In the last post that presented the highlights of the FLEX trial that tested the benefit of adding Ebritux (cetuximab) to standard chemo. The trial was technically positive, statistically significant, but the results in the overall population of the trial were so marginally superior with Erbitux that a very logical follow-up question is whether we might identify certain subgroups of patients who benefit more, enough to definitely add Erbitux, while not pursuing it for other subgroups that appear to benefit much less.

It's important to add a word of caution about interpreting information gleaned from patient subsets. Clinical trials are generally designed to have enough patients to show differences in the entire trial with everyone. Nevertheless, when you dissect it ten different ways, the smaller subgroups don't have adequate numbers to show significant differences, even when they may really exist. At the same time, doing multiple different comparisons escalates the chance that you'll randomly find differences that aren't real, and that occur just as a product of random chance. In fact, trials are generally powered to consider a difference as significant only if the probability of the difference happening by chance is less than 5%, but if you slice and dice the results to do ten different subgroup tests, the likelihood of at least one coming up positive just by chance is now 22%.

So these comparisons aren't really conclusive and are better suited for shaping our ideas for future research than for guiding our treatment decisions. In reality, people (present company included) tend to pick and choose which subgroups they focus on to support their inclinations, while discarding other comparisons.

To provide the general picture, there's a figure called a forest plot that shows the performance of multiple subgroups simutaneously. Here, the size of the different subgroups is represented by the size of the black ovals, and the position of the oval to the relative to the vertical line shows whether the Ebritux group did better (if the oval is to the left of the vertical line, sometimes referred to as "unity") or the chemo alone arm did better (if the oval falls to the right).

FLEX Trial Forest Plot

(Click on image to enlarge)

You can see from the figure above that the majority of the groups had a modest benefit with Erbitux, falling just a little to the left of the vertical line, but there were some groups that seemed to do a little better or worse with Erbitux, so we'll talk more about some of them.

The trial had a pre-planned division by patient race, since we have come to recognize that the behavior of lung cancer in Asian patients can be quite different from that typically seen in Caucasian patients. But the side by side comparison of the Asian patients, who constituted 11% of the trial population, with the Caucasian population, who comprised 84% of the patients, highlights how very different these groups really were:

FLEX Trial Asian vs. Caucasian Populations

The Asian group had far more women, overwhelmingly had adenocarcinoma histology subtype, actually had more never-smokers than former or current smokers, and had a particularly good performance status. They were also much more likely to be offered an oral EGFR tyrosine kinase inhibitor (TKI) like Tarceva (erlotinib) or Iressa (gefitinib). I imagine that this is because oncologists were far more likely to still recommend an EGFR inhibitor for Asian female never-smokers than for Caucasian current or former smokers.

But what is most striking is the fact that Asian patients had a median survival that was twice as long as the survival in Caucasians (19.5 vs. 9.6 months). It's not clear whether that's because more Asians received an oral EGFR TKI, whether the underlying biology of their cancer was just much more favorable, or some combination. Nevertheless, the study also noted that median overall survival numbers within the Asian population trended toward being better for those who didn't get Erbitux (20.4 vs. 17.6 months, not a statistically significant difference).

Was Erbitux actually harmful for Asians? It's possible, but I think it's more likely that the reason they appeared to do worse with Erbitux was that oncologists were less likely to give them another EGFR inhibitor, namely an oral EGFR TKI, after Erbitux (73% received it after chemo alone, vs. only 50% after chemo/Erbitux), as they presumed that a second drug against the same EGFR axis wouldn't be helpful. I believe that may be a mistake and that the patients who were going to do well with an EGFR TKI might have still done well with it after Erbitux, and I wouldn't deny them that opportunity.

The investigators had said before the results came in that they were particularly interested in the results in Caucasian patients, independent of how other races did. Here, the survival curves separate earlier and show some median overall survival differences that appear somewhat more compelling than the 1.2 month difference in the entire trial population:

FLEX Trial Caucasian Results

As shown in the table included above, the benefit in Caucasians was pretty comparable in patients with squamous vs. adenocarcinoma NSCLC histology. The forest plot at the beginning of this post makes it appear that the benefit is clearly less in patients with an adenocarcinoma, but that must be a by-product of these results being pulled by the non-benefit of Erbitux in Asian patients, who almost all had adenocarcinomas. For Caucasians, the absolute benefit with Erbitux corresponding to an approximately 20% relative improvement in survival vs. chemo alone, and in the same ballpark as what we'd seen with Avastin.

You'll note that the benefit in the forest plot looks like it's less in men and those with a good performance status than for women or the minority of enrolled patients with a marginal performance status, respectively. Not surprisingly, we've never seen slides from the company that show survival curves by sex, and I'd presume that's because they don't want oncologists to think that maybe there's no value of Erbitux for men, who still comprise the majority of lung cancer patients, or for good performance status patients.

So this is how subgroup work gets used selectively. Companies slice and dice the full picture, and then we see a tendency to highlight the pieces that are most encouraging, and the pieces that don't support the conclusions they'd like to leave people with are just kind of ignored, with a hope that nobody will notice and ask about them.

People will tend to use subgroup information in different ways. I think this information is most useful when there's a plausible biological explanation for differences and when they're seen in several similar trials. For me, it's encouraging to see that the benefit with Erbitux is comparable for patients with squamous NSCLC to that seen in patients with an adenocarcinoma. I think for most oncologists, Avastin is their leading targeted agent they'd add to chemo for a patient with an adenocarcinoma and that the leading question is whether the benefit with Erbitux is enough to recommend it for Avastin-ineligible patients.

There's still more to cover about Erbitux for NSCLC, including the association of rash with patient outcomes and the predictive value of molecular markers, but we'll cover these topics later.