Sometimes even when things go wrong, they can still work out in your favor. When Exelixis (Nasdaq: EXEL) announced a delay in presenting the results of its phase 3 trial testing cabozantinib in patients with advanced medullary thyroid cancer, I proposed three possible reasons for the delay:

• Patients taking cabozantinib are living longer than expected.
• Patients taking placebo are living longer than expected.
• Both.

Turns out none of them were the case -- at least not when looking at the median progression free survival, a measurement of how long it takes for the tumor to start growing again.

Exelixis was looking for an eight-month median progression free survival for the placebo arm and hoping for at least a 75% increase for patients taking cabozantinib, so at least 14 months. The results announced today pegged the median progression free survival at just 4.0 months for the placebo and 11.2 months for patients taking cabozantinib. Nearly tripling placebo's median progression free survival is a solid win for cabozantinib -- there's less than a 0.01% chance that it happened randomly -- but the results were still nowhere near what Exelixis expected.

The estimate being off shouldn't be a big surprise; the median survival estimate was just a guess. Exelixis tightened the enrollment criteria in the phase 3 trial to include only patients with progressing disease, so the earlier phase 1/2 trial, which included less severe patients, wasn't much help in estimating how long it would take for the more-advanced patients to progress.

Medians vs. averages
But then why did the trial take longer than expected if the median progression-free survival was actually shorter than the estimate? We'll have to head back to third-grade math to explain this one.

The median of a group of numbers is the one in the middle. Unlike an average -- where you add up all the numbers and divide by the number of members in the group -- the numbers at the extreme don't matter for a median. So the median of 11,12,13,14,15 is 13, but the median of 11,12,13,14,30 is also 13.

Exelixis only released top-line data, but the simple explanation is that there are some patients in one or both of the groups that are outliers, taking a lot longer than the median to progress. Since the trial could only be stopped after a certain number of progression free survival events, the outliers delayed the readout even though they didn't affect the median.

Houston, do we have a problem?
If the outliers are predominately in the arm that got cabozantinib, there's likely a subset of patients that are responding well to cabozantinib, perhaps because their tumors have a genetic mutation that makes the cells more susceptible to killing by cabozantinib.

If you go back and look at the trial data for Bristol-Myers Squibb (NYSE: BMY) and Eli Lilly's (NYSE: LLY) Erbitux and Amgen's (Nasdaq: AMGN) Vectibix you'd find outliers on the other side -- those that didn't respond well to the treatment because they had a mutation in KRAS, which renders the drugs ineffective.

Having outliers in the cabozantinib arm is an interesting scientific puzzle that I'm sure Exelixis will follow up on, but it shouldn't affect the approval. Both Erbitux and Vectibix were approved, and then their labels were subsequently changed when the scientific discovery was made.

If the outliers are in the placebo group only, that could potentially spell trouble for an approval.

Placebo controlled trials are only accurate if the patients are evenly balanced between the two groups. Things like age and gender have to be controlled for in each trial, and then there are factors for each disease. Diabetes drug makers like MannKind (Nasdaq: MNKD) and Amylin Pharmaceuticals (Nasdaq: AMLN) for instance, have to match up the HbA1c levels to ensure there are similar HbA1c levels across the treatment groups.

The Food and Drug Administration could argue that having outliers only in the placebo group implies that the trial wasn't balanced. Of course the outliers landing in just the placebo group could have happened randomly as well, but then we're getting well beyond third-grade math.

Until we see the full data it's hard to handicap the risk of the FDA might balk, but I'd guess it's fairly low. First, there's a good chance that it can be explained by a response to cabozantinib, but even if the outliers were only in the placebo arm, it may not be the end of the world. There aren't any drugs approved to treat medullary thyroid cancer patients this late in progression -- AstraZeneca's (NYSE: AZN) vandetanib is approved for earlier use -- and the FDA tends to be more lenient with drugs for unmet medical needs.