Introduction to Biostatistics Dennis O. Dixon, Ph.D. Formerly, Mathematical Statistician National Institute of Allergy and Infectious Diseases, NIH 2011
Introduction to Biostatistics
Dennis O. Dixon, Ph.D. Formerly, Mathematical Statistician
National Institute of Allergy and Infectious Diseases, NIH
2011
Running Example
Study the problem of babies born with HIV infection: • What is the extent of the problem? • What is the evidence? • How good is the evidence? • Can the problem be eliminated / reduced?
2
Concepts to Be Covered
• Sample v. Population • Estimate v. Parameter • Curse of variation (Sample Size) • Comparing Parameters (Hypothesis
Testing) • Clinical Trial Design • Interim Trial Monitoring
3
Sample v. Population
4
Research Question
What proportion of babies are infected with HIV at birth? Terminology: Proportion = fraction = rate, a number between 0 and 1, ratio of numerator and denominator
5
Research Question
What proportion of babies are infected with HIV at birth? • In a particular place (country, district,
hospital)? • In a particular time period? • Born to certain types of parents?
6
Example: “Among 1000 babies born to HIV-infected mothers in Hospital X in 2008 the proportion of babies who were infected was 0.25 (250/1000).”
7
Estimate v. Parameter
• What if there were actually 5000 babies born to infected mothers at Hospital X in 2008, but only a sample of 1000 of those babies were tested for HIV?
• The proportion 0.25 is an estimate of the actual proportion, which could be as low as 250/5000 = 0.05 or as high as 4250/5000 = 0.85.
• Of what value is the estimate? 8
More Generally
• Often the parameter of interest is not observable, even in principle.
• The population of interest, and hence the parameter of interest, is hypothetical instead of actual.
• Data from Hospital X, interest in all of Nigeria
• Estimate important only insofar as it sheds light on a parameter
9
Other Examples of Parameters
• Probability that a negative cancer screening test is correct
• Average survival time of men after first heart attack
• Effect of nutritional supplementation on infant mortality
• Probability that my baby will be infected at birth
10
Curse of Variation
11
Kinds of Samples
• Representative – Fair – Random • Convenient – Systematic • Skewed – Biased These are all context-specific: Any sample is
representative relative to some population, not always the population of interest
12
To Assure a Good Sample
• Carefully identify the members of the population of interest
• Choose the sample at random (by lottery), giving each member the same chance of selection
13
Estimates Based on Good Samples
When estimating a proportion from a large sample (N),
– p (estimate) will tend to be close to π (parameter)
– In fact, in hypothetical repeated samples, p and π will be no farther apart than one standard error of p, denoted sep, 68% of the time, and no farther apart than 2 x sep 95% of the time, where sep
2 = π(1- π)/N
14
Estimate Precision When π = 0.25 and Various N
N se0.25 (68% limit)
95% limit
10 .14 .27 100 .0433 .0866
1000 .0137 .0274
So when π = 0.25, an estimate based on 1000 babies will be between about 0.22 and 0.28 95% of the time.
15
Inference
• Precision expresses the variation of an estimate when the parameter has a fixed value (whether known or unknown)
• Given the estimate, what can be said about the unknown parameter?
16
Confidence Intervals
• Noted earlier that Prob{ - 2sep < p – π < 2sep } = 0.95
• This is a description of the variability in p that would be seen in repeated samples
• When p has been calculated from a set of data, the expression inside { } can be rearranged to [ p - 2sep < π < p + 2sep ], commonly referred to as a 95% confidence interval for π
17
Note
• Without knowing the true value of π you can’t find the true value of
sep = square root of [ π(1- π)/N ] • Substituting p for π in the formula is OK
unless N is small (< 30) or π is close to 0 or 1
18
Comparing Parameters
19
Is an Intervention Any Good?
• Suppose the infection rate in babies born to infected mothers is 0.25 (this is the true rate πo)
• Does a short course of AZT to mother and baby reduce the infection rate?
• In a sample (series) of 100 births only 15 babies are infected.
• Is the benefit of AZT proven?
20
Example
• With 15 infected babies out of 100 tested, the 95% confidence interval for πAZT is
[ 0.15 – 0.07, 0.15 + 0.07 ] = [ 0.08, 0.22 ]
21
Example
• With 15 infected babies out of 100 tested, the 95% confidence interval for πAZT is
• [ 0.15 – 0.07, 0.15 + 0.07 ] = [0.08, 0.22 ] • Strict interpretation: Over repeated similar
samples, 95% of the confidence intervals will include the true value of πAZT
22
Example
• With 15 infected babies out of 100 tested, the 95% confidence interval for πAZT is
[ 0.15 – 0.07, 0.15 + 0.07 ] = [ 0.08, 0.22 ]
• Informally, all plausible values of πAZT (with 95% confidence) are between 0.08 and 0.22
23
Possible Argument • The estimate 0.15 based on 100 observa-
tions has a confidence interval [0.08 - 0.22]
95% CI
π 0.08 0.22
24
Possible Argument
95% CI .25
π 0.08 0.22
Since the interval is all below the infection rate without AZT (0.25), one can reasonably conclude that πAZT is smaller than πo
25
Formalise: Tests of Hypotheses
• Ho: πAZT = πo stands for the null hypothesis that AZT does not change the infection rate
26
Formalise: Tests of Hypotheses
• Ho: πAZT = πo stands for the null hypothesis that AZT does not change the infection rate
• Ha: πAZT < πo or πAZT > πo stands for the alternative hypothesis that the rates are not equal
• Both hypotheses refer to the unknown value of πAZT
27
Tests of Hypotheses
• Data will be compatible with Ho or not; if not, reject Ho in favor of Ha
• Otherwise do not reject Ho – not the same as accepting Ho
• A popular way to decide what is compatible is using a confidence interval for the parameter of interest
28
Hypothesis Testing Logic
• Obtain an estimate (p) and its standard error from a good set of data
• Form the 95% confidence interval for the parameter (π)
[p – 2sep, p + 2sep] • Reject Ho: π = πo if πo is not within the
interval; otherwise do not reject
29
Possible Error – Type I
• By design, this procedure sometimes will reject Ho even when it is true, because 5% of the time the estimate will fail to be close to the true parameter value
• Incorrectly rejecting Ho is called a Type I error
• Example: claiming AZT lowers the infection rate below 0.25 when it doesn’t
30
Type I Error Rate
• P(reject Ho when Ho is true) = α • Probability of rejecting the null hypothesis
when the null is true • False positive error rate • Probability of rejecting that πAZT = 0.25
when AZT makes no difference
31
Possible Error – Type II
• Not rejecting Ho when Ha is true is also an error, called a Type II error
• Example: not claiming AZT lowers the infection rate below 0.25 when it does
• The researcher wants to avoid both errors and can do so to some extent through careful study design
32
Type II Error Rate • P(do not reject Ho when Ha true) = β • False Negative error rate • Probability we do not conclude that πAZT
differs from 0.25 when it truly does • Power = 1-β = P(reject Ho when Ha is true)
33
Possible Type II Error
.25
π 0.14 0.30
34
Controlling Error Rates
• Type I errors are bad, because we do not want to use ineffective treatments
• Never rejecting eliminates Type I errors
35
Controlling Error Rates
• Type I errors are bad, because we do not want to use ineffective treatments
• Type II errors are bad, because we do not want to overlook effective treatments
• Some Type II errors are worse than others
36
Alternate Form
• Reject Ho: π = πo at the 0.05 level if p < πo – 2sep or p > πo + 2sep
• Otherwise do not reject • Rearranging, the criterion is
p – πo < – 2sep or p – πo > 2sep Same as Z < -2 or Z > 2,
where Z = (p - πo)/sep
37
(Significance) Level
• 2 and 0.05 are related: The critical value – 2 – is specified so that α, the probability of a Type I error, will be 0.05
• This exactly matches the property of the 95% CI that the interval contains the true value of π 95% of the time
• Critical values 1 and 3 correspond to significance levels 0.32 and 0.0027
38
How α Affects CI Length
α = 0.0027
α = 0.05
α = 0.32
39
P-value
• Smallest level α for which the observed sample would reject Ho
• Given Ho is true, probability of obtaining a result as extreme or more extreme than the actual sample
• Measure of the strength of evidence in the data that the null is not true
• “p is significantly < πo with P = 0.045.”
40
Comparing rates when both must be estimated
• Different formula for Z but same procedure • To test Ho: π1 - π2 = 0 at level 0.05, reject if
Z < -2 or Z > 2, where
Z = (p1 – p2)/sediff and
sediff = square root of [ se12 + se2
2 ]
41
Possible Type II Error
0
πAZT – πo -0.17 0.07
A reduction of the transmission rate from 0.25 to 0.08 is not ruled out, so do not claim Ho is true
42
Power Depends on Sample Size
• Power = P(reject Ho when Ha is true) = “Probability of rejecting the null hypothesis if the alternative hypothesis is true”
• More subjects è smaller se è smaller CI è higher power
43
Other Cases
Z formulation works in almost all situations with plenty of data
Estimate – parameter under null seestimate
44
Comparing Average Values of Measurements (such as
weight change) To test Ho: µ1 = µ2 at the 0.05 level, reject if
Z = (m1 – m2)/sediff
is either < -2 or > 2; otherwise do not reject, where se1
2 = sum(m1i - m1)2/n1 (n1-1), and n1 is the number of observations used for m1
45
Note
To be precise, critical values (1, 2, and 3 for α = 0.32, 0.05, and 0.0027) will depend on the properties of the sampling distribution of the estimate
46
Summary • Inferences from data always have
limitations • Estimates of parameters of interest are all
imprecise to some degree, should come with confidence limits
• Hypothesis test results are always subject to either Type I error (in case of a rejected hypothesis) or Type II error (in case of a non-rejected hypothesis)
47
Clinical Trial Design
48
Example: ACTG 076
Reduction of Maternal-Infant Transmission of Human Immunodeficiency Virus Type 1 with Zidovudine Treatment
49
Basic Design Components • Background and rationale • Clear statement of primary hypothesis to
be tested or parameter to be estimated • Other objectives • Concise statement of design • Methods and analysis plan
50
Designing a Clinical Study = Writing a Protocol
• Road map for conduct of study • Anticipate problems • Facilitates communication with potential
collaborators, funding agencies • Assists in manuscript preparation
51
ACTG 076
Background and Rationale • 25% of babies born to HIV+ mothers were
also infected in the early 1990s • Adults with AIDS became very ill more
slowly if given AZT • Perhaps reducing disease burden in HIV+
mothers with AZT and/or treating babies with AZT would reduce transmissions
52
ACTG 076
Primary Hypothesis Can infection rate of babies be reduced safely by administering AZT to mothers during pregnancy and delivery and to infants for 6 weeks?
53
Basic Design Components ü Background and rationale ü Clear statement of primary hypothesis to
be tested or parameter to be estimated • Other objectives • Concise statement of design • Methods and analysis plan
54
ACTG 076
Other Study Objectives • What other characteristics of mothers
(demographics, medical history, etc.) or infants (gender, birth weight, etc.) affect the transmission rate?
• Do AZT effects vary among subgroups?
55
ACTG 076 Statement of Design
A randomized, placebo-controlled trial of “short-course” AZT vs. standard obstetrical care alone, to enroll 636 “assessable” mother-infant pairs, with mothers between 14 and 34 weeks’ gestation and not given AZT in the current pregnancy, in the U.S. and France … 56
Methods • Selection of subjects
– Inclusion criteria – Participating sites, investigators
• Treatment and treatment administration • Data and sample collection and handling • Statistical considerations • “Trial conduct”
57
Specifying the Population of Interest and the Study
Population
58
Study Eligibility
• The characteristics of the study population influence the generalizability of study results
• Not easy to reach a conclusion about both adults and children if only adults enroll in the study
• Rarely feasible to match the target population fully
59
Inclusion Criteria • Definition of patient population
– Specific (but not too restrictive) • Inclusion criteria
– Disease or condition under study • Prior myocardial infarction, smokers
– Other information • Age • Sex • Area of residence or hospitalization
60
Exclusion Criteria • Participants must not have any specified
criterion • Conditions making study difficult or
impossible • Persons for whom one regimen or other is
inappropriate or unethical – Women with symptomatic HIV illness should
not be denied antiretroviral therapy in ACTG 076
61
Exclusion Criteria • Practical concerns
– Aged under 18 – Critically ill
• Circumstances making determination of outcome difficult or impossible – Expected to leave area – Unable to communicate in language study
team uses – Pregnancy (in studies of treatment for
women) 62
Common Mistakes • Unnecessary exclusion/inclusion criteria • Plans for the trial made without any
reliable data on participant availability – Pilot recruitment
• Unrealistic timetable for recruitment or no recruitment goals
• Revision of sample size calculations to make them consistent with recruitment realities
63
Treatment Definition • Specify as much as possible without
interfering with patient management • Realize that generalizability often lost in
quest for specificity • Specify criteria for withdrawal from study
or deviation from protocol • List concurrent medications, procedures,
etc. that are prohibited or permitted
64
Methods ü Selection of subjects
- Inclusion criteria - Participating sites, investigators
ü Treatment and treatment administration • Data and sample collection and handling • Statistical considerations • “Trial conduct”
65
Case Report Forms
• Collect relevant data, relative to protocol • Facilitate efficient and complete data
collection, processing, and analysis • Provide units of measure (e.g. pounds,
kilos) • Include instructions for completion,
participant ID number, Study name, but not participant’s name
66
Outcome Definitions
• Be specific and as clear as possible • Primary vs. secondary outcomes • Standard clinical definitions
– Textbook: usually not specific enough – Consensus conference
• Definition of new infection – Recognized expert body (WHO, AHA)
67
ACTG 076 Outcome Definitions
• Infected: at least one positive HIV culture of peripheral-blood mononuclear cells, at birth, 12 weeks, 24 weeks, and/or 78 weeks, from certified labs
68
ACTG 076 Outcome Definitions
• Infected: at least one positive HIV culture of peripheral-blood mononuclear cells, at birth, 12 weeks, 24 weeks, and/or 78 weeks, from certified labs
• Not Infected: all cultures through 78 weeks negative
• Censored: negative through < 78 weeks
69
Sample Size Determinants
• Variables of interest – type of data e.g. continuous, categorical
• Desired significance level (α) • Desired power for effect/difference of
clinical importance (1 – β) • Standard deviations of continuous
outcome variables • Analysis plan
70
Sample Size for Comparing Proportions
Total Sample Size for α = 0.05
π1 π2 β = 0.2 β = 0.1
0.60 0.50 780 1040
0.40 200 260
0.30 90 120
0.30 0.20 590 790
0.15 250 330
0.10 0.05 870 1170
Excerpted from Table 7-3 of Friedman et al. (1998) 71
ACTG 076 Statistical Considerations
• Focus on infection rates at 18 months of life, based on testing at birth, 12, 24, and 78 weeks of life
• Rate estimates account for censoring • α = 0.05 (overall, allowing for three
interim and one final analysis) • β = 0.20 for a 1/3 reduction from 0.30 • Total N = 636 assessable pairs
72
Sample Size for Comparing Proportions
Total Sample Size for α = 0.05
π1 π2 β = 0.2 β = 0.1
0.60 0.50 780 1040
0.40 200 260
0.30 90 120
0.30 0.20 590 790
0.15 250 330
0.10 0.05 870 1170
Excerpted from Table 7-3 of Friedman et al. (1998) 73
MOP: Manual of Procedures or Manual of Operations
• Can another investigator step into the study (or reproduce it) at any time?
• Study organization • Which data to be collected, how • Follow-up schedule, with visit windows • Specimen collection and handling • Adverse event collection, reporting
74
Randomization
75
Randomization: Definition • Random Allocation
– Known chance of receiving a treatment – Cannot predict the treatment to be assigned – Not a random sample from target population
• Eliminate Selection Bias – Outcome measurements may still be biased
• Similar Treatment Groups – Does not eliminate baseline differences – If baseline differences they arose by chance
76
ONE Factor is Different • Randomization tries to minimize
differences between two or more groups other than treatment assignment
• Observe the consequences • Attribute difference to causality
• In truth, a rarity and cannot test
77
Simple Randomization
• Randomize each patient to a treatment with a known probability – Corresponds to flipping a coin (head/back)
• Could have imbalance in # / group or trends in group assignment
• Could have different distributions of a trait like gender in the two arms
78
Block Randomization
• Insure the # of patients assigned to each treatment is not far out of balance – Block size = 6, 2 study interventions (A and B)
• AAABBB, BAABAB, ABABAB…. • Variable block size - additional layer of
blinding • Different distributions of a trait like gender
in the two arms possible
79
Stratified Randomization
• A priori certain factors likely important (e.g. known risk factors, research site)
• Randomize so different levels of the factor are balanced between treatment groups
• For each stratum perform a separate block randomization
80
Cluster Randomization • Unit of randomization
– School/Clinic/Hospitals – Providers
• Outcome measurement – Students – Patients
• Special analysis methods needed
81
Masking/Blinding
• Disguise or conceal actual treatment assignment
• Prevent even unintentional bias in subjective assessments by subjects and researchers
82
Masking/Blinding
• Disguise or conceal actual treatment assignment
• Prevent even unintentional bias in subjective assessments by subjects and researchers
• Specify criteria for unmasking, people to be unmasked
83
Intent to Treat vs. Completers
• ITT = Intent To Treat analysis – Includes everyone randomized
• MITT = Modified ITT analysis – ITT, but only include people who start the
intervention they are assigned to • Completers or Adherers analysis
– Only the well behaved
84
Phase I to Phase IV Trials
• National Cancer Institute: Dictionary of Cancer Terms
• Phase I - The first step in testing a new treatment in humans
• Test the best way to give a new treatment (for example, by mouth, intravenous infusion, or injection) and the best dose
85
Phase I
• Dose increased little at a time • Find the highest dose that does not cause
harmful side effects • Little known about possible risks and
benefits of the treatments being tested • Trials usually include only a small number
of patients who have not been helped by other treatments
86
Phase II
• A study to test whether a new treatment has an anticancer effect (for example, whether it shrinks a tumor or improves blood test results) and whether it works against a certain type of cancer
• Surrogate endpoints
– Note: many things shrink tumors and do not improve survival/quality of life (ACRT/SCTS story)
87
Phase III
• A study to compare the results of people taking a new treatment with the results of people taking the standard treatment (for example, which group has better survival rates or fewer side effects)
• “Definitive”
88
Phase IV • After a treatment has been approved and
is being marketed, it is studied in a phase IV trial to evaluate side effects or other indications or populations that were not apparent in the Phase III trial
• Thousands of people are involved in a Phase IV trial
• Different population from Phase III • Duration of use
89
Data and Safety Monitoring in Clinical Research
90
Usual Experiment Timeline
Design
Preparation
Experiment
Analyze
Report
91
Clinical Trial Timeline
Design and Preparation
Recruit Intervene Observe
Report
92
Clinical Trial Timeline
Design and Preparation
Recruit Intervene Observe
Report
Overlap provides the opportunity to monitor 93
Definition
(Data and safety) monitoring is the planned review of study progress to enhance the quality of study conduct, detect problems, and allow appropriate changes to the study.
94
Presentation Outline
• Example – monitoring by independent Data and Safety Monitoring Board (DSMB)
• Data and safety monitoring basics – monitoring all kinds of studies
• When to use a DSMB
95
ACTG 076 • AZT able to slow progression of
HIV in adults with advanced disease
• AIDS Clinical Trials Group Protocol 076 designed to assess both safety and efficacy of AZT in preventing transmission of HIV from infected women to their babies
96
ACTG 076 • Powered (80%) to detect a 33%
reduction of transmission rate (through 78 weeks of life) relative to projected rate of 30%
• Target N was 748; began April 1991 • Projected accrual to take at least 5
years and 15% dropouts
97
ACTG 076 Monitoring Plan
• DSMB met twice a year to monitor safety • Efficacy reviews planned after each 1/3
of projected infant infections (with censoring the number of events plays the role of the number of subjects)
• 1st efficacy review took place in February 1994, based on mothers enrolled up to December 1993 and their babies
98
Group Sequential Boundaries
99
P = 0.00006
At First Interim Analysis
100
ACTG 076 • After careful review of data quality and
completeness, toxicity, transmission rates, DSMB recommended stopping
• Trial leaders and NIAID agreed • AZT provided to those in control group • U.S. Public Health Service Guidelines
modified • Ref: Connor et al. (1994) NEJM
101
Presentation Outline
ü Example – monitoring by independent Data and Safety Monitoring Board
• Data and safety monitoring basics – monitoring all kinds of studies
• When to use a DSMB
102
Data and Safety Monitoring: Why?
• To identify any safety problem rapidly • To identify logistical problems • To evaluate continued feasibility of trial • To determine if trial objectives have
been met and trial may be terminated early
103
Data and Safety Monitoring: What? Logistics
• Enrollment • Baseline Data, Comparability • Protocol Compliance • Specimen Collection • Data Quality Develop specific benchmarks
104
Realities of Subject Recruitment
• Early estimates unrealistically high • Takes a major effort • People presumed eligible for study during
planning will disappear mysteriously as soon as the study starts
• Recruitment will be more difficult, cost more, and take longer than planned
105
106
Prepare! • Collect reliable data to estimate
participant availability • Commonly enroll half of those screened • Decide on general recruitment approach • Outline steps in recruitment process • Establish necessary recruitment contacts
107
Recruitment Mistakes/Problems • Attempting recruitment without the
support of colleagues • Taking access to medical records for
granted • Failing to secure enthusiasm and
commitment of staff • Inadequate publicity
108
CONSORT Diagram
# screened
# entered
# analyzed
# ineligible (reasons) # refused (reasons)
# withdrawn (reasons) # lost to follow-up # discontinued therapy # completed
109
Protocol Adherence Missed Visits
Baseline Visit 1 … Last Visit # Expected … # Missed …
% Retained …
Overdue Forms Baseline Visit 1 … Last Visit
# Expected # Received % Complete
110
Baseline Characteristics
Gender % Female
Age Median, Min, Max
Stage of Disease % Early, % Advanced
Prior Therapy % None, % Surgery, % Chemo
111
076 Database as of 12/93
AZT Placebo Total Women enrolled 239 238 477
Still pregnant 24 24 48
Withdrew before giving birth 11 10 21
Births 205 204 409
At least one culture 180 183 363
No cultures 25 21 46
112
Reasons for Lack of Cultures
Withdrawal 6
Death 1
Too young 20
Results not submitted 19
113
Data and Safety Monitoring: What?
Outcomes • Adverse Events • Interim Variables • Response Variables (Endpoints)
114
Serious Adverse Events
Line Listing showing Event Study entry, treatment start dates Event start, stop dates, final resolution Relationship to research procedures Other relevant patient characteristics
115
Adverse Event Summaries Lab Abnormalities and Clinical Signs
Tables of frequencies, by AE type and severity
• Include all those treated • Sort by body system • Count 1st occurrence for each volunteer • Summarize across types
116
Adverse Event Summaries Body System MedDra Term
Mild
Moderate
Severe or Worse
Total
INFECTIONS INFESTATIONS
3(30%)
1(10%)
0(0%)
4(40%)
Gastroenteritis 1(10%) 0(0%) 0(0%) 1(10%)
Sinusitis 1(10%) 0(0%) 0(0%) 1(10%)
URI 0(0%) 1(10%) 0(0%) 1(10%)
Viral infection 1(10%) 0(0%) 0(0%) 1(10%)
Table: Frequencies (relative frequencies) of adverse events among 10 subjects treated
117
Efficacy Summaries
Summarize study endpoints: • % treatment success • Average AUC, antibody response, etc.
118
Data and Safety Monitoring: Who?
• Ethics Committee(s) • Sponsor • Regulatory Agencies • Data and Safety Monitoring Board
(DSMB, DSMC, DMC, External DMB, etc) • Investigator(s) • Safety Monitor
119
Why Data and Safety Monitoring Boards?
• To ensure regular and systematic interim monitoring
• To provide an objective assessment of the interim data
• To protect confidentiality of interim treatment comparisons
120
Generally Accepted Principles
• Certain types of trials should have formal DSMBs
• DSMBs should be multidisciplinary • A charter should describe the operations
and procedures of a committee • DSMB members should be free of conflicts
of interest • Interim data should be considered highly
confidential 121
An Independent DSMB Is One in Which No Member Has
• Any basis for preferring the outcome to be in one or the other direction
• Any ability to influence the trial conduct in a role other than that of DSMB member
122
Confidentiality of Interim Results
• Interim comparative data generally considered highly confidential, because
• Knowledge of interim data could affect – patient entry – patient care – patient assessment – sponsor action complicating interpretation of results
123
Scope of DSMB Responsibilities
• Evaluating accumulating data with regard to efficacy and toxicity
• Recommending termination or continuation of study
• Recommending other study modifications • Reviewing study protocol • Assessing study conduct • Recommending additional analyses
124
Monitoring Recommendations
• Continue Protocol Unmodified • Modify Protocol • Terminate Trial
125
DSMB May Recommend Stopping If • A safety issue has emerged • The trial has already shown
efficacy • Interim results preclude a
positive finding • Operational difficulties are
insurmountable • External information
undercuts the scientific rationale for the trial
126
Decision Making Process is Complex
• Internal consistency • External consistency • Benefit/risk balance • Current vs. future patients • Clinical and public health impact • Statistical issues
127
ALL TRIALS NEED MONITORING BUT NOT ALL TRIALS
NEED DSMBS
Data and Safety Monitoring Regulations, Policies
• Regulations NONE • Policies - NIH
– All trials need a plan – describe in application (2000 NIH Guide)
– Phase III trials must use a DSMB (1998 NIH Guide)
129
Data and Safety Monitoring Policies, Guidelines
• Policies - NIAID – Clinical Terms of Award – NIAID Policy on DSMB Operations
(2006, 2008, 2009) • FDA Guidance on DMCs (2006) • ICH Good Clinical Practice Guidances
130
When Are Independent DSMBs Needed?
• Large randomized trials with mortality or major morbidity endpoints
• Trials for which assessment of serious toxicity requires comparison of rates
• Trials of novel, potentially high-risk treatments
131
Independent DSMBs Generally Not Needed for • Single-arm trials • Early phase trials • Short-term trials of treatments to
relieve common symptoms • Any trial for which there is no ethically
compelling need to monitor the interim comparisons of safety or efficacy
132
TRIAL LIFE -- OVERSIGHT
START END
IRB
DSMB DSMB
End of Year 1
End of Year 2
IRB IRB DSMB DSMB
IRB
IRBs and DSMB review the proposed “Data and Safety Monitoring Plan” as part of the initial protocol review
133
Summary
Monitoring • Contributes to quality and ethics of study
conduct • Requires planning • Can take a variety of forms based on risks
of participation and study size and complexity
• DSMBs serve as a model 134
Resources: General Books • Hulley et al (2007) Designing Clinical Research,
3rd ed. Lippincott Williams & Wilkins, Philadelphia
• Bland (2000) An Introduction to Medical Statistics, 3rd. ed. Oxford University Press
• Armitage, Berry and Matthews (2002) Statistical Methods in Medical Research, 4th ed. Blackwell, Oxford
• Friedman, Furberg, DeMets (2010) Fundamentals of Clinical Trials, 4th ed. Springer, New York
135
Resources: General/Text Books
• Altman (1991) Practical Statistics for Medical Research. Chapman and Hall
• Fisher and Van Belle (2004) Wiley • Simon et al. (2003) Design and Analysis of DNA
Microarray Investigations. Springer • Rosner Fundamentals of Biostatistics. Choose
an edition. Has a study guide, too.
136
Resources: FDA Guidance
• http://www.fda.gov/cdrh/ode/odeot476.html (devices, non-diagnostic)
• http://www.fda.gov/cdrh/osb/guidance/1620.html (diagnostics)
• And all the ones listed before
137
Resources: URLs • Sample size calculations simplified
– http://www.tufts.edu/~gdallal/SIZE.HTM
• Stat guide: research grant applicants, St. George’s Hospital Medical School (http://www.sgul.ac.uk/depts/chs/chs_research/stat_guide/guide.cfm)
– http://tinyurl.com/2mh42a • Software: nQuery, EpiTable, SeqTrial, PS (http://
biostat.mc.vanderbilt.edu/twiki/bin/view/Main/PowerSampleSize)
– http://tinyurl.com/zoysm
138
Software
• Most is expensive and some have yearly license fees – Sometimes universities have agreements and
costs less • Some is hard to use, some is easy
139
Other Software
• STATA (Windows/Mac/UNIX) – Good for general computation, survival,
diagnostic testing – Epi friendly – GUI/menu and command driven – Active user community – www.stata.com
140
Other Software
• SAS (Windows/UNIX) – Command driven – Difficult to use, but very good once you know
how to use it – Many users on the East coast – www.sas.com
• SPSS, EpiCure, many others
141
Statistical Calculators • www.randomization.com • http://calculators.stat.ucla.edu/
– “Statistical Calculators” – Down recently
• http://statpages.org/ • http://www.biostat.wisc.edu/landemets/ • http://www.stat.uiowa.edu/~rlenth/Power
142