The Role of a Data and Safety Monitoring Board in Phase III Clinical Trials

John M. Lachin

The Biostatistics Center

Department of Statistics

The George Washington University

=================================

Interim Monitoring - History

NIH clinical trials - UGDP, CDP, etc.

Traditional sequential methods

Cornfield: Bayes Rules

Armitage et al.: RST plans

Group sequential

Spending functions

Stochastic curtailment

1978 NIH Guidelines

1988 FDA Guidelines

1990 PMA Guidelines

1997 ICH Guidelines

=================================

Objectives

Protocol and data quality

Patient safety - do no harm

"Ethics" - make the best treatment available as soon as possible

Cost savings -

- treatment effective

- treatment ineffective

Others

- select best dose or best new drug

- re-evaluate N

=================================

NIH Model

Audience:

- scientific and clinical community

Therapies:

- interventions (non-pharmacologic)

- surgical procedures

- new uses of established agents

- competing agents

- orphan drugs

- novel new agents

Mechanism:

- publish then treat

=================================

Industry Model

Audience:

- FDA then medical community

Therapies:

- new agents and devices

- new indications

Mechanism

- FDA review and approval,

publish, treat

=================================

OBJECTIVES -- NIH Vs INDUSTRY

Patient Safety:

Do No Harm

NIH and Industry models both place premium on safety

Largely not a statistical issue

Mechanisms for safety monitoring differ

=================================

Ethics- Offer Best Treatment ASAP

Applies to NIH model

Does not necessarily apply to Industry Model

If stop early

- Cannot offer agent to general public until FDA approval

- Interim monitoring or early termination may backfire and delay FDA approval

=================================

Cost-Savings

Yes, if stop due to toxicity

Yes, if stop due to ineffectiveness

???, if stop only due to effectiveness

=================================

Interim Monitoring Procedures

NIH Model

Data entered as collected in the clinic, daily or weekly

Data edited for errors immediately

DSMB meets periodically (Q6-12 Mos.)

- Reviews analyses of all outcome data

- Effectiveness, adverse effects and benefit/risk assessed between groups

Statistical Adjustments (boundaries)

Evidence Vs stopping rules

Basis for termination: Committee Gestalt

=================================

NIH Model Criteria for Termination for Effectiveness

Compelling, conclusive results

Intention-to-treat analysis

Not mere statistical significance

All objectives of trial met. No further gains even if continue

Examples:

- UGDP: Stopped too early??

- BHAT: Simple

- DCCT: Complex

- AZT in Pregnancy: Regulatory considerations

=================================

Interim Monitoring Procedures Industry Model -- Clinical Monitor

Data harvested periodically by CRAs

Data often not entered until study end

Safety monitored by Sponsor

- Adverse Event reports

- Clinical Monitor

- No ongoing statistical analysis of relative risks between groups

=================================

Interim Monitoring Procedures Industry Model -- The DSMB or IDMC

Data must be collected, entered and edited for errors in a timely manner

DSMB has less flexibility than NIH model

- Precise assessment of type I error

- Operations pre-specified as much as possible: outcomes and statistical methods

Regulatory implications

- Choice of primary outcome

- Choice of primary analysis

=================================

AZT in Pregnancy

Primary Outcome:

Incidence of HIV in offspring at 1 year

Not the cumulative incidence curve

Monitoring Procedure:

Z-test for Kaplan-Meier cumulative incidence at 1 yr.

Not the logrank test

=================================

General Issues in Interim Monitoring

Primacy of a single outcome vs. multiple outcomes

Safety or toxicity adequately assessed

Benefit : risk adequately assessed

Completeness of data (losses to follow-up; incomplete interim ascertainment)

Consistency among related measurements

Consistency across clinics and subgroups

Whether termination affects:

- precision of final results

- credibility of trial

=================================

Possible Issues in FDA Review

Operational process may appear to introduce bias

- Clinical vs. data monitoring

- Study management

Patient years of exposure for safety

Adequate numbers of events

Time course of effectiveness

- short vs. long-term effects

- logrank test vs. proportions test at fixed time

Adequacy of documentation

=================================

Recommendations

Not recommended for routine practice in industry sponsored trials

Recommended when

1. Safety concerns pre-exist

2. Early termination for effectiveness will not jeopardize FDA review and approval

3. A single dominant outcome measure is employed

=================================

Recommendations (cont.)

If conducted, sponsor should not participate in review by DSMB.

- Committee should be external.

Statistician member of DSMB should not be associated with the study.

- Operational statistician should not attend meetings.

- Preferably independent analyses as well.

=================================

Recommendations (cont.)

Early termination criteria should be explicitly described a priori:

- number of planned looks

- approximate information times

- outcomes to be monitored

- statistical techniques.

Complete documentation

DIA298sl..DOC DIA meeting 2/98 in California