group)
• the treatment on which the patient is being evaluated
Intervention Arm (Treatment)
• the new or experimental treatment
Control Arm (Standard)
• the standard or current treatment, sometimes no treatment
Placebo
• sham new or experimental treatment CHL 5225 H The Language of Clinical Trials USEFUL TERMINOLOGY
Multicentre
Multi centre trial
• patients recruited from more than one clinical site
Multinational trial
• patients recruited from more than one country
Masking (Blinding)
• singleblind: patients are masked to treatment arm
• d bl bli d patients and person evaluating th patients are masked
doubleblind: ti t
d
l ti the ti t
k d
• tripleblind: patients, evaluators and clinicians masked
• often evaluators are the clinicians, in which case doubleblind = tripleblind CHL 5225 H The Language of Clinical Trials PHASES
Phase I – Dose finding studies
Usually done on healthy individuals
Often contracted out to Contract Research Organizations (CROs)
O’Quiqley t l Stat i M d
O’Q i l et al. St t in Med 1991 10 1647 1664
10:16471664.
Biometrics 1990 46: 3348.
Biometrics 1996 52: 673684.
Storer. Biometrics 1989 45: 925937.
Piantadosi and Liu. Stat in Med 1996 15:16051618. CHL 5225 H The Language of Clinical Trials PHASES
Phase II – Safety and efficacy studies
• Usually done as a single arm
• Common in oncology
Herson (1984). Chapter 15 in Buyse, Staquet, Sylvester.
Cancer Clinical Trials. Oxford: Oxford University Press.
Simon. Controlled Clinical Trials 1989 10: 110.
Thall and Simon. Controlled Clinical Trials 1994 15: 463481. CHL 5225 H The Language of Clinical Trials
PHASES
Phase III – Comparative studies (Randomized Controlled Trials (RCTs)
Patients randomized between two or more treatment arms
Early Phase III – Efficacy (Explanatory) Trials
• Smaller sample size
• Often single country
• Smaller number of clinical sites
• Often placebo controlled (i.e. intervention compared to nothing)
• Selected patients (those most likely to benefit from new intervention)
• Subclinical outcomes (those most likely to be sensitive to change)
(
y
g )
• Tight protocol control (compliance, followup)
• Short time horizon
• Positive results lead to late phase III trial
• Answers question “Can the new intervention work?” (biological question) CHL 5225 H The Language of Clinical Trials
PHASES
Phase III – Comparative studies (Randomized Controlled Trials (RCTs)
Patients randomized to two or more arms
Late Phase III – Effectiveness (Pragmatic, Management) Trials
(Pragmatic
• Larger sample size
• Often multinational
• Large number of clinical sites
• New intervention usually compared to best alternative (standard care)
• Unselected patients
• Clinical outcomes (those most important to p
(
p
patients and clinicians)
)
• Loose protocol control
• Longer time horizon
• Positive results lead to adoption of new intervention
• Answers question “Will the new intervention work?” (Clinical question) CHL 5225 H The Language of Clinical Trials PHASES
Example: new drug from reducing high blood pressure
Early Phase III – Efficacy (Explanatory) Trials
• 200 or 300 patients, 1 or 2 clinical sites, one country
• Placebo controlled
• Oth
Otherwise health patients, not previously t t d
i h lth ti t
t
i
l treated
• Demonstrate compliance in a “runin”
• Measure blood pressure as primary outcome
• Monthly visits to clinical site
• Include in the analysis only those who took 80% of drug
• One year time horizon CHL 5225 H The Language of Clinical Trials PHASES
Example: new drug for controlling high blood pressure
Late Phase III – Effectiveness (Pragmatic, Management) Trials
(Pragmatic
• Large multinational trial with numerous clinical sites
• Intervention compared to best active comparator
• All patients with hi h bl d pressure
ti t ith high blood
• Use heart attacks, strokes and death as primary outcomes
• Might include quality of life (QofL) and economic (cost) outcomes
• Patients visit clinical site every 6 months
• Include all randomized patients in analysis
• Five to ten year time horizon CHL 5225 H The Language of Clinical Trials PHASES
Phase IV – Postmarketing (RCTs)
• Done by drug companies after drug has been approved
• Large sample size
g
p
• Sometimes multinational
• Usually large number of clinical sites
• Intervention compared to best available alternative
• Unselected patients
• Clinical and safety outcomes
• QoL and economic outcomes
•L
Loose protocol control
t
l
t l
• Longer time horizon CHL 5225 H The Language of Clinical Trials DESIGNS OF PHASE III TRIALS
Parallel Groups
• Most common
• Patients individually randomized between two or more arms
• Simple or controlled randomization
Crossover
• Patients individually randomized to two or more
sequences of treatments
• i.e. A then B or B then A
• Patients act as their own control
• Best for chronic diseases and “short” acting interventions
• Used mostly for early phase III CHL 5225 H The Language of Clinical Trials DESIGNS OF PHASE III TRIALS
Cluster Randomized
•G
Groups of patients are randomized b t
f ti t
d i d between t
two or more arms
• Groups can be defined by (i) family practice, (2) town, (3) hospital
• Simple randomization
• Paired randomization
• Stratified randomization
• Usually late phase III Beck, C. A. et al. JAMA 2005;294:309317. CHL 5225 H The Language of Clinical Trials DESIGNS OF PHASE III TRIALS
Factorial
• Patients randomized “twice”: A vs PlaceboA and B vs PlaceboB
• Four groups of patients A/B, A/PlaceboB, PlaceboA/B
and PlaceboA/PlaceboB
• Two trials for the price of one
• Allows for the examination of the interaction between A and B
• Usually late phase III Little, P. et al. JAMA 2005;293:30293035. CHL 5225 H The Language of Clinical Trials METHODS OF RANDOMIZATION
Parallel Groups (patients enrolled sequentially)
• Simple randomization
• each patient has same probability of being allocated to a
particular arm
• advantages: simplest analysis; concealment guaranteed, even
for
f unmasked t i l
k d trials
• disadvantages: clinician don’t sleep well; imbalance with respect
to important prognostic factors possible CHL 5225 H The Language of Clinical Trials
METHODS OF RANDOMIZATION
Parallel Groups (patients enrolled sequentially)
• Blocked randomization (block size = b)
• after k*b patients have been allocated there will be an equal
number of patient allocated to each arm, k = 1, 2, . . .
• b = (no. of arms)*(blocking factor), blocking factor = 1, 2, . . .
• example: 2 arms, blocking factor = 2, block size = 4
• possible blocks: (ABAB) (BABA) (AABB) (BBAA) (ABBA) (BAAB)
(ABAB),(BABA),(AABB),(BBAA),(ABBA),(BAAB)
• allocation list created by randomly sampling blocks with replacement
• advantages: ensures the same number of patients on each arm
• disadvantages: for unmasked trial concealment not guaranteed;
clinician don’t sleep well; imbalance with respect to important
prognostic factors possible CHL 5225 H The Language of Clinical Trials
METHODS OF RANDOMIZATION
Parallel Groups (patients enrolled sequentially)
• Stratified blocked randomization (block size = b)
• patients stratified by important prognostic factor and a blocked
allocation list is created for each stratum
• example: suppose age is expected to affect outcome
• create 3 strata: 1825; 2645; 4665
• each age stratum has their own allocation list
• most multicentre RCTs stratify randomization by clinical site
• often use more than one block size randomly
• advantages: ensures each each arm will have the same age distribution
• disadvantages: for unmasked trial concealment not guaranteed; with
several factors the number strata become too large; analysis is more
complicated, i should i l d stratification f t as covariate
li t d i.e. h ld include t tifi ti factor
i t CHL 5225 H The Language of Clinical Trials
METHODS OF RANDOMIZATION
Parallel Groups (patients enrolled sequentially)
• Adaptive randomization
• next patient is allocated to the arm that minimizes the imbalance with
respect to prognostic factors according to some loss function
Loss function : L w age ( A B ) w sex ( A B )
where i average age on treatment arm i
g g
and i proportion of males on treatment arm i
L A LB allocate patient to A
L A LB allocate patient to B • advantages: handles large number of prognostic factors
• disadvantages: complicated; appropriate analysis not known CHL 5225 H The Language of Clinical Trials
N=100 N = 200 N = 500 N = 1000 N = 2000 0.4 Pr
robability 0.3 0.2 0.1 0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 Variance Inflation
V i
I fl ti CHL 5225 H The Language of Clinical Trials INTENT TO TREAT PRINCIPLE
Patients are analyzed in the arm to which they were randomized,
regardless of what treatment they received and whether or not
g
y
they were eligible to be randomized
This is the analysis that accurately assess the impact of adopting
the intervention
Often used for “Effectiveness” trials
Meant to minimize “selection” bias and improve credibility CHL 5225 H The Language of Clinical Trials DRUG COMPANY FUDGE ON THE INTENT TO TREAT PRINCIPLE
Intent to Treat Population (Sample)
All eligible patients
Safety Population (Sample)
All eligible patients who had a least one dose of drug
Efficacy Population (Sample)
All eligible patients who had 80% of the drug they were allocated to CHL 5225 H The Language of Clinical Trials TYPES OF OUTCOME VARIABLES
Time to Event
• Very common
• Time to death, relapse, stroke, symptom relief, discharge from hospital
• Usually characterized by censoring
• l t t followup
lost to f ll
• administrative (analysis done before everyone has an event)
• Survival curves to describe data
• Logrank test to compare treatment arms
• Cox proportional hazards for covariate (prognostic factor) adjustment Fig 1. KaplanMeier plot of time to progression in the
intenttotreat population in each treatment group Jones, S.E. et al. J Clin Oncol; 23:55425551 2005 Fig 2. KaplanMeier plot of survival in the
intenttotreat population in each treatment group Jones, S.E. et al. J Clin Oncol; 23:55425551 2005 KaplanMeier Curve of Survival Free From Atrial Fibrillation CHL 5225 H The Language of Clinical Trials TYPES OF OUTCOME VARIABLES
Time to Event
Treatment contrasts of interest
• ratio of median survival (from survival curves)
• hazard ratio (from Cox proportional hazards)
• diff
difference in mean survival i more clinically relevant
i
i l is
li i ll
l
t
• must assume a parametric form for the survival distribution
• not estimable from Cox proportional hazards
• survival curves only provide restricted means, unadjusted for
covariates
• Inverse probability weighting provides restricted means while
adjusting f covariates
dj ti for
i t CHL 5225 H The Language of Clinical Trials
TYPES OF OUTCOME VARIABLES
Time to Event—Difference in mean survival
1 / (difference in mean survival) = Number Needed to Treat (NNT)
NNT is the number of patients y need to treat on the new
p
you
treatment (T), rather than the standard (S), to expect to realize one
extra unit of effectiveness (in this case one year)
Suppose mean survival on S = 5 years
and mean survival on T = 5.25 years
NNT = 1 / 0.25 = 4
Total expected survival time for 4 patients on T = 4*5.25 = 21
Total expected survival time for 4 patients on S = 4*5 = 20
45 CHL 5225 H The Language of Clinical Trials TYPES OF OUTCOME VARIABLES
Binary
• Very common
• Patient died, relapsed had stroke became free of symptoms
died relapsed,
stroke,
(within a specific time period)
• Often a composite outcome (e.g. a patient would be considered
a failure if he or she died OR had a stroke OR had a
major bleeding event
• Often no censoring
• Results presented in 2 by 2 contingency table
• Fisher exact test or Chisquared test to compare treatment arms
• Logistic regression for covariate (prognostic factor) adjustment
• Life table methods (survival tables) for censored data OneYear Followup Results by Treatment Group Wazni, O. M. al.
Wazni O M et al JAMA 2005;293:2634 2640
2005;293:26342640. 28/88=32% 16/87=18%
15/88=17%
11/87=13% Valgimigli, M. et al. JAMA 2005;293:21092117. CHL 5225 H The Language of Clinical Trials
TYPES OF OUTCOME VARIABLES
Binary—Parameters of Interest
Let pT and pS be the probability of the event occurring in a patient that is
randomized to Treatment and Standard, respectively
Relative Risk (RR) = pT / pS
If RR = 0.8, then patients on T have 20% lower chance of experiencing event Odds Ratio (OR) 1 pS
pT (1 pT ) RR
pS (1 pS )
1 pT 1 pS
If event is “rare”, say pT and pS <= 0.1, then
is close to 1 and OR is
1 pT
close to RR
OR has better statistical properties
Logistic regression, used for covariate adjustment, yields ORs CHL 5225 H The Language of Clinical Trials
TYPES OF OUTCOME VARIABLES
Binary—Parameters of Interest
Risk Difference (RD) = pS  pT Number Needed to Treat (NNT) = 1 / RD NNT = the number of patients that need to be treated with T
rather than S to expect to prevent one event
If pS = 0.1 and pT = 0.05, then NNT = 1/0.05 = 20
Expected number of event from treating 20 patients with S = 0.1*20 = 2
Expected number of event from treating 20 patients with T = 0.05*20 = 1
0.05 20 CHL 5225 H The Language of Clinical Trials TYPES OF OUTCOME VARIABLES
Continuous
• Less common
• blood pressure, symptom scores, quality of life (QoL)
• Usually characterized by repeated measurements
• Oft characterized by missing values
Often h
t i db
i i
l
• Repeated measures ANOVA to compare treatment arms
• GLM for for covariate (prognostic factor) adjustment
• Area under the curve (AUC) methods for QoL yields
qualityadjusted survival time Shear, K. et al. JAMA 2005;293:26012608. CHL 5225 H The Language of Clinical Trials
Continuous O t
C ti
Outcome
ANOCOVA y y T e
post,i
post i
0 1 pre i T i i
pre,i where Ti is dummy indicator for treatment
i.e. Ti = I(patient i on new treatment arm)
Ttest on difference y
y T e
post,i pre,i
0 T i i
Ttest on postrandomization observation y T e
post,i
0 T i i V y
y 2(1 )V y
post,i pre,i
post,i CHL 5225 H The Language of Clinical Trials Patient followup–Compliance
“Dropouts”
“D
t ”
• referring to patients who do not complete allocated treatment
• often used only for intervention arm (drug company)
• poor use of terminology
• “noncompleters” is a more accurate term
“Dropins”
• referring to patients allocated to control who receive intervention
• poor use of terminology
• “Crossovers” is a more accurate term
Much better to characterize the treatments patients receive using
proper statistics CHL 5225 H The Language of Clinical Trials Patient followup–Compliance
Consider a trial with a 8 doses Doses
Patients
Proportion 3 4 5 6 7 8 200 180 164 150 140 100 1 0.9 0.82 0.75 0.7 0.5 CHL 5225 H The Language of Clinical Trials Journal Reporting of Randomized Clinical Trials
p
g Example:
Tariot PN et al.
Memantine Treatment in Patients With Moderate to Severe Alzheimer
y
g
Disease Already Receiving Donepezil
JAMA 2004; 291:317314 Study Flow Tariot, P. N. et al. JAMA 2004;291:317324. SIB and ADCSADL19 by Visit (Observed Case) and at End Point (LOCF) CHL 5225 H The Language of Clinical Trials
Sample Size and Power Curves
p Probabil of rejec
lity
cting null
hypothesis
s 1 1 0.5 0 0.5 True treatment difference CHL 5225 H The Language of Clinical Trials
Sample Size and Power Curves Power curves will pass through (0,) because we’re testing the hypothesis
H: = 0 vs. A: > 0 at the level (i.e. probability of reject H when it’s true is ) Steepness of the power curve will increase with sample size CHL 5225 H The Language of Clinical Trials
Multiplicity and Controlling Type I Error (i.e. false positives) Multiple arms
Multiple outcomes (endpoints)
Multiple subgroups
Interim analyses CHL 5225 H The Language of Clinical Trials
Exercise Search through recent issues of
Journal of th A
J
l f the American M di l A
i
Medical Association
i ti
New England Journal of Medicine
Lancet
Journal of Clinical Oncology
and located one or two reports of randomized clinical trials
Read the articles and identify some of the issues we discussed today
y
y ...
View
Full
Document
This note was uploaded on 02/23/2012 for the course CHL 5225H taught by Professor Andrewwillian during the Spring '12 term at University of Toronto Toronto.
 Spring '12
 AndrewWillian

Click to edit the document details