Ultimately new treatments must undergo testing in people before they can be marketed as new drugs or devices. Testing stuff in people raises all sorts of ethical issues. No matter how many animals have been cured by your next great thing, people may be a bit different and there will be risks, known and unknown. The evolution of these protections can be read, in beautiful moving prose, in The Immortal Life of Henrietta Lacks: internal review boards (IRB), informed consent, and other protections have developed over time since the end of World War II and the revelation of Nazi atrocities and experiments.
Clinical research initially seems pretty simple. Get people, give them a new treatment, and see if they get better. Unfortunately, patients in experiments (or not) may get better spontaneously, especially if their doctors expect it. The placebo effect led to the double-blind clinical trial. Neither the subjects nor the investigators know who is getting the test agent, eliminating a major source of bias in outcomes.
While masking doctors and subjects to study conditions removes bias, it also removes some safety. What if the treatment has an unexpected side effect, especially a deadly one? What if the treatment is more effective than imagined and a significant effect can be demonstrated with only half the original number of subjects? In either of these cases, it would be unethical to continue the trial, but the investigators cannot unmask the data without ruining an ongoing study. As large, multi-center studies became more common in the late 1960's the data safety monitoring board (DSMB) became a common feature of such trials.
Clinical trials can be the brain-child of an investigator, although many are conceived of by the sponsor, especially in the case of pharmaceutical trials. Large trials frequently involve a steering committee, a group of investigators who devise the study and supervise its performance, analysis, and publication. In general, everyone in the orange boxes in the figure will be masked to subject assignment. If they break the code, then the study must be terminated.
Qualified investigators also compose the DSMB. The committee must include a statistician and physicians with expertise in the disorder under study and potential side effects. The statistician accesses unmasked data periodically and the committee examines the results. Is recruitment going as expected? If patients are stratified for a characteristic, is balance being achieved? Are side effects or mortality occurring in excess in one of the treatment groups? Any of these areas could lead the DSMB to recommend premature termination of the study to the investigators, steering committee, or sponsor. The DSMB also reports to the IRB who can recommend premature termination as well.
A recent article discussed three large clinical trials in which data were unmasked prematurely by the sponsor or investigators, without consulting the DSMB and/or steering committee. I have discussed this article and some of these issues elsewhere.
Bottom line: The DSMB helps protect subjects from a number of outcomes, including undue risks of bad outcomes, missing out on an effective treatment, or being a subject in a trial that ends before the answer is really found.