How Sweet It Is

One of the more interesting escapades in the bio-stuff sector over the last couple of years has been the effort by MannKind to get its version of inhaled insulin, named Afrezza, approved (no, I don't own any or short any). Before yesterday, that'd gotten them two CRLs (complete response letter, aka rejection) from FDA over the course of a few years. So, they tried a third time. The company is backed by one Al Mann (yes, he bought a company and changed its name to include his), who invented an insulin pump. He and his researchers devised a nano-particle (or nearly so, at least) carrier for a particular type of insulin, and an inhaler for delivery.

Seemed like a good idea, but there were various stumbling blocks along the way to yesterday (insert Beatles' tune snippet).

The two previous submissions were direct: MannKind sent the data and FDA sent back the rejections. This third try was going along similarly, until a few months ago when FDA said it would convene an Advisory Committee (aka, adcom) panel to review the submission prior to the decision date (called PDUFA, which you can Wiki if you're interested; not germane to this tale). Adcoms are, supposedly, panels made of outside experts to review the submission as filtered into a Review Document which the FDA distills from the submission and its staff review. Companies don't get to choose any of the panelists, and each adcom can have a different number of members with differing areas of expertise. Having a drug required to go before an adcom is generally considered an additional risk to approval, as opposed to a coronation. The review documents are (nearly) universally snarky. Managements get Tight Sphincter Syndrome as a result.

So, yesterday rolls around, and the panel takes seriously the review doc point of view that may be Afrezza doesn't work quite as well as the company asserts. This issue comes down to a quant question: is an average from the control better than the average from Afrezza? The measure is A1c (again, Wiki if need be) for Type 1s. The panel was viewable for $200, but I didn't have a ox in the ring, so I did not pass GO and pay $200. But Adam Feuerstein, as he often does, devoted his blog to commentary on the proceeding in real time (the replay is still up, if you're interested).

In the end, from Feuerstein's feed, it didn't seem that MannKind and FDA ever addressed the core issue with A1c and Type 1s. Which I thought was rather odd.

One of the benefits of Afrezza (again, according to the company) is that both the type of insulin used and its method of delivery yield fewer and less severe hypoglycemic events. In other words, blood glucose levels will have smaller excursion when compared to either a placebo or other drugs (insulins or not). The data which caused the kerfuffle was that Afrezza had a slightly higher A1c than control. It's easy to see, absent specific data of course, that with fewer outliers, particularly hypo crashes, A1c for Afrezza would calculate higher than controls which do evidence crashes.

Yet, from Feuerstein's feed, neither MannKind, the panel, nor FDA presented any data to confirm or refute the obvious arithmetic. Very odd. In the end, the panel accepted the *idea* that Afrezza produced fewer hypos and said yes with only 1 NO vote. Let the partying begin.

Quants talk about Type I and Type II errors (again, run to the Wiki). But there is another, much less discussed, variety sometimes called Type III error. This is the error of asking the wrong question. Nobody, apparently, asked the right question: "show my the numbers". Now, quantifying hypos in field trials is not trivial, but still...

One of the side effects of the adcom approval (not to be confused with FDA approval, which awaits) is a flurry of postings on message boards and blogs of all kinds. Including this one. One of the notions that got mentioned was whether, or not, the apocryphal Apple iWatch could include glucose monitoring. When I read that, my immediate reaction: "wrong question, dude". The issue isn't whether Apple can build some silicon and software to display a glucose level; it certainly can. Most anyone could.

And, it turns out, it's been tried for more than a decade.

The issue is: how to measure glucose, accurately, externally with a constantly moving device on a constantly moving body part. Turns out, not so easy. ARM and Intel won't be of much help. Boston Scientific or Novo or Pfizer, could be. There have been health metric measuring watch-like devices for some years. That's not new. What hardware Apple might control which would be useful and moated in the watch sector? No idea. But the iPhone certainly proved that having exclusive access to the signal bit of hardware makes a big difference.