Skip to content

The Evolution of Clinical Trials

Boiling oil versus egg yolks

One early clinical trial was accidental. In the 16th century, a common treatment for wounded soldiers was to pour boiling oil on their wounds. In 1537, the surgeon Ambroise Pare, attending French soldiers, ran out of oil one evening. He tried a substitute: egg yolks, turpentine and oil of roses. The next morning, he found the soldiers treated with boiling oil were (unsurprisingly) writhing in pain, while the soldiers treated with his ad-hoc substitute were doing better.

Dr. Arun Blatt’s 2010 article in Perspectives in Clinical Research recounts other early precursors to the modern clinical trial, including one documented in the Bible (book of Daniel). King Nebuchadnezzar of Bablyon had directed a meat and wine diet for his subjects, but several of his relatives objected and he allowed them to eat only vegetables and water for 10 days. At the end of the time they appeared better nourished than others, and were permitted to continue to eat vegetables.

1946 – Streptomycin

In 1946, the British Medical Research Council conducted a randomized clinical trial of the efficacy of streptomycinin the treatment of tuberculosis (TB). Before the advent of streptomycin, TB was was a major killer (25% of all deaths in Europe in the 1800’s), and the prevailing medical treatment at the time of the streptomycintrial was to surgically induce the infected lung to collapse.

The successful 1946 streptomycintrial is commonly regarded as a landmark in the development of clinical trial methodology, combining together for the first time the elements of randomization, meticulous recruitment and careful specification of treatments, including a control treatment.

A 1943 British experiment with a common cold remedy also came close to the standards for a modern-day randomized controlled trial (RCT). Instead of true randomization, the study alternated treatment-control on a patient-by-patient basis. This was not a huge departure from the eventual standard of randomization, but the study did not achieve the noteworthiness of the streptomycin study, partly because the remedy proved ineffective.

Both the 1943 and 1946 studies incorporated the concept of blinding. In a single blind study,the physicians and other care-givers are aware of who gets what treatment (active or control), but the patients are not. In a double blind study, the physicians and care givers are also unaware of who gets what treatment. The 1946 study was the first, though, to set out a rigorous rationale and method for randomization, to avoid bias.


The idea of a placebo, or inactive, treatment was traced by Bhatt to an 1863 experiment in the U.S. Physician Austin Flint treated 13 patients suffering from rheumatism with an herbal extract (presumed to be medically inactive), comparing results to those of the established remedy. There are two competing strands of thought which can confuse discussion of placebos.

  • The scientific community recognizes the placebo effect, in which any treatment, no matter how chemically/medically ineffective, can command a certain improvement just by virtue of the patient perceiving that he or she is being treated.
  • The clinical trial administrator, by contrast, is interested in testing the effect of a specific chemical compound or treatment protocol. They include a placebo as the “no treatment” option for the control group, so that the only thing that distinguishes the treatment group from the no-treatment group is the active component of the treatment.

Therefore, the result of a controlled clinical trial is the incremental improvement of the active component of the treatment over the placebo component.


In 1957, the German pharmaceutical company Chemie Grünenthal developed and brought to market a drug with purported wide applicability – it was advertised to treat morning sickness, anxiety, insomnia, and other conditions, and sold over the counter. At that time, though it had been 10 years since the Streptomycin trial, effective drug relation was not in place. Over 10,000 babies were born with birth defects caused by thalidomide; half of them died.

The US Food and Drug Administration (FDA) had been in existence for nearly 100 years at the time, and was spurred to develop mich tighter rules for establishing both the efficacy and safety of new treatments under study. The FDA and other regulators can establish a set of principles and rules governing clinical trials and drug development, but they cannot place staff in all the pharmaceutical labs and medical clinics to observe procedures and enforce rules.

Instead, since the 1980’s, the industry itself has created independent Data and Safety Monitoring Committees (DMC’s) to oversee clinical trials. In the words of Jay Herson, author of Data and Safety Monitoring Committees in Clinical Trials and instructor for the course Independent Data Monitoring Committees in Clinical Trials here at the Institute, the members of such committtees should consider themselves

“…responsible for the stewardship of the trial. This implies both the preservation of credibility of the trial and the aegis of patient safety.”

Note the dual emphasis:

  • integrity of the data
  • safety of the patient

Trial Phases

Modern clinical trials with drugs are typically divided into phases.

  • Phase 1: Small scale studies to determine if the drug is safe, and to establish appropriate doses
  • Phase 2: Larger scale studies to determine if the drug is effective, and to continue to evaluate safety and side effects
  • Phase 3: Large scale, longer studies to formally establish efficacy with sufficient evidence to submit the drug to regulators for approval

Phase 4 is also referred to, though it does not have the character of an experiment – it involves the post-market monitoring of adverse events.


The most recent chapter in clinical trial design concerns flexibility. Early on, trials specified a very specific design – so many patients, such and such dosages, seeking a treatment effect of a certain size. This was important to avoid selection bias. Selection bias occurs when you choose data in a way that adds bias, often to confirm a hypothesis you are interested in. For example, if you do not specify a certain number of patients for the study, you might feel free to stop the study at a point when the data, just owing to the vagaries of random variation, just happen to lean strongly in favor of your desired result. A naive statistical analysis of just that final snapshot might, incorrectly suggest statistical significance. An inflexible study design prevents the investigator from putting a thumb on the scale in this fashion.

However, inflexibility also leads to inefficiencies and can result in effective treatments being withheld from the public longer than necessary. If a treatment is clearly superior after results with 200 patients, then it is not right to delay its introduction until all scheduled 400 patients have been recruited and evaluated. In the last two decades, academic research, followed by commercial development of products and services, then finally by regulatory acceptance, has yielded more flexible clinical trial designs that rest upon a solid statistical foundation.

You still must have a design, but, for example, the design can allow for multiple “looks” at the data, at specified times, to assess whether a treatment is effective or ineffective. A trial can be stopped early if it demonstrates either, but the standard for statistical significance is correspondingly more stringent than if you just had one inflexible stopping point.

The leading agent in the development of flexible trial design software and services has been Cytel Corp., led by Cyrus Mehta and Nitin Patel. For an excellent tutorial in flexible trial design, see the Institute’s course Adaptive Designs for Clinical Trials, developed and taught by Cytel statisticians. (Historical note: Nitin Patel was instrumental in launching – he helped teach the initial data mining courses at the Institute, and co-authored the book Data Mining for Business Analytics that underlies five current courses at the Institute, including the Predictive Analytics sequence).

For an excellent introduction to the subject, consider the course Introduction to Statistical Issues in Clinical Trials. It is taught by Nand Kishore and Vidyadhar Phadke, both senior biostatisticians at Cytel.