I recently wrote a post about combining chemo and radiation after surgery for unsuspected stage IIIA N2 node-positive NSCLC (if discovered prior to surgery, it is more common to administer chemo and/or radiation before surgery). In the comments that followed it, Carlos brought to my attention a study that I neglected to mention a recently published trial. This is the prematurely closed CALGB 9734 trial, (abstract here), which intended to enroll 480 patients with “occult” N2 nodal involvement, but it was closed due to poor accrual, with a mere 44 patients over two years, and two of them were actually ineligible. The trial was designed to test the value of adding radiation after chemotherapy for this setting, and all patients started with four cycles of carboplatin and paclitaxel given every three weeks for cycles. Following that, patients were randomized to be observed or receive daily radiation up to 50 Gray over five weeks, with 74% of those assigned to RT completing it (11% progressed before starting RT, and 15% refused it). The trial design is shown here:

(click to enlarge)

The results of this very small trial supported the potential value of radiation added to chemo, since the median failure-free survival (progression or death) was 16.8 months for the chemo alone arm and almost exactly twice that, at 33.7 months, for the chemo followed by radiation arm. Despite the large difference in those numbers, the difference was not statistically significant. Despite that, there were no differences in one year overall survival, which were 72% and 74% for chemo and chemo followed by RT, respectively. The authors couldn’t draw any real conclusions from the study.

How do we reconcile all of these results? How can there be a doubling of median failure-free survival and no improvement in overall survival at one year? First, the trial was very, very small, with only about 10% of the patients they intended to enroll on the trial. With fewer patients, even large differences are likely to be attributable to chance alone. It’s very possible to flip a coin five times in a row and get heads four of those times (result: 80% vs. 20%) just by randomness (probability more than 5%), but it’s remarkably unlikely to flip a coin 20 times and have it come up heads 16 times (result again: 80% vs. 20%), and in this case you should suspect that there is some real difference between the heads and tails sides of that coin, with a probability less than 5% for seeing this, so it would be statistically significant. Sometimes in oncology we have trials that show big differences but are too small to statistically significant, and they’re generally ignored, unless you put a bunch of similar small trials together for a meta-analysis (I’ve discussed several of these before in the lung cancer literature, and some have driven our practice). A median result, which is the point where half of the results are above and half are below that number, is a particularly inaccurate measurement in small samples. So it’s hard to read much into that big difference in median failure-free survival.

At the same time, that one-year survival being almost the same between the two arms may underestimate a real difference, since patients tend to do worse early and better late with more aggressive approaches (it’s a short-term investment in hopes of doing better long-term). For instance, no patient with leukemia undergoes a bone marrow transplant so they can do better than they would otherwise in the next six months. It’s a major up-front toxicity risk to get a potential payoff of better longer-term survival. The same principal applies, but not as extreme a situation, when using multimodality therapy to treat locally advanced NSCLC. Sure enough, although the one-year survival was essentially identical between the two arms, 53% of the patients receiving RT after chemo remain without evidence of recurrence, compared with 33% of the patients receiving chemo alone. As with failure-free survival, this large absolute difference was not statistically significant because of the small number of patients in the trial arms. But it helps us get a better sense of where reality may lie.

Taking the results of this trial along with the rest of the limited results we have for post-operative treatment of stage III N2 NSCLC, my conclusion is that the best evidence does favor sequential treatment with chemo and then radiation, although nobody can say anything definitive based on what information we’ve got. The problems with this CALGB trial suggest that SWOG may have a hard time completing its planned trial in this treatment setting, although its target is only 120 patients, and there are more institutions participating as part of the SWOG network than with CALGB. But for now, we’re going to have to rely on good judgment and good hints for years before we ever see results of a study that can really tell us exactly what to recommend for patients with stage IIIA N2 NSCLC who undergo surgery without pre-operative chemo and/or radiation.