GRACE :: Cancer Basics


Dr West

All about subgroups: Why do we value the results in some trial subgroups but not others?

download as a pdf file Download PDF of this page

Here’s an important summary of how to interpret results from cancer trials that are increasingly hyped to the general public. You need to be a knowledgeable consumer of this information.

Over the past decade, one of the biggest developments in lung cancer, and in fact many kinds of cancer, has been the identification of important subgroups within the broad categories. Where 10-12 years ago we categorized patients as “advanced NSCLC” and didn’t get much more granular than that, we now routinely look for differences depending on different tumor histologies (adenocarcinoma, squamous, or other), the presence or absence of driver mutations, perhaps even smoking status. We expect to see differences, and in many cases identification of specific subgroups within a larger trial has led to major changes in treatment recommendations. At other times, however, you may see/read cancer news highlights that describe results as being significantly different for one group or another in a trial that was otherwise negative. What makes the results of one subgroup analysis credible and practice-changing, while oncologists view others with far more skepticism?

The biggest factor is whether the subgroup was defined prospectively or retrospectively. For a subgroup to be prospectively defined, this is saying that even if the trial includes all patients with advanced NSCLC, there’s reason to hypothesize that, for instance, results might be different for those with adenocarcinoma vs. squamous NSCLC, or those with a particular mutation or not.  In such cases, the investigators are saying before the trial begins that they are interested in specific groups that make biological sense to identify before the trial is conducted. In some cases, the trial is even done looking for large enough patients in key subgroups to test the hypothesis not only in the overall trial population but in one or more subgroups. The results of analyses from these subgroups tend to be viewed as credible and valuable.

In contrast, retrospective analyses of different subgroups is a way to express curiosity whether some groups might have done better than others, but they were not identified before the trial began. Sometimes this analysis takes the form of asking, in a trial that is positive for a significant effect in the overall population, whether the favorable effect of the new treatment was seen in everyone but not, for instance, in the patients with squamous NSCLC or elderly patients, etc.  Very often, it’s a way to try to salvage something positive from a negative trial and is essentially viewed as a “fishing expedition”.  This means that if a trial is negative, perhaps the investigators can highlight a positive result or two in one of the subgroups.  However, when there is no prospective identification of the subgroups, it can be tempting and very possible to slice and dice the data until you find subgroups that happen to show a difference, perhaps just randomly. If you keep sifting through the wreckage of negative results, changing which groups you care about (maybe looking at those who had a significant response and not just stable disease during first line therapy, or maybe now checking the results in people who quit smoking more than 10 years ago), eventually you’ll find some provocative differences, if for no other reason than just by chance alone. If a statistically significant result is defined as one that occurs by chance alone less than 5% of the time, then if you look for results in more than 20 subgroups, chance alone will lead one to be positive – and now you’ve got a great spin to report on your negative trial!

This isn’t to say that retrospective identification of subgroups isn’t important – for instance, it led to things as important as the finding that patients with squamous NSCLC have a high risk of life threatening or fatal bleeding complications on Avastin (bevacizumab) – but a retrospective subset analysis rarely leads to firm conclusions. This is because working backward really should lead only to refinement of the research question, so we usually need to use the results we’ve cherry picked to help refine a subsequent trial in which the focus of the study is only on that subgroup, or at least that subgroup is prospectively defined and the trial designed to test the value of the new treatment for that subgroup.

It’s very common and appropriate to have clinical research be an iterative process, and it’s very important and valuable to identify relevant subgroups retrospectively. But the key is to recognize that this kind of subgroup analysis isn’t as valid as one that is prospectively defined. When people are asked to think of a number from 1 to 100 for a decision, usually the person asking will write down the number before hearing the answers. If not, how could you believe that being told the person was thinking “39” wasn’t because they like the person who guessed 29 more than the person who guessed 66. This is the same principle. You’re far more credible if you state your expected answer before seeing the results.

I hope that’s helpful. As always, I welcome your questions.

One Response to All about subgroups: Why do we value the results in some trial subgroups but not others?

  • ssflxl says:

    Dr. West

    Thanks for addressing this subgroup issue that I brought up in my vaccine question. I think it’s always great to be prospective and predictive, but sometimes you don’t actually know what subgroup would benefit, and you need to do retrospective analysis. sometimes new “subgroup definitions” come up in these trials. you don’t have a crystal ball to set up the trial looking at certain subgroups. In that case, you need to rely on retrospective analysis.
    Secondly, would a retrospective analysis of subgroup be more convincing if the numbers are very different – not just a tad higher, but quite a bit different.



Leave a Reply

Ask Us, Q&A
Cancer Basics Expert Content


download as a pdf file Download PDF of this page

GRACE Cancer Video Library - Lung Cancer Videos




2015 Acquired Resistance in Lung Cancer Patient Forum Videos

download as a pdf file Download PDF of this page

Join the GRACE Faculty

Lung/Thoracic Cancer Blog
Breast Cancer Blog
Pancreatic Cancer Blog
Bladder Cancer Blog
Head/Neck Cancer Blog
Kidney Cancer Blog
download as a pdf file Download PDF of this page

Subscribe to the GRACEcast Podcast on iTunes

download as a pdf file Download PDF of this page

Email Newsletter icon, E-mail Newsletter icon, Email List icon, E-mail List icon

Subscribe to
   (Free Newsletter)

Other Resources

download as a pdf file Download PDF of this page

Biomedical Learning Institute