Stefano Garzon

Research · Oct 23, 2026

Data monitoring committee: structure and function

Concept

The data monitoring committee (DMC, also called DSMB) reviews a trial’s accumulating data while the trial is ongoing. It recommends continuation, modification or termination. Interim monitoring has ethical, scientific and economic grounds: harm from the intervention, clear benefit, futility and data quality problems.

Friedman prefers the term DMC because the task is to weigh risks and benefits, not safety alone.

Independence

An investigator who knows the trend of the data has difficulty continuing to enroll and treat participants without bias. In later phase trials, therefore, those who monitor should have no involvement with participants or investigators. Independence also protects the committee from pressure by investigators and sponsors.

Local ethics committees receive isolated serious adverse events and cannot assess accumulated evidence by treatment group. That role falls to the DMC.

Early phase, single-center or simple-intervention trials may not need an external committee.

Membership

Membership is multidisciplinary: clinicians in the field, people experienced in trials, epidemiologists, biostatisticians and, in NIH-funded trials, often a bioethicist or patient advocate. No single statistical procedure settles a stopping decision, and no single person has all the required experience.

Priorities, in order, are participant safety, investigators and ethics committees, the sponsor and the regulatory agency.

Meeting format

In industry trials, a statistical center independent of the sponsor often prepares the interim analyses. This reduces the chance that interim results circulate within the sponsor. Regulatory agency representatives generally do not attend the closed session.

Blinding of the committee

Labelling groups as A and B, with unblinding when trends emerge, is acceptable. Using different codes for each table (A/B for baseline, C/D for efficacy, E/F for toxicity) prevents a joint assessment of risk and benefit. Friedman does not recommend this practice.

Frequency

Long trials often meet every 4 to 6 months, with extra meetings as needed. Another option is to convene at prespecified fractions of expected events, for example 10, 25, 50, 75 and 100%. Friedman recommends a final meeting with the complete data before publication.

Repeated testing

Testing the null hypothesis repeatedly at the 5% level raises the false positive rate: about 8% with two tests, 14% with five and almost 20% with ten. In the Coronary Drug Project, clofibrate crossed the nominal 5% threshold on five occasions, and the final mortality curves were nearly identical. Formal methods are covered in interim analyses and early stopping.

Factors in the decision

Before stopping, the committee considers baseline imbalance, assessment bias, missing data, adherence and concomitant interventions, secondary and safety outcomes, consistency across subgroups and centers, trends over time, results of similar trials, and credibility of the result.

In CAST, encainide and flecainide increased total mortality (56 of 730 vs 22 of 725; relative risk 2.5). These arms were stopped with a small fraction of the expected events. In TRACER, with 12,944 patients, vorapaxar did not significantly reduce the primary outcome (HR 0.92; P = 0.07), and intracranial hemorrhage was 1.1% vs 0.2% (HR 3.39). Follow-up was terminated after a safety review. In ILLUMINATE, torcetrapib increased cardiovascular events (HR 1.25) and death (HR 1.58), and the trial was stopped.

Common errors when reading a paper

What to check when reading a trial

  1. Presence of an independent DMC and its membership.
  2. Who prepared the interim analyses and whether they were separated from the sponsor.
  3. Number and timing of interim analyses and the prespecified stopping rule.
  4. Reason for stopping, if any, and the number of events at that time.
  5. Consistency of the result across outcomes, subgroups and centers.

References