This is theoretical and abstract.
The prior post here, regarding Scott Page's work on the power of diversity in his new book "The Difference", deals with predictions more than estimations. The difference is crucial.
What is being pooled is a set of models, not a set of data points where each of the points estimates some true value plus noise.
More precisely, each of the models is really a subdimensional sampling of a higher dimensional reality. Maybe the reality has 50 key factors, and each model pays attention to only 5 or so of those factors and does the best estimation it can using those.
So, in an abstract sense, the primitive elements being pooled are really reference frame choices.
That means that we are looking at a computing system, in the most abstract sense, which isn't processing "content" but is processing "context".
The equation presented by Page that I saw involves summing a set of terms, each of which takes current reality and views it through a different lens resulting in a "content" estimate.
I'm suggesting that this equation and physical reality could be refactored a different way, and seen instead as summing the lenses first, then using the aggregated and synthesized new hyper-lens to view reality. This "aperture synthesis" approach is probably mathematically equivalent, but touches a different part of our intuition and suggests different experiments.
In fact, the transition has some parallels to going from an array of light sensors, such as a common house-fly's compound eye, and changing to a model of a retina, which means there are light-sensors and also, critically, higher-vision centers that process data on an image level, not a pixel level.
This in many ways describes the sense of a peak-performance teamwork window, or peak Baha'i "consultation", where the people have in many real ways given up individual "ego" and become, in crucial ways, a larger system that operates as if a single larger being. It is as though the slime-mold individual cells have flowed together and, for the moment, formed a larger living being that can act as if it was a single living thing, but then when it reaches its destination dissolves again into individuals.
The above part of this post describes changing the idea of a "central [content] processing unit" (CPU) to a "central [context] processing state" (CCPS).
Something like this appears to me to be necessary to overcome, for example, the breakdown of health system hospitals and the clinical medical area into hundreds or thousands of tertiary specialties, each of which forms silos and a type of fractal-shaped world in which "data" about the patient is observed by these tertiary specialists.
The key problem with "The Electronic Health Record", which seeks to "consolidate" all these disparate system's data, is that most of the meaning of the data is context-sensitive.
This is mathematically equivalent to the world described in the Hilbert Space of cosmological General Relativity, where local warps and curvature of the underlying space-time metric distort not only the thing being observed, but also the observer.
Digression into general relativity:
In Relativity, it is not possible to simply take a "set of measurements" made in context A, and lift them off the map, and plop them down unaltered into context B, and then compare them to what an observer in context B measured and get anything that makes sense.
For example, if two observers are racing towards each other at 99.995% of the speed of light, each, if they can see the other, will observe the other's clock to be moving more slowly than their own. Initially, we say this is clearly a "paradox". How could both of these things be true? They seem "inconsistent". At this point most incorrect interpretations of the situation say "nothing can be measured, everything is relative" and go off the deep end.
What Einstein said is very different. He said that, if you take observation A, and slide it through space over to context B, and as you slide it, adjust it for the sliding process and the change in space-warp that is occurring under you, when it reaches context B, it will agree entirely with what a credible observer in context B observed. This process of "parallel transfer" of tensor data is crucial to reconciling observations made in different reference frames or differently warped contexts by perfectly credible, perfectly accurate observers who are themselves embedded in those reference frames.
In fact, Einstein went on to say that you could equivalently parallel transfer both observations to a "flat" reference frame, a perfect one with no distortion, where they would both have changed, but would now agree entirely with each other. He said, there is indeed a "proper" reference frame that could be used, if you want, in which to do data comparisons and aggregation.
Return to the discussion
The reason I raise general relativity is not to confuse things, but to point out that there are solid physical models and existing mathematical tools for dealing with context-sensitive data. They are not for the timid.
However, they provide in some way a guide to what is probably going to be needed to get a central "Electronic Health Record" (EHR) to work - namely, that the hard work be done to relate each of the sub-specialties to a "proper" or "flat" central reference frame, so that the data can be distorted and transformed properly as it is moved from the tertiary specialy subsystem to the central data repository.
Distorted? Transformed? (The audit trail people have run screaming from the room.)
Yes, distorted. When a clinical oncologist says that a person "has cancer" this does not mean the same thing at all as when a patient says he "has cancer." The words look identical, but the meaning is different. What is being said, and the whole worldful of implications of what is being said, is entirely different. You cannot just take the words that mean something within one context and remove them from context and plop them down somewhere else in a different context. That is an illegal operation, mathematically, that is more wrong than "distorting" or "correcting" the words for the context as best one can.
And, if you did, the tertiary specialists would never want to "read" the central EHR record summary of the patient, because they would either get the wrong idea of what was being said, or be totally confused. Their first question for every "observation" would have to be (and typically is) - "Who said that?" They don't mean just what person said that, with what level of expertise, but in what specialty, in what context, with what level of sophistication and discrimination were those exact words selected?
And, are those, in fact, the exact words, or did some coding clerk, generically trained, take the carefully selected words of a specialist and replace them with some sort of universal least-common-denominator phrasing?
No, we can see that the context of tertiary specialists IT systems in fact hold much, perhaps 99% or more, of the meaning of any given set of words, of what is being tacitly or implicitly said in shorthand by that exact choice of language.
So, they would actually never want to go to the Clnicical data central repository to "read all about" the patient. What they would want is that any and all relevant data be transformed the other direction, be brought into their tertiary context world, and stated precisely in their own shorthand. Then they can find it more useful than misleading.
As with Hilbert space or general relativity's "tensors", some data corresponds to rank-zero data, ie, scalars, and can in fact translate from one place to another by just lifting it up and moving it. Number of children is invariant. Temperature, list of drugs currently being given, schedule for next tuesday, home phone number, etc. These are not context-sensitive.
But much of the clinically significant readings are, in fact, extraordinarily context sensitive. Some mean absolutely nothing outside of the small circle of specialists.
I suspect that this problem, of a fractal metric underlying clincial data and practice "silos", is one that will be the shoals upon which many EHR ships will be lost.
technorati tags:EHR, healthrecord, silos, specialties, communication, fractal, metric, Hilbertspace, relativity, Einstein
No comments:
Post a Comment