Cancer Trials Terminology: Study Design & Stats

February 7, 2007 - 10:20 pm email to a friend

Print This Post view / write comments

(No Ratings Yet)

Loading ...

Dr West

Like most medical specialties, oncology is part art and part science. There’s plenty of room for individualizing treatment plans, but as a specialty we try be evidence-based. These are treatments that can be very helpful for patients, but also can have significant side effects, so we want to be guided by as much information as possible about the anticipated risks and benefits of treatment. I’ve been using the terminology of oncology trials throughout all of these discussions, so I wanted to take some time to discuss what the terms mean and how to interpret a survival curve. Whether here or from other sources, what you may read can have details that are not necessarily obvious. Here are some of the basics of oncology terminology.

First, trials can be retrospective, which means looking back at results of patients being treated a certain way, or prospective, which means that patients who have a similar cancer and stage are assigned a uniform treatment plan. Prospective trials are generally more informative, but retrospective reviews of information can provide good hints of whether certain patients respond well to a treatment, for instance, or whether others with a certain tumor histology develop a particular side effect.

As I mentioned in my discussion of drug development, phase III trials are randomized, which means that there is essentially an electronic coin flip between treatment A and treatment B. Usually in phase III trials we are testing a new approach vs. the prior standard treatment. Trials can be open-label, in which the doctor and patient know exactly what treatment is being given, single-blinded, in which the doctor/medical team know the treatment but the patient does not, or double-blinded, which is when neither medical team nor the patient know the treatment a patient is getting. Double-blinded trials generally include a placebo, an inactive IV or pill that appears indentical to the active medication. This is to clarify whether the differences between arm A and arm B are truly because of the drug or because of the placebo effect, which describes the range of effects people ascribe to a drug even when it has no active properties. This can be important for many reasons, because patients with progressing cancer may feel increasing pain or cough or fatigue that they ascribe to a new medication rather than to the underlying disease. By the same token, coming off of harsh chemotherapy can leave people feeling better, so a trial of a new treatment that starts after completing challenging treatment may leave people feeling better because of the new drug or just because they’re not doing the harder treatment anymore. Finally, there’s a potentially powerful psychosomatic effect from taking a drug that everyone believes is going to be the next great thing. A placebo helps determine what the active drug is really doing.

(click to enlarge)

Patients are often reluctant to be enrolled on randomized trials, particularly where the other arm includes a placebo. These trials are randomized because we truly don’t know whether the arm with the active drug is superior to the other arm. Sometimes, the active drug shows no benefit and has significant side effects, and sometimes the arm with the active drug has a worse survival than the placebo arm. So the only way we can clarify whether the new approach is better or worse than our standard approach is to do a prospective, randomized trial. Sometimes randomized trials include a cross-over design, in which patients on the placebo arm can receive the active drug after they demonstrate disease progression.

A few other definitions are useful. Stratification means that the randomization procedure takes into account certain variables that are programmed in, to ensure that the two arms are balanced for important clinical factors such as patient sex, smoking status, etc. Sample size is the number of patients the trial is designed to enroll on a trial, usually designated with the symbol N. The larger the trial, the greater its ability to detect a significant difference between the arms. Small trials may have very large differences in how two treatment arms perform, but with small numbers, there is such a strong component of chance that the results may not be statistically significant despite the large differences in outcome. Statisical significance is designated by the symbol p, which is the probability that the results could occur by chance alone. For instance, a difference in survival of 2 months between treatment A and treatment B may have a p-value of 0.05, which means there is a 5% chance the difference could be just by luck and not because of the efficacy of a drug. The standard convention is to define statistical significance at p = 0.05, so if a result has a p value of 0.05 or lower, translating to a 5% or lower likelihood that the differences could be achieved by chance alone, this is a convincing and meaningful difference in outcomes.

The results of our trials are described in various ways. There are a few grading scales for tumor shrinkage, but shinking in volume by usually around 50% is a threshold for an objective response. This can be a partial response if it exceeds that threshold but still shows evidence of disease, or a complete response if there is no evidence of disease. Remission is the same general concept as a complete response, but that term is generally used for “liquid tumors” like leukemias or lymphomas (tumors of blood cell origin), and we more commonly describe a similar situation in solid tumors (like lung, colon, breast, prostate, etc.) as no evidence of disease (NED), or no evidence of residual disease (NERD). On the other side, there is progressive disease, which is a finding of a new cancer lesion or growth beyond a threshold of approximately 30% in volume overall. In between an objective response and progression is stable disease.

We also care greatly about survival differences. Overall survival is the percentage of people who remain alive from the start of a trial to a defined time point, such as one or two years. We also report median overall survival, which is the point at which half of the patients on an arm have died. This is not an average, which can be distorted by a few people doing remarkably well on the tail of a curve (so a few people who are cured can skew the results powerfully as follow-up increases). There is also progression-free survival, which is like overall survival in that it is the percentage of patients who remain without disease progression at a particular time point, and this is also often presented as a median value. Disease-free survival is the duration of time in which no disease is evident. Time to progression is similar to progression-free survival, except that time to progression is measured in time, and progression-free survival is a percentage of patients who remain without progression. Does anyone’s head hurt?

Lastly, there are survival curves, also known as Kaplan-Meier curves, and hazard ratios. A huge proportion of our oncology results are presented as survival curves, which show the proportion of patients on a trial still alive, or still without progression, over time, progressing from left to right. A plateau on the curve toward the right side represents potentially cured patients, where we aren’t seeing any further deaths or progression with ongoing follow-up. We all like to see plateaus on survival curves, and the higher, the better. Gaps between two curves on such a graph represents the improvement in survival (overall or progression-free) between one treatment approach and the other. The overall space between two curves, over the entire time of follow-up, is captured by the hazard ratio, which is a decimal that if below 1 represents the improvement in survival on an investigational approach, while a number greater than 1 represents a worsening of survival with the investigational approach. For example, if there is a hazard ratio of 0.6 with a new agent compared to placebo, this represents a 40% improvement in survival. Conversely, a hazard ratio of 1.25 means that there was a 25% worse survival with the new agent.

That covers the highlights. If there are gaps and particular questions I didn’t cover, please chime in.

We’ll get back to lung cancer topics next. This was just a little detour to make sure everyone is equipped with enough of the lingo to follow these stats-heavy discussions.

Bookmark to: