Randomized Controlled Trials随机临床试验

Xu Xiong, MD, DrPHSchool of Public Health and Tropical Medicine

Tulane University

Why Do RCTs?

• Compare new drugs, treatment, medical and health care technology against the current clinical standard

• Evaluate the effectiveness of prevention• Assess the usefulness of programs for

screening and early detection of disease• Evaluate the impact of new policies in health

care and health care financing• Make a causal inference

RCTs: Basic Study Design

RCTs

• The major distinctions between cohort studies and RCTs:– Intervention (exposure)– Randomization

• Group assignment is unknown to researcher and subject whenever possible - “blinding”

Identifying Study Participants

• Representative of the reference population.

• Population versus hospital-based.• Depending on the nature of

intervention, e.g.,– Vitamin E/C for preventing pre-eclampsia– Preventing congenital syphilis in Africa

Old/Current Treatment

• The treatment, drug, existing behavior, health care system, or even a placebo that we use as a reference against the “new treatment”.

• For both ethical and practical reasons the old treatment should be the best currently available treatment.

Placebo• Specifically - an inert substance that looks,

tastes and smells like the agent being tested.• Any alternate drug, treatment, behavior

change that closely replicates the treatment of interest, so that the participants are unable to determine what treatment they are receiving (i.e. used to blind/mask participants).

• Placebo effect – any effect attributable to the expectation that the regimen (treatment) will have an effect (Last JM)

Randomization

• Assignment of an individual or group to a treatment arm/study group by a means of chance.

• Normally by use of a random number table, computer generated random number pattern or other random generator (coin, dice, etc.).

Primary Goal of Randomization

• Remove all subjective biases which may be introduced by investigators and participants – Investigators can’t choose their favorite patients.– Participants can’t choose the treatment they receive.– Any differences in characteristics of the study groups

at baseline should be purely due to chance, and chance alone.

Randomization does not always work

Study Population

Treatment Arm 1 Treatment Arm 2

Randomization

2000

1000 1000

1000 men, 1000 women

550 men, 450 women 450 men, 550 women

Stratification

• Stratify the population by certain variables to help improve comparability of study arms/groups.

• Will guarantee comparability for the stratified variables within the study arms/groups

Stratification

Masking - Definition

• Last JM – “Procedures intended to keep participants in a study from knowing some facts or observations that might bias or influence their actions regarding the study”

Masking (Blinding)

• Three levels of masking– Participants– Allocaters/ Data collectors– Investigators/Analysts

• Basic intention is to remove conscious or sub-conscious bias (prior opinion) on the part of participants or investigators.

Masking (Blinding)

• Participant Masking: Achieved through proper randomization and use of placebos.

• Investigator Blinding

• If both participant and investigator are masked, then the study is termed: “Double Blinded”

Phases of Clinical Trials - FDA

• Phase 1: Small studies to assess human safety

• Phase 2: Efficacy and safety

• Phase 3: Randomized controlled trials

• Phase 4: Post-marketing surveillance

Generalizability

• The ultimate goal of any Randomized Controlled Trial is to generalize the results of a trial to the general/reference population.

• Must consider Internal Validity and External Validity

Problems and Issues relating to RCTs

• Ethics

• Non-participation

• Non-compliance

• Crossover

• Loss to follow-up

• Sample size

• Multicentered Collaborative Trials

Ethics in RCTs

• Is randomization ethical?

• Can truly “informed” consent be obtained?

• Under what circumstances can a trial be stopped prior to the original plan?

(DSMB: Data and Safety Monitoring Board) • Laws and regulations (HIPAA, …)

Non-Participation

• Selection of a study population may be limited/difficult.

• Try as hard as possible to get a population representative of reference population but also voluntary.

• Non-participation may limit the ability to even conduct a study.

Refusal to Participate

Non-Compliance• After randomization, participants may

overtly or covertly not take the assigned treatment. They “do not comply” with their allocation.

Non-Compliance

• The net effect of non-compliance on a controlled trial is a reduction in the differences seen between study groups.

• Compliance monitoring– May be done by keeping track of treatments

taken– Biological testing for treatment evidence

Loss to Follow-up

• Complete loss of contact, either through personal movement or death, that results in a participant not being able to be fully followed or treated.

• Must be assessed to ensure generalizability

• High loss to follow-up may indicate a poor study

What do you do with the non-compliers, etc.?

Sample Size Estimates

Example: Sample Size Calculation for RCTs

Single Institution vs. Multiple Site Collaborative Controlled Trials

• Multi-center trials can be beneficial when:– Generalization is important– Limited number of patients are available from a

single clinic– We wish to bring together expertise

• However, multi-center trials add a significant amount of overhead to a study.

• http://www.clinicaltrials.gov/

• http://www.who.int/ictrp/en/

• https://www.clinicaltrialsregister.eu/index.html

Registration of Clinical Trials

Data Collection in RCTs

• Prognostic profile at entry

• Treatment

• Outcome

• Criteria for outcome, data management, and acquisition must explicit and written out.

Data Collection – Treatment Data

• Assigned:– The treatment arm/group that the subject was allocated to

(placebo or active).– Treatment A, Treatment B, or Treatment C

• Received:– The actual treatment that the participant received.– During the study, the investigators must assess compliance

to the treatment– Interview/ specimen collection

Outcome Data

• Look for “improvement”, i.e. the desired effect

• Also must record any and all “side effects”– Beneficial– Harmful– SAEs (severe adverse events)

Intention To Treat (ITT)• What do you do with the non-compliers, etc.?• Intention to Treat analysis (ITT) – “All patients allocated to

each arm of treatment regimen are analyzed together “as intended” upon randomization, whether or not they actually received or completed the prescribed regimen. Failure to follow this step defeats the main purpose and advantage of random allocation and can cause serious bias.”

–A Dictionary of Epidemiology, 5th Edition

Intention to TreatThe validity of a randomized controlled trial depends greatly on the process of randomization. Randomization insures that both measurable and immeasurable factors will balance out on average. If a factor other than the treatment itself could possibly influence an outcome measure in your study, then randomization provides the best possible insurance that patients with this factor are equally likely to receive either one treatment or the other. This prevents many types of bias that can occur in a non-randomized trial.

Results of RCTs

• Risk of incident event/death• Relative Risk (the same as cohort studies)• Efficacy• Number of patients who need to be treated (NNT)• Survival curves/Cumulative incidence

Results of RCTs

• Efficacy = Relative risk reduction

• A measure of the reduction in risk attributable to the treatment of interest.

= % e.g., if RR = 0.7?

Question about efficacy• In RCTs, we only study factors that may

be beneficial (not harmful)

• Therefore, in calculating efficacy, we will be talking about a risk reduction (i.e., RR<1)

• If it turns out the incidence is higher in the treatment group then we stop the study and the efficacy is considered 0.

Results of RCTs

• Number of patients who need to be treated (NNT) to prevent one event

• Assesses the impact on medical practice of the investigated treatment.

Results of RCTs

NNT =1

0.10 – 0.07=

1

0.03= 33.3

In order to prevent one event, we need to treat 33 patients.

The incidence in the non-treatment group is 10 %The incidence in the treatment group is 7 %

Survival Curves

Estrogen/Progestin and Heart Disease

Grodstein et al. NEJM 1996

141

59

0

20

40

60

80

100

120

140

160

Incidence Rate, per 100,000person-years

Never usedCurrently use

Nurses Health Study: a prospective cohort studyN=59,337; follow-up period: up to 16 years

Age-Standardized Baseline Characteristics of Participants in the Nurses’ Health Study in 1990

Grodstein et al. NEJM 1996

• Women’s health

• Observational evidence

• Hard evidence lacking

• Political reasons

Why Women’s Health Initiative (a RCT)?

WHI. JAMA 2002

Flow Chart of Women’s Health Initiative

Baseline Characteristics

Grodstein et al. NEJM 1996

Estrogen plus Progestin and CHD and Stroke

WHI. JAMA 2002

Estrogen plus Progestin and Colorectal Cancer and Hip Fracture

WHI. JAMA 2002

RCT – GOLD STANDARD

• The randomized controlled trial is an experiment.

• Provides the strongest evidence for inferring causality compared to observational studies.

• Must be wary of poorly conducted studies.

• Study designs

• Uses

• Randomization

• Blinding

• ITT

• NNT

Summary