-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 1
Bios 6648: Design & conduct of clinical researchSection 1 - Specifying the study setting and objectives
1. Specifying the study setting and objectives
1.0 BackgroundI Where will we end up?:
(a) The treatment indication(b) Inference upon trial completion
I The scientific method
1.1 Defining the study population1.2 Defining the study question
(a) Defining the interventions: What is the treatment?(b) Phase I-IV clinical trials(c) Statistical structure of the outcome space(d) 1-sided versus 2-sided questions
1.3 Case study (Rocket-AF trial)
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 2
1.0 Background
Where will we end up?(a) The treatment indication
I The FDA approves a new drug for a specific “indication."I The study setting/objectives must inform the intended
indication (or vice versa).I Components of an indication include:
I TreatmentI Disease or conditionI Patient populationI Therapeutic objective
I See my written notes for section 1 for further discussionI Example: Exercise 1.
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 3
1.0 Background
(a) Specifying the treatment indication
Notes (See also Section 1 written notes):I Disease:
I Diseases can be defined by symptoms (COPD), causalagents (meningococcal meningitis), or treatment (MDRtuberculosis).
I The definition of a disease often changes with factorsunrelated to the disease (e.g., new treatments or a newcategorization of symptoms)
I Population of patients:I The target patient population is similarly dynamic.I Example: better diagnostic tools may detect cancer at
earlier stages or make it easier to detect later-stagedisease.
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 4
1.0 Background
(a) Specifying the treatment indication (con’t)
I Treatment:I The way in which a treatment is delivered may change.I New formulations may allow oral instead of IV delivery,
which might extend the use to other populations or otherforms of the disease.
I Ancillary treatments (standard of care) is always changing.I Desired outcome
I Primary clinical outcomes versus surrogate outcomes(Vioxx; mammography; colon cancer screening)
I Unanticipated or anticipated beneficial (sildenafil citrate) orharmful (rosiglitazone) effects.
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 5
1.0 Background
Where will we end up?(b) How will we summarize trial results?
I Inference upon trial completion(i) What is statistical inference?(ii) Four required elements(iii) Properties of estimators(iv) Interpretation of interval estimates
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 6
1.0 Background
Inference upon trial completion(i) Why review statistical foundations?
I We are discussing the study setting and objectives.I As a scientific experiment, the results of a clinical trial are
used to rule out (or rule in) hypotheses about treatmenteffects. The standards for rejecting (or acceptinghypotheses) are based on statistical criteria.
I We need a basic understanding of statistical foundationsin order to discuss the scientific setting and the role ofuncertainty.
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 7
1.0 Background
Inference upon trial completion(i) What is statistical inference?
Underlying Population
Sample
Statistics
Inference
θθ denotes unknown center
about θθ
Sample summary measure: θθ̂
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 8
1.0 Background
Inference upon trial completion(ii) Four required elements of statistical inference
We use θ̂ (observed trial result) to estimate the true underlyingvalue θ
1. Point estimate: θ̂ is the “best" estimate of θ.2. Interval estimate: Values of θ that are consistent with the
trial results.3. Expression of uncertainty (p-value): To what degree is a
particular hypothesis (the “null" hypothesis) consistent withthe observed trial results?
4. Decision: Based on the above measures, what decisionshould be reached about the use of a new therapy?
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 9
1.0 Background
Inference upon trial completion(ii) Four required elements of statistical inference
I Example:CHEST trial: Ghofrani,et.al. NEJM (2013); 369: 319-29:Riociguat for the Treatment of Chronic ThromboembolicPulmonary Hypertension.
I Trial: Randomized double-blind placebo controlled trial inpatients with inoperable CTEPH.
I Results:“...By week 16, the 6-minute walk distance ( had increasedby a mean of 39 m in the riociguat group, as compared witha mean decrease of 6 m in the placebo group(least-squares mean difference, 46 m; 95% confidenceinterval [CI], 25 to 67; P
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 10
1.0 Background
Inference upon trial completion(ii) Four required elements of statistical inference
Example (con’t)I Setting:
I Primary endpoint: 16-week change in 6-minute walkdistance (6MWD)
I Summary of outcome: mean change denoted byθ1 (riociguat) and θ0 (placebo)
I Measure of treatment effect: difference in 6mwd:θ = θ1 − θ0.
I Observed effect:I Observed summary outcomes: θ̂1 = 39m; θ̂0 =-6mI Observed treatment effect: θ̂ = θ̂1 − θ̂0 = 46m
(by least-squares regression).I Inference:
I Point estimate: 46mI Interval estimate: 25m to 67mI Uncertainty: p < 0.001I Decision: to use or not to use?
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 11
1.0 Background
Inference upon trial completion(iii) Properties of estimators
Desirable properties of:I Point estimate
I Unbiased and consistent: the long-run average of θ̂ is veryclose to θ
I Small variance (Uniform Minimum Variance UnbiasedEstimator)
I Interval estimateI Correct coverage probability (e.g., 95% of all 95%
confidence interval include θ).I As narrow as possible while maintaining the correct
coverage probability.I P-value
I Correct sizeI Decision
I Decision criteria maintain the appropriate type I statisticalerror rate.
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 12
(iv) Interpretation of an interval estimator
What is the interpretation of a 95% confidence interval?
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 13
(iv) Interpretation of an interval estimator
What’s wrong with the following picture?
θθL θθ̂obs θθu
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 14
(iv) Interpretation of an interval estimator
Proper depiction of an interval estimator:
θθL θθ̂obs θθu
If θ < θL or if θ > θU then the observed result θ̂ = θ̂obs would beunusual.
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 15
(iv) Interpretation of an interval estimator
I The scientific objective is to identify the hypotheses thathave (or have not) been ruled out by the trial’s results.
I Let U(θref |θ̂obs) represent a statistical measure of theconsistency between the trial’s result θ̂ = θ̂obs and thehypothesis θ = θref .
I By usual frequentist criteria this measure is equal to thesmaller of:
P(θ̂ ≥ θ̂obs|θ = θref )P(θ̂ ≤ θ̂obs|θ = θref )
I Reject the hypothesis θ = θref when U(θref |θ̂obs) is small;specifically when:
U(θref |θ̂obs) <α
2
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 16
(iv) Interpretation of an interval estimator
We seek the values of θ that cannot be rejected; specifically:I Find the set of θref such that U(θref |θ̂obs) ≥ α/2 usingα = 0.05.
I If θ̂ ∼ N (θ,V ) then the non-rejection region is given by[θL, θU ] where
θL = θ̂obs − 1.96√
VθU = θ̂obs + 1.96
√V
I (See graph above)
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 17
(iv) Interpretation of an interval estimator
What assumptions are necessary to assure that the aboveinterval has the correct properties?
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 18
(iv) Interpretation of an interval estimator
What assumptions are necessary to assure that the aboveinterval has the correct properties?
I No assumption about the distribution of the individual dataelements is necessary.
I The estimated effect θ̂ must follow a Normal distribution:I Central limit theorem assures θ is Normally distributed as
long as the sample size is not too small.I For interpretation as a non-rejection region, we must know
the mean-variance relationship.
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 19
I A note on mean-variance relationships:
θθL θθ̂obs θθu
I With a mean-variance relationship the confidence intervalcan have the correct coverage probability, but may not be anon-rejection region.
I Interventions often change both the mean and the variance.I We will return to this issue in chapter 7.
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 20
(iv) Interpretation of an interval estimator
In summary...I Scientific decisions must consider the magnitude of the
effect (point estimate) and the hypotheses that remainviable based on the trial’s results (the interval estimate).
I I will appeal to these considerations as I describe thescientific setting.
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 21
(iv) Interpretation of an interval estimator
Example (CHEST trial CI: 25m to 67m):I 16-week improvement in 6MWD with riociguat was 46m
better than the improvement with placebo. The variabilityin the results allows us to rule out (with 95% confidence)improvements that are more than 67 meters greater withriociguat or less than 25 meters greater with riociguat.
I Note: the following is incorrect:I There is a 95% chance (or probability) that the true
underlying difference is between 25 meters and 67 meters.I Note: the following is sometimes accepted, but
misleading:I We are 95% confident that the true underlying difference is
between 25 meters and 67 meters.
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 22
1.0 Background
The scientific method
I The scientific method is an iterative process of posing andevaluating hypotheses using carefully designedexperiments.
I A clinical trial is an experiment and should be built oncarefully-framed hypotheses:
I What is the treatment?I What is θ (the measure of treatment effect)?I What are important differences?I What differences support recommending use of a new
treatment?I The trial must be designed to be informative relative to the
hypotheses (the scientist game)I Upon completion the range of viable hypotheses that
remain is determined by the experimental results
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 23
1.0 Background
The scientific method: The scientist game
Assignment: Try the scientist game:htpp://www.emersonstatistics.com/ScientistGame
I Careful consideration of what you want to know upon trialcompletion is essential. The ‘obvious’ choice is often notthe best choice.
I The scientist game is illustrative of the scientificimportance of all aspects of the design including:
I Specification of the treatmentI Selection and definition of the outcome(s)I Choice of control groupI Definition of design hypothesesI Statistical standard for evidenceI Choice of sample size
What was your choice?
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 24
1.0 Background
The scientific method: The scientist game
The scientist game illustrates how it is possible to select an entirelyuninformative experiment to test a hypothesis. An explanation of game and itsapplication to experimental design and clinical trials is given on the abovewebsite. The possible choices for your experiment and whether or not they areinformative (indicated by a +) are:
Possible ExperimentsHypotheses A b b B a B a Aletter, size, script + - - - - - - -letter, size + - - - - - - +letter, script + - - - - - + -size, script + - - - - + - -letter + - - - + - + +size + - - + - + - +script + - + - - + + -all coincidence + + + + + + + +
The best choices are B, a, or A because they will reduce the number ofpotential hypotheses from 8 to 4.
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 25
1.1 Defining the study population
Defining the study population
I Study population:I Subjects in the study are representative of the study
population.I Study results are intended to estimate effects in the study
populationI Study population is defined by:
I Eligibility criteria: Characteristics of individuals who will beinvited to participate in the trial
I Exclusion criteria: Individuals who will be excluded even ifthey meet the eligibility criteria (usually for safety reasons).
I “Target" population:I The population in which the treatment/intervention will be
used.I The treatment indication is for the target population.
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 26
1.1 Defining the study population
Defining the study population
Reasons why the study population might differ from target:(a) Ethical:
I Cannot force participationI Sometimes you must exclude “vulnerable populations"; e.g.:
I PrisonersI Pregnant womenI ChildrenI These populations might be studied after efficacy is
established in the study population.
(b) PracticalI E.g., restriction to major regional centers** Beware of excessive restriction
(c) Scientific:I Compliant (exclude subjects who are likely to be
non-compliant).I Restrict to subjects who will be able to complete follow-up
visits.I Exclude co-morbid conditionsI Require basic level of health(?)
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 27
1.1 Defining the study population
Defining the study population
Reasons why the study population might differ from target:(d) Special considerations in placebo-controlled trials:
I Restrict to patients who can (ethically) receive placeboI (Sometimes this restriction is used even if the new drug
would be offered to the patients who are excluded.)
But remember:“If it is ethical to conduct a placebo controlled trial, then it isunethical not to... (Lloyd Fisher)"
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 28
1.1 Defining the study population
Defining the study population
Selected examples for illustrationI OCEANS trial (Carol AghajanianC., et.al. J Clin Oncol
(2012); 30:2039-2045:A Randomized, Double-Blind, Placebo-Controlled Phase IIITrial of Chemotherapy With or Without Bevacizumab inPatients With Platinum-Sensitive Recurrent EpithelialOvarian, Primary Peritoneal, or Fallopian Tube Cancer
I Problems with gastro-intestinal perforation in early trialsI Restricting eligibility to platinum-sensitive recurrent cancers
appears to have eliminated those problems.I See Eligibility Criteria in paper (bottom of first column page
2040).
I CHEST trial (see first exercise)I Review JK notes on chapter 12.
-
Date: 4 Sep 2013
1. Setting andObjectives1.0 Background
(a) Treatment indication
(b) Inference uponcompletion
1.1 The study population
1.2 The study question
(a) What is the treatment?
(b) Phase I-IV trials
(c) Nature of the clinicalquestion
(d) 1-sided vs 2-sidedquestions
1.3 Case study: Rocket-AF
Bios 6648- pg 29
End:Section 1.1
Date: 4 Sep 20131. Setting and Objectives1.0 Background1.1 The study population1.2 The study question1.3 Case study: Rocket-AF