Hello Mr. Peabody

Just call me Sherman, Mr. Peabody's faithful pet boy who accompanied him on all of his trips in the Wayback Machine. Let me explain. My break from writing (although it is a bit pretentious to say "break" when you have made only one blog post) was due to a pending job change (thankfully of the Robert Frost-type). In going through the process of packing books, an old pair of texts I had read years ago on Catastrophe Theory caught my eye, one by Arnol'd and one by Gilmore.

I had just finished Fooled by Randomness by Taleb and traveled a few pages into his second book, The Black Swan: The Impact of the Highly Improbable, so my brain was/is focused on rare events and started to draw a few tentative lines between the world of the black swan and the mechanics of catastrophe theory.

One of the concepts that Taleb stresses is his "triplet of opacity", which he defines as

1. an "illusion of understanding", where individuals believe that they comprehend the world more than they actually do
   
2. "retrospective distortion", or the proverbial adage "hindsight is always 20/20"
   
3. "overvaluation of factual information", or the curse of knowledge
   

Simply put no matter what your description or understanding of a system, it is incomplete and that serves as a potential breeding ground for black swans, aka catastrophes.

Starting with the work of mathematician Rene' Thom, catastrophe theory describes a geometric interplay between small changes in parameters in nonlinear systems that lead to sudden, large changes in the system. The figure to the right is a cusp catastrophe, and think of the green arrow as the path of a single, known parameter. Movement in the direction perpendicular to that parameter (toward the front of the figure) is due to an unknown parameter. As long as that unknown parameter is in the back of the figure, the system behaves as described. But once that parameter passes the start of the cusp, that smooth transition due to changes in the known parameter becomes a black swan, the system rapidly falls from the upper state to the lower state along the red path. Easy example...you are turning the volume knob up and down on a stereo for a time, enjoying the rising and falling of your favorite song. The knob setting is the known parameter, and the system state is the volume. The unknown parameter is a small child hiding behind the stereo, fighting temptation to pull out the plug. When he does, the system fails and to you (still not seeing the child) it looks random...a black swan...a catastrophe.

Now my mind is wondering about the structure of things, especially in biology. Does the start of cancer live in the cusp? Is the quiet accumulation of mutations that "unknown parameter" that silently slides toward the cusp? The human body is a complex system, that is only partially understood...

Feeling Frail

Talked to a friend today who is trying to get a biotech startup off the ground. The company has an interesting product that could provide quick genetic information on cancer. What is really missing now in the world of molecular diagnostics is a way to rapidly assess disease status. By rapid, I mean less than a day and ideally within the span of an office visit. There has always been a big push for diagnosis from a molecular standpoint because that mode of diagnosis as the potential to be the most rapid, sensitive, etc. and (hopefully) least invasive for the patient. You can see the initial stages of the later in the growing debate over merger of radiology and pathology on the Lab Soft News blog. What is missing now to a large extent is a platform (and possibly a mindset) to monitor the response of a cancer to therapy and give guidance for therapy selection. There are a few instances of this happening (CML and gleevec come to mind), but there needs to be more.

Anyway, back to 4/C exam prep. The topic of frailty models just cruised past. The ability to adjust survival based on covariates is a powerful modeling approach. It reminds me of sliding mode control theory that I learned as a engineer...adjust the dynamics to fit the current model. In actuarial math, I have already studied/learned how to correct for inflation and how to splice models together for simple cases, but the generation of really finely-grained models seems like the next logical step. Is what molecular diagnostics really has in store for the actuarial world? Chris Anderson's excellent book The Long Tail speaks to this and clearly shows the growth/exposure of finely-grained consumerism via the web and other modern technologies. I really wonder with all of the data that these technologies can generate, are the models (and modelers) there and ready for the integration? From a recent article in CAP Today, It looks like at least one molecular lab is talking to/working with insurance companies.

Should be fun...back to 4/C.

First Post

First note on the personal-yet-private notepad. Why am I doing this? It is a way for me to keep track of things as I merge/expand/etc. my medical knowledge with all things actuarial. Right now, I am studying for 4/C, with 1/P, 2/FM, and the stats VEE under my belt. Depdending on progress, I may add in one of the two parts of 3. Currently leaning to the CAS-side of the street (I know, it should be SOA for health), but it is just a lean. ERM is my main focus now. Time for more study questions on inflation and expected values from the ASM guide...