From a Clinical Perspective

Enrichment Design Studies Should Enhance Signals of Effectiveness

by Robert Temple, M.D., Deputy Center Director for Clinical Science

Enrichment is a broad term used to describe attempts to find a study population in which the effect of a drug can be most readily demonstrated -- if, in fact, the drug is effective. People have been using enrichment approaches since clinical trials were first conducted and there are many practical enrichment maneuvers that are common in clinical trials. In fact, CDER is in the process of writing a guidance document to help people better understand the topic. What has changed in recent years is a growing ability to find individualizing characteristics of patients, like genetic characteristics, that can permit these kinds of enrichment.

There are basically three kinds of enrichment: noise reduction, prognostic enrichment, and predictive. Enrichment won’t save a drug that doesn’t work but it will help find one that does.   

Noise Reduction

Noise reduction is the elimination of needless variability and of what seems to be therapeutic responses that are not caused by the drug. Noise reduction is one of the variety of ways researchers try to include people in trials whose responses can be measured precisely and correctly so if they have a drug effect it can be detected. 

For example, at the start of trials it’s common to have a period of treatment with a placebo when testing antidepressants or other drugs where there is a large response in a placebo group. If you can eliminate people who have a significant response that is not drug-related then the difference between active treatment and the placebo will be easier to detect. If everybody’s disease goes away in the placebo group there will be nothing left for the active drug to do, and therefore no  ability to show effectiveness in the treatment group. Whether the improvement in a placebo group is truly a placebo response or is just a population that was getting better spontaneously does not really matter. If there is too much improvement, there is no disease left to treat and no effect can be demonstrated.

Sometimes measurements can be done precisely in people and sometimes they cannot. You can screen a population for an antihypertensive trial to eliminate people whose blood pressure is very variable. That variability will make it difficult to show an effect of the drug. Another example is to try to identify people who will comply with the therapy, such as by testing compliance in a lead-in period and randomizing only good compliers; it’s not always easy to do this, but it can make a difference.   

Prognostic Enrichment

The second major method of enrichment is called prognostic enrichment. This mostly applies to studies where you are trying to show that a drug reduces a bad outcome, like heart attack or death. In order to succeed you need a population that has a reasonable number of these events. If they’re too healthy, the group won’t have any events and your drug will look like it doesn’t do anything.

One of my favorite examples is the first study of an angiotensin converting enzyme inhibitor in heart failure that showed an effect on survival. It was called the Consensus Study and was done with a drug called enalapril in a population with New York Heart Association Class IV heart failure. The NYHA Functional Classification provides a simple way of ranking the extent of heart failure. It places patients in one of four categories based on how much they are limited during physical activity. Class IV patients are unable to do any physical activity without discomfort. These were very sick people. In fact, the mortality rate during the six months of the trial was more than 50 percent.
In this study of only 253 patients, it was possible to show that enalapril decreased mortality by about 40 percent. The study was able to show results in a small number of patients because the event rate was so high. Later studies in less ill people needed many thousands of patients.

It’s important to recognize that the study group didn’t necessarily have a bigger percent effect from the drug than the less sick people studied later. But since the study group had lots of events, it showed -- with a relatively small population -- that the drug helped.   

Predictive Enrichment

The third kind of enrichment is predictive enrichment -- trying to find a population that responds to the particular drug in question. One way to do this is by testing a population first, identifying patients who respond, and then randomly allocate  those patients to the drug and placebo. This is called a randomized withdrawal study. It uses a population that has responded so that it should be  very capable of showing that the drug works. This approach doesn’t tell you about the effect in the overall population, nor does it define who the drug works in, but it does tell you that in at least some population the drug really works. In many cases, this is the first thing you want, and need, to know. 

But what has everybody really excited is the possibility that you can find a genetic or physiological characteristic that predicts response to a particular therapy, the real hope of rational individualization of therapy.

There have been some recent examples of this. A number have to do with cancer which is, in a sense, a genetic disease (tumor cells often have genetic changes that account for their uncontrolled growth.) It has become possible to identify particular characteristics on the surface of the cancer cell or sometimes genetic characteristics that predict whether a particular kind of drug will treat that tumor. 

A classic example is the drug herceptin. When used for breast cancer it works in people whose tumor cells have high levels of the HER2-neu receptor but it doesn’t work nearly as well in people who don’t have that receptor. Studies of herceptin in patients with high levels of the Her2 receptor showed a far greater response than would have been seen in an unselected population. 

Another example is a recently-approved drug for cystic fibrosis that reverses the genetic defect that causes cystic fibrosis. The drug, Kalydeco, works in only a small fraction (about 4 percent) of the people with the disease who have a particular genetic abnormality. Although the drug works in only four percent of patients with cystic fibrosis, it has a dramatic effect in that population. A study in unselected patients with cystic fibrosis -- with only 4 percent responders -- would have had great difficulty detecting a response.

There are also several new treatments for Hepatitis C. They work faster and better than the previous treatments, but only work in patients with Hepatitis C Type 1; they don’t work nearly as well in Types 2 and 3 but there are other drugs under development to treat these conditions. 

These examples are all called predictive enrichment because they use  a population that will respond much better than an unselected population, making detection of an effect much easier and avoiding treatment of people who cannot respond but could be harmed.

Now, there are some issues with predictive enrichment, especially being reasonably sure the predictor does what it is supposed to do. That is, you want the enrichment characteristic to predict the good responders, and correctly identify people (the unselected patients) who willnot respond (you don’t want to miss people who could benefit) You want test to be very accurate – find all the people who will respond and not miss any who could benefit, while also identifying people who will not respond. But no test is a perfect predictor so it’s very important fully to characterize the test you use to identify responders.

An issue to consider in any enrichment design is how much you need to study the people who don’t seem to have the enrichment characteristic. This is something to be worked out over time. But identifying a population that is more likely to show an effect is a wonderful first step. is important. Predictive enrichment comes closest to what people hope for when they refer to individualization of therapy. Enrichment design studies help you reach this kind of individualization. 

Bob Temple is currently the Deputy Center Director for Clinical Science at the FDA. He completed his residency at the Columbia Presbyterian Medical Center.

###

Robert Temple serves as CDER’s Deputy Center Director for Clinical Science and also Acting Deputy Director of the Office of Drug Evaluation I (ODE-I). He has served in this capacity since the office's establishment in 1995. 

Dr. Temple received his medical degree from the New York University School of Medicine in 1967. In 1972 he joined CDER as a review Medical Officer in the Division of Metabolic and Endocrine Drug Products. He later moved into the position of Director of the Division of Cardio-Renal Drug Products. 

In his current position, Dr. Temple oversees ODE-1 which is responsible for the regulation of cardio-renal, neuropharmacologic, and psychopharmacologic drug products. Dr. Temple has a long-standing interest in the design and conduct of clinical trials. He has written extensively on this subject, especially on choice of control group in clinical trials, evaluation of active control trials, trials to evaluate dose-response, and trials using “enrichment” designs.