Understanding Overall Survival and Response Rate - 1245602

Sat, 07/21/2012 - 02:34

When I read that a study of 1st line chemo treatment shows a response rate (RR) for FOLFOX of 56% and an overall survial (OS) of 24 months, can I assume that the overall survival statistic of 24 months includes the death rate of the 44% who showed no response?

And if this is true, wouldn't it make more sense to also present layered results and look at the overall survival of those who showed complete response, partial response, and no response as it seems these statistics would be radically different?

Revision log message
Created by FeedsNodeProcessor

Dr West

I'd actually add a different perspective. The median survival includes patients who do well, do "average", and who do poorly. You can sometimes see "layered" results, but it's not a preferred way to report results. Why? Because all trials show that people who get better results do better, and because we don't know beforehand who is going to respond and who won't (it's not statistically valid to siphon off patients who aren't doing well and then preferentially report on those doing better).

-Dr. West

dr. weiss

Generally, the three major measures of tumor control are response rate, PFS (progression-free survival) and OS (overall survival). Measurements of side effects are at least equally important.

The relationship between RR, PFS and OS has changed over time and opinions on the “best” measure have also varied. Traditionally, new drugs were approved by response rate. Back when carbo/taxol was still THE standard of care for NSCLC, it was approved on response rate. RR measures the % of patients who have at least 30% shrinkage of their cancer. But is it the most important measure for all, or even most patients or is it a so-so surrogate for other things that matter more?

I believe that only two factors really matter—duration of life and quality of life. Traditionally, the drugs that had the highest response rates had the most activity against cancer. By controlling cancer growth, they improved duration of life. Similarly, while cancer drugs have very significant side effects, lung cancer is so brutal that they improved quality of life by stopping cancer-related suffering more than they harmed it by causing side effects. But RR was always a poor surrogate. Imagine two treatments. One is very aggressive and gives huge response rates; for the sake of the example, let’s say that the average patient has 50% shrinkage. But, patients can only take a short duration of therapy, and the responses are short-lived, say 3 months, before the cancer starts rapidly growing again. In contrast, another treatment has 0% response rate (granted, I’m picking an extreme example just to get a point across) because all patients have only 5% shrinkage. But, for this treatment, it last much, much longer—it’s a full year before the cancer grows again. For most patients, this second treatment would be better even though its RR is lower. The notable exception would be the patient with terrible pain not responsive to pain drugs, for which response is needed urgently to improve symptoms.


In asking the question one thing I was trying to get at is whether one could, at the end of 1st line therapy, modify one's expectation for OS duration upward (from the original 24 months) based having a complete or partial response considering that nearly half of patients have no response and pulled the average down. I was assuiming, of course, that the no response group had a significantly shorter OS.



It is certainly a great start. Congratulations Peter!

In a blog/post here http://cancergrace.org/cancer-101/2012/03/17/the-principle-of-letting-t… Dr. West ended the discussion with,
"In fact, in all sorts of cancer settings where the best treatment approach is ambiguous, starting with a good systemic therapy is typically a very good approach that burns no bridges and allows us to learn more about the pace of the cancer and its sensitivity to good treatment. These factors are critical in refining prognosis and helping to guide us in the recommendation for a management approach that won’t represent undertreatment or overtreatment."

I hope this helps you Peter and njliu,

Dr West

There are individual studies that show results for particular populations, but this reference is really the latest that describes staging breakdown in NSCLC:


The publication is five years old, and it uses data that go back well before then, but otherwise you're looking at results in one selected group of people or another, so it's hard to consider these results generalizable to a wide audience.

-Dr. West

dr. weiss

I apologize for the delay in posting this--I wrote it on a plane on Sun night and meant to post on landing, but got side tracked:

I’m sitting on a plane from Columbus Ohio back home to NC. During takeoff, I used the time to catch up a bit on my journal reading. Just 10 minutes after hitting “send” on my last post, I read a relevant editorial that I’d like to share. It’ sby Christopher M. Booth and Elizabeth A. Eisenhauer and was published in JCO in Apri.

First, the authors point out that while advanced colorectal and advanced ovarian cancers have data specifically linking PFS as a surrogate for OS, breast, prostate and NSCLC do not. They specifically cite a NSCLC lung study that was negative on this subject (I vaguely recall reading the article a year or two ago, but don’t want to comment further until I have a chance to review it. I’m still on the plane without internet and suspect that while I will have time to copy/paste this post into GRACE before bed, I won’t have time to critically evaluate an article).

The authors also make some philosophic points about why a PFS improvement might not translate to OS:

1. Sometimes the change in tumor size required for progression (by definition, 20%) may be so small (especially in tumors that are small to start with) that progression doesn’t have an impact on time until death.
2. Statistically, it is sometimes easier to reach a PFS endpoint than an OS endpoint (differences in PFS tend to be bigger than differences in OS on an absolute basis). On the other hand, with sufficient followup, a PFS difference should eventually show up an OS difference if there truly is a relationship.
3. PFS differences are subject to more biases than OS differences. Time to death is very objective. On the other hand, progression is subject to biases such as time of evaluation and human judgment re tumor sizes.
4. A drug may really delay progression, but could at the same time make a tumor biologically more aggressive.
5. Post prog. Rx