A Tale of Two Tails

The tale of immuno-oncology hasn't been of the fairy variety, you know, with the happy ending. The number of approved IC therapies is less than earlier predicted, and has led to a bit of silliness from certain quarters. Nano-cap drug companies attempting some better mouse trap (so they claim, of course) IC have not gotten much done. Realizing that clinical trial results have been non-stellar, sponsors have turned to a new excuse for trial failure: IC only works over longer (years) periods than previously used for older agents (and earlier IC trials), aka chemotherapy. This argument goes by the name 'long tail'.

The notion behind long tail is that IC takes much longer to be therapeutic than old fashioned chemo. Most, if not all, of this argument lies on the foundation of trials not yet reaching event termination; the control (SoC or placebo) can't be the reason deaths haven't yet occurred, so it must be that the IC under trial must be extending survival. Not explained is how it could be that these patients are making it past the control's survival (and/or previous trials) in the first place? You don't avoid dying from cancer in order to get to the presumed long tail if the therapy hadn't been working during the early part of the trial. Now would you? It's not like a precocious 8 year old skipping third grade, now is it?

Here's the larger problem. Cancer survival (and most clinical trials) is measured by median survival. The reason is straightforward: with any disease, some survive longer whether on SoC, placebo, or no intervention at all. Survival distribution is known to be skew right; replete with outliers, no matter what happens. Data folk always turn to some other measure than mean in the face of outliers, and most often median is where the turn goes.

One doesn't even need fancy graphs to see why long tail may be utterly bogus. Consider these two distributions of survival:

A: 1 2 3 4 10 11 12 13 14
B: 1 2 3 4 10 110 120 130 140

The answer is that both measure 10 for median survival. The long tail zealots will claim that B is superior, but it isn't. It hasn't reduced the number of early events, nor increased the number of late events. I don't expect that any real trial would be this wacky, but only to show how median rejects the effect of outliers, per se. Without a rational MoA that explicates simple extension past the base median without changing the counts, such a result has to be accorded to random chance, because it is. A long tail effect, if one chooses to continue that name, can exist if the IC therapy changes the distribution to be less right skew by reducing the number of events before the base median and increasing the number of events after the base median.

Since the P&D folks have been extrapolating from blinded data, good luck. If PII trials have been stellar at 12 or 18 or 24 months, there's some reason to believe that PIII at 36 or 48 would at least squeak by at stat sig; although history shows that PIII trials (nearly?) always come in worse than PII, across the board. But, if as history shows, PII trials (and/or previous PIII) of these compounds have failed, then where are the patients coming from to fill out that long tail? Again, to ride that long tail, patients have to get past the base median, and not be matched by another fatality before the median. Arithmetic has rules.