Skills Lab: Classifying and Defining Surrogate Endpoints

Medscape

07-05-2025

This transcript has been edited for clarity.
Hello, everyone. This is Dr Bishal Gyawali, from Queens University, Kingston, Canada, continuing with our Skills Lab series on how to interpret the clinical trial publication well. In the past few videos, we have discussed the methods section and we touched briefly on surrogate endpoints.
This is a pretty big topic, so we will continue discussing surrogate endpoints today.There is much confusion about surrogate endpoints. There are several questions that people ask about surrogate endpoints. In these videos, we're trying to answer those, but it may not be comprehensive. If you want to get a comprehensive understanding of almost all the questions regarding surrogate endpoints in oncology, I highly recommend you read this publication, which we have recently published and is open access.
Continuing with surrogate endpoints, we discussed in the last presentation about how to differentiate between prognostic and predictive surrogate endpoints. Today, let's talk about how we classify and define them.
We classify them in different ways, and one of those ways is surrogate endpoints in the setting of early cancer drug trials, as in neoadjuvant and adjuvant setting trials, and in advanced-setting trials. Within the advanced setting, we have different types of surrogate endpoints, such as response rate, clinical response, partial response, and overall response rate, as well as progression-free survival.
These are usually defined on the basis of criteria known as RECIST. RECIST stands for Response Evaluation Criteria in Solid Tumors, and that's classically what almost every clinical trial uses to define the surrogate endpoints in the advanced setting.
First, let's talk about the goal of the RECIST criteria. Why was this developed in the first place? It was not developed to make therapeutic decisions or intended for drug approvals. The whole idea around RECIST was to help clinical trials define success or lack of success early on, without the need to wait for overall survival.The goal was to make those decisions about whether these should be tested in a phase 3, from a phase 2, and so on. You can read the RECIST publication in detail, but briefly, to classify under RECIST, you need a target lesion. There are conditions on what can classify as a target lesion and what cannot.
Progression of disease, or PD, is classically defined as at least 20% increase in the sum of the maximum tumor diameters, with an absolute increase of at least 5 mm, or a development of a new lesion that was not there before, or death of the patient. Any of these will classify as disease progression, so we classically keep talking about progressive disease as a 20% increase in the size of the tumor.
Complete response, as the name indicates, is complete disappearance of all lesions. Partial response is shrinkage in the sum of the maximum tumor diameters by at least 30%. Anything in between, meaning the tumor size has decreased less than 30%, or the tumor size has increased less than 20%, is called a stable disease. Anything in between disease progression and partial response is stable disease.
Overall response rate, sometimes called objective response rate,isthe complete response rate plus the partial response rate, so the percentage of patients who had any response is the overall response rate. We also measure the duration of response, which is the duration after achieving response until the tumor starts to grow again.
There are also some other, what I call make-believe, meaningless endpoints. For example, people talk about the clinical benefit rate, which is theoverall response rate plus the stable disease rate. The idea is that if the disease does not progress, then that's a clinical benefit. These do not correlate with any meaningful endpoint. People should not focus on the clinical benefit rate.
Why we're defining this is to see why t hese surrogate endpoints do not necessarily correlate with overall survival or quality of life: (a) because these were not intended to be used as such; and (b) the way we define them is quite arbitrary. As you can see, there is nothing magical that happens at 20% progression or 30% shrinkage of the tumor.
You can imagine that if a person has a disease progression of 18% vs 22%, that's not going to cause any meaningful difference. A disease might grow by only 10%, but if it's at a sensitive location, such as intracranial, then it could lead to some symptoms. Not all disease progression events will be the same; therefore, they do not correlate with clinical endpoints.
Now let's talk about the early cancer trial setting. In the neoadjuvant/adjuvant setting, we have endpoints such as disease-free survival, invasive disease-free survival, event-free survival, metastasis-free survival, relapse-free survival, andpathologic complete response.
Pathologic complete response deserves a special mention because it's used more as a prognostic marker. It's not reliable as a predictive marker, but it's reliable as a prognostic marker. That means, irrespective of how you achieve pathologic complete response, you'll have a good prognosis. That does not necessarily mean that a drug that improves pathologic complete response will improve survival.
Event-free survival is basically disease-free survival, but in the neoadjuvant setting. Of course, it's not feasible to go through the definition of each and every endpoint here, so again, if you want to have a more comprehensive understanding, I refer you back to that publication where all these endpoints are defined.
Another way of defining endpoints is that there can be response-based endpoints, such as the objective response, complete response, partial response, and pathologic complete response that we talked about, vs time-to-eventendpoints.This is important because they are measured and analyzed in a different statistical way.
Time-to-event endpoints are any endpoints that are measured in terms of the time it takes for the event to happen. This includes disease-free survival, progression-free survival, and metastasis-free survival. All of these are time-to-event endpoints.
Let's spend some time talking about response rate because it's getting more and more frequently used in clinical trials and also as a basis for approval, but it's a very weak surrogate. The correlation with overall survival is even weaker than progression-free survival.
The other interesting thing about response rate is that there is no way to define success. If we are measuring response rate in a phase 2 or a single-arm trial, is a 30%response rate enough? Is a 50% response rate enough? Is a 20% response rate enough? In how many patients? In 50 patients? In 40 patients? There is no way to define success here. As I mentioned, this was not meant for approval or clinical use. This was meant for transitioning from phase 2 to phase 3.
In fact, we did a comprehensive study on the response rate of placebo in randomized controlled trials of cancer drugs. You'd expect that placebo would have a 0% response rate, but we were surprised to find that even placebo had a 1% response rate. Complete response rate was, of course, zero with placebo.
When you put that in context with what we're approving nowadays, this is 1% response rate overall. There were some typical examples. For example, I'm showing this trial in renal cell carcinoma where interferon was compared with placebo. In this particular case, placebo had a higher response than the drug. The response from the drug was 4.4% and placebo had a 6.6% response rate.
In another example from desmoid tumors, placebo had an overall response rate of 20%. If you compare that with some of the drugs that we have approved on the basis of response rate, like a 12% response rate in the CheckMate 032 trial, a 15% response rate in the EZH-20221 trial, and a 70% response rate in KEYNOTE-224.
These are not good enough response rates, but we're approving drugs on the basis of these very small response rates. We need to be very careful. I think that, even if we were to rely on the response rate, it has to be in the context of a randomized trial where we're comparing the response rate of the drug against the response rate of the comparator arm. Just relying on the response rate in a single-arm trial is very likely to give us false positives.
Let's also talk about progression-free survival. We talked about progression-free survival in the last video, but I'd also like to mentionthat there are some specific occasions where progression-free survival absolutely cannot be relied on, such as in later-line therapy trials or in situations where the prognosis is so bad that survival events happen within a year.
For example, if we're talking about second- or third-line cholangiocarcinoma, in pancreatic cancer trials, patients have a median survival of 6-8 months — and not even 6 monthsif we're talking about second or third line.
In that situation, there is no sense to use progression-free survival because you can measure survival so quickly, and there is no point in having a drug that does not improve survival. When the prognosis is very short, especially for later-line therapy trials, progression-free survival cannot be used.
For combination therapies, when we are combining drug A plus B vs drug A alone, of course if we combine two drugs, it will extend progression compared with one drug alone. The point to ask is whether combining up front is better than giving the same drug sequentially. Only overall survival can answer that.
For maintenance therapies, these are situations where we continue on a treatment for a long time, where the standard of care actually is to do no treatment. If you're giving an active drug instead of not giving a drug, of course you'll extend progression. Does that lead to improved survival outcomes and should we subject all of our patients for that extended duration of treatment? In these situations, progression-free survival cannot be used.
Let's spend some time on the adjuvant setting as well. In the adjuvant setting, I find it even trickier to rely on a surrogate endpoint, such as disease-free survival, as opposed to overall survival. In the adjuvant setting, by definition, we're treating all the patients, some of whom may not need treatment and some of whom may not benefit from treatment, but 100% of the patients will get toxicities.
In this particular example, in this Kaplan-Meier graph, at the 2-year landmark, 22% of the patients have relapsed or died even after getting the drug. There are 67% of patients who did not relapse even if they only received placebo. We are benefitting only that small margin of patients, that 11% of patients who benefited from getting the drug. Again, in this situation, we should also ask whether the patients who received placebo in this trial actually received the drug when they relapsed and how it affected overall survival.
We had this publication where we advocate to use the approach of three E's — evidence, ethics, and economics — to decide whether or not treating all the patients upfront on the basis of disease-free survival or surrogate endpoints is better than treating only the patients who relapse at the time of their relapse.
That ends our video for today. In the next video, we'll continue discussing some additional considerations about surrogate endpoints.