ICTW, Cordoba, Argentina

Clinical Research Design & Methodology: Phase III Trials

Ian Tannock, MD, PhD, DSc

Princess Margaret Cancer Centre & University of Toronto,

Toronto, Canada

ICTW, Cordoba, Argentina

To understand:• The sequence of clinical trials that is necessary to develop

an agent showing preclinical activity to become an approved anti-cancer drug

• The hierarchy of evidence from clinical trials• Basic principles underlying the design of Phase 3 clinical

trials• Common problems relating to the design, analysis and

interpretation of Phase 3 clinical trials• How the results of related clinical trials are combined in a

meta-analysis

Learning Objectives

Dr. Tannock

ICTW, Cordoba, Argentina

Anticancer Drug Development: Clinical Phases

Phase 1: Evaluate toxicity

Study drug disposition (pharmacokinetics)

Determine dose for Phase 2

Phase 2: Estimate antitumor efficacy

Further define toxicity

Phase 3: Compare outcomes with usual standard of care

Phase 4 (after registration):

Additional post-marketing safety assessment

Confirm results from fast-track approval trials

Dr. Tannock

ICTW, Cordoba, Argentina

Phase 3 Trials

Phase 3 trials have the goal of determining whether a new treatment provides sufficient benefit to patients that it should replace (or add to) current standard treatments.

Phase 3 trials generally include people with a single type of cancer, with randomization between the new treatment and an accepted “standard” treatment.

The new agent can be tested alone (i.e. standard vs. new treatment) or in combination with standard treatment (i.e. standard + new treatment vs. standard treatment).

They may be double-blind (e.g. standard + new treatment vs. standard treatment + placebo)

Dr. Tannock

ICTW, Cordoba, Argentina

Hierarchy of Evidence Used to Decide if Results of Clinical Trials Should Influence Clinical Practice

1. High quality randomized phase 3 trials, or meta-analyses

2. Small randomized trials

3. Non-randomized trials with concurrent controls

4. Non-randomized trials with historical controls

5. Expert committee review, case reports, retrospective studies

Dr. Tannock

Not included are large population-based outcome studies that can assess the influence of a new

strategy on a less selected population.

ICTW, Cordoba, Argentina

Asking a Good Question

Phase 3 trials require huge resources (collaborators, patients, $$$$)

It is therefore important that:– They address a clinically important question– There is substantial evidence for effectiveness of a new

treatment in preclinical models– There is evidence that the new treatment is safe and

tolerated (phase I)– There is good preliminary evidence of anti-tumor activity

(phase II)

Dr. Tannock

ICTW, Cordoba, Argentina

Phase 3 Trials

The primary endpoint (outcome measure) should be a measure of patient benefit (e.g. overall survival [OS] or QL)

If other endpoints are used they should be shown to be surrogates for OS or QL

Phase 3 trials are large (several hundred patients)

Entry criteria should be as broad as possible so that the results will apply to the general population of patients with the type of cancer under investigation

Dr. Tannock

ICTW, Cordoba, Argentina

DFS and PFS as Surrogate Endpoints in Phase 3 Trials

– When phase III trials show large early increases in PFS, and there are few alternative treatments, ethical considerations may require the new treatment be given to patients randomized to the control arm. This may make it difficult to detect an influence on survival.

– Examples include sunitinib for renal cell carcinoma and vemurafenib for BRAF-mutated melanoma

– This does not apply to treatments for (e.g.) metastatic breast or ovarian cancer

Dr. Tannock

ICTW, Cordoba, Argentina

Evaluation of Toxicity

Therapeutic benefit of a new treatment depends on a balance of efficacy and toxicity

All new agents add toxicity

Toxicity is usually reported and graded as per NCI CTC criteria, with rigid requirements for reporting serious (grade 3-4) toxicity

However toxicity is both under-recognized and under-reported in phase III trials – and chronic toxicity associated with many targeted agents may only become apparent after the trial is completed

Dr. Tannock

ICTW, Cordoba, Argentina

Principles of Design of Phase 3 Trials

Sample size depends on ....

The expected level of survival in the control arm

The difference in survival (δ) between arms that you wish to detect or rule out

α or type 1 error is the probability that a difference ≥ δ is a false positive result (usually α = 0.05)

β or type 2 error is the probability of failing to detect a real difference ≥ δ, i.e. of a false negative result (usually β = 0.1 or 0.2)

Dr. Tannock

The “power of the study” is then 1- β or 0.9 or 0.8

ICTW, Cordoba, Argentina

Phase 3 Trials: How to Determine Sample Size

Sample size for trials can be estimated from on-line calculators. e.g. http://www.openepi.com/v37/SampleSize/SSCohort.htm

Roughly 300 patients will be needed to detect or rule out a 15% absolute difference in survival & about 1,000 patients for a 10% difference!

Thus phase 3 trials require a large sample size to detect or rule out differences in outcome that might reasonably be expected

They require collaboration between multiple sites and are usually organized either by cooperative groups or companies

For company-organized trials it is essential that there be safeguards to ensure the validity of the data

Dr. Tannock

ICTW, Cordoba, Argentina

Phase 3 Trial: An ExampleA phase 1 trial indicates that a new drug, miraculin at 50mg/m2 IV every three weeks, can be added safely to standard gemcitabine/cisplatin

A phase 2 trial suggests that miraculin is active as second-line therapy for people with metastatic urothelial cancer (TCC)

MiraPharma agree to sponsor the following randomized phase 3 trial to determine if miraculin will become part of first-line treatment for this disease:

Gemcitabine + cisplatin + miraculin

Gemcitabine + cisplatin + matched placebo

Primary endpoint is

overall survival

Dr. Tannock

450 patients with metastatic TCC

and no prior chemotherapy

ICTW, Cordoba, Argentina

Survival Curves and Hazard Ratios

Patients are recruited to clinical trials at different times and have different length of follow up.

In actuarial survival curves patients alive at last follow-up are “censored”. The tail of the curve may depend on few patients (given as numbers under curves) and is subject to error.

Proportional hazards applies when the ratio of ‘events’ (e.g. deaths) in the experimental and control arms remains roughly the same over the time of observation.

Dr. Tannock

ICTW, Cordoba, Argentina

Survival Curves and Hazard Ratios

The Hazard Ratio (HR) is ratio of events in experimental to control arm in any given time interval.

HR < 1 with 95% confidence interval excluding 1.0 implies statistical benefit of experimental treatment.

Note that statistical significance is not the same as clinical significance

Dr. Tannock

ICTW, Cordoba, Argentina

Survival Curves and Hazard Ratios – An Example

Dr. Tannock

Baselga J, et al. N Engl J Med 2012; 366:109-119

ICTW, Cordoba, Argentina

Common Errors in Design, Analysis and Interpretation of Phase 3 Trials

Control arm is not an accepted standard treatment

The primary endpoint has not been shown to convey benefit to patients (i.e. not a surrogate for survival or its quality)

Sample size too small to detect or rule out a reasonable difference in outcome (failing to find a difference is not the same as proving no difference)

Sample size so large that difference in clinical outcome is statistically significant but not clinically meaningful

Making conclusions on the basis of secondary endpoints

Failure to provide detailed report of toxicity

Dr. Tannock

ICTW, Cordoba, Argentina

Post-marketing (Phase 4) Studies…

...evaluate a wider population of patients treated with a new therapy.

They can help to better define its tolerance and side effects in a wider population.

Often the unstated purpose of such studies is to encourage oncologists to become familiar with the new treatment (sometimes with financial reward) so that they will continue to use it when it is marketed.

Dr. Tannock

ICTW, Cordoba, Argentina

Meta-Analyses

Similar (but not necessarily identical) trials are combined (e.g. any adjuvant chemotherapy vs. none for women with primary breast cancer).

In a patient-based meta-analysis (preferred) the data are analyzed like one large trial to produce survival curves.

Date of randomization and death (or last follow-up alive) are known for all patients on multiple trials.

Data are displayed as a Forest plot: individual trials are represented by a square proportional to sample size, with whiskers showing confidence interval for Hazard or Odds Ratio.

Literature based meta-analysis combines published survival curves from reports of individual trials but is subject to error.

Dr. Tannock

ICTW, Cordoba, Argentina

Meta-Analyses: Example of a Forest Plot

Dr. Tannock

Dowsett M, et al. J Clin Oncol 2009; 28:509-518

Meta-Analysis of Aromatase Inhibitors vs. Tamoxifen