In the early 1990s, two brothers, Gregory and David Chudnovsky, built a supercomputer in their Manhattan apartment.
In September 2015, Andrew Tyser, M.D. and his colleagues at the University of Utah published the results of their attempt to discern patterns of patient behavior using data mining techniques.
Increasingly, orthopedic care is being driven by data mining studies.
The story of the Chudnovsky brothers 25 years ago illustrates the pitfalls and promise of mining natural phenomena data.
The Chudnovsky Brothers
At the time the brothers built their supercomputer there were less than 10 true supercomputer models in the world. The Cray Y-MP C90, Thinking Machine’s CM-5, the Hitachi S-820/80, the nCube, the Fujitsu parallel machine, NEC SX-3 and Chudnovsky’s computer.
The Cray Y-MP C90 cost about $30 million. Gregory and David Chudnovsky’s computer cost about $70, 000. It was cheap because the brothers built it from mail order processors, routers and cables which they wired themselves. The “computer” extended through most every room in their Manhattan apartment. The apartment itself became the outside shell of Chudnovsky’s super computer—which they named m-zero—“m” for Machine; “zero” for success.
m-zero ran day and night. It burned about two thousand watts of power and raised the temperature of the small apartment to well over 100 degrees. Twenty-five fans blew constantly to keep the apartment livable. If it ever shut off, it would die. Seven monitors were windows into m zero’s guts.
m-zero was a parallel processing supercomputer with about 16 circuit boards chewing through the brother’s “inhumanly” complex algebraic calculations. Around the clock.
The Chudnovsky brothers are considered to be among the world’s greatest living mathematicians. Gregory Chudnovsky, for example, was the first winner of the MacArthur Genius award in 1981. The two brothers are currently the Distinguished Industry Professors at New York University’s Polytechnic School of Engineering. At the time of m zero they were with Columbia University.
The brothers used m zero for their most esoteric problems—problems which developed into systems of equations which had crashed other super computers.
Gregory, for example, had kidney problems. He went to the hospital for CAT scans. The pictures were, of course, rudimentary. Remember, this is the early 1990s. The brothers got ahold of the magnetic tape with the CAT scan data and took it back to their apartment. They put the data through m zero which gave them back a series of high definition cross-sectional, color images of Gregory’s body far more detailed than any CAT image. It took a few weeks. But the brothers relished the “interesting mathematics” the job required.
Indeed.
Pi
The great task to which the brothers applied m zero was to calculate Pi.
Pi is the ratio of a circle’s circumference to its diameter. Approximately 3.14 diameters equal 1 circumference—for any circle. It is a ubiquitous number. Since Pi describes a circle, it’s used in trigonometry, geometry, number theory, statistics, thermodynamics, mechanics and electromagnetism. The ancient Chinese and Indian mathematicians calculated Pi.
Pi is also an infinite number. It has no apparent end.
In 1674 Gottfried Wilhelm Leibniz, created a formula to calculate Pi which has been called one of the most beautiful mathematical equations ever. His series showed that Pi over four equals one minus a third plus a fifth minus a seventh plus a ninth—and so on. The numbers extend on into infinity.
But you never arrive at infinity. Paradoxically this infinite series is tethered to a basic shape in geometry—the Circle.
Pi is a fixed point. It does not move. It’s the math that wobbles around it. The Leibniz series, for example, begins at 2.66 then goes to 3.46, then 2.89, then 3.34. It circles around Pi getting ever closer until finally it converges on 3.14159265359….
So the Chudnovsky brothers decided to apply m zero to observe Pi. To push the calculation of Pi further than it had ever gone before. The brothers attacked Pi ferociously.
“We are looking for the appearance of some rules that will distinguish the digits of Pi from other numbers, ” Gregory explained at the time.
Mid-year 1991, the brothers had pushed Pi to two billion two hundred and sixty million three hundred and twenty-one thousand three hundred and thirty six digits past the decimal point. It was double the existing world’s record—which had been set by the Chudnovskys in 1989. The closest competitor was Yasumasa Kanda who was using a half-megawatt Hitachi supercomputer at Tokyo University.
At around the three-hundred-millionth decimal place of Pi, the digits went to 88888888—eight eights in a row. Later, ten sixes emerged: 6666666666. Somewhere past the half-billion mark the numbers count: 123456789.
But at the end of it all no pattern emerged. Pi proved to be infinitely long, non-repeating and absolutely, unalterably orderly.
Pi may well be the most perfect random data sequence that has ever been discovered. Pi is more unpredictable than anything known to man. Except Pi is not random. The fact that it can be calculated by a simple formula means that it is orderly.
In the world of mathematics, that makes Pi a transcendental number.
Biologic Data is Transcendental
The Chudnovsky brothers used raw computing power to seek patterns within Pi. Because Pi expresses transcendental natural phenomena, it combines the appearance of infinite randomness with a fundamental orderliness.
The same may be said of biologic processes.
Payers and hospitals are repeating Chudnovsky’s attempt to tease patterns from biologic phenomena. Certainly, 25 years after m zero, payers have plenty of computing capacity to throw at the problem.
Yet, where are the straight lines in the human anatomy? Squaring the circle—which is the essential task of Pi—is also what using data to describe biologic events is like. In terms of human anatomy and biologic processes Pi is everywhere. Everywhere are transcendental math problems.
Everywhere, in short, are the fingerprints of God.
Patient Age, Geographic Location of Treatment Impact Satisfaction With Outpatient Orthopedic Procedures
This paper, Patient age, geographic location of treatment impact satisfaction with outpatient orthopedic procedures, by Andrew Tyser, M.D. of the University of Utah appeared in the September 2015 issue of the Journal of Bone and Joint Surgery.
The author’s objective was to try to uncover patterns from 12, 177 academic, orthopedic, outpatient clinical encounters of 7, 258 patients who had completed the Press Ganey Medical Practice Survey between December 2010 and October 2013.
Seeking Patterns and Predictability
In fact, not only did the authors seek the underlying pattern, but they hoped it would be robust enough to have predictive qualities.
To try to meet that lofty goal, the study authors added additional data points to the Press Ganey numbers—such as each patient’s gender, age, status of employment, health insurance, zip code and the orthopedic subspecialty involved.
These were patients with a wide variety of underlying musculoskeletal disease processes including such chronic degenerative diseases as osteoarthritis, spinal stenosis, chronic wounds and degenerative disc disease.
Squaring the Biologic Circle
Determining how patient’s feel—physically or emotionally—is the predominate tool for studying the effectiveness of musculoskeletal treatment. Visual Analog Scale (VAS). Knee Injury and Osteoarthritis Outcome Score (KOOS). Western Ontario & McMaster Osteoarthritis Index (WOMAC). Press Ganey Medical Practice Survey.
These main tools are often supplemented with MRI images, CAT scans or X-rays.
But it still feels like we are, collectively, Gottfried Wilhelm Leibniz in 1647 where we make one calculation which puts us over the mark, make a second calculation which puts us under the mark and with each iterative step—whether by this research team or subsequent ones—wobble around the core biologic phenomena.
Our basic tool, the patient’s subjective assessment, as flawed yet honest as it is, has shown ability over time (and with lots of iterative steps) to eventually converge onto truth.
Age and Location Determines Treatment Outcomes
The University of Utah researchers took the answers from the Press Ganey Medical Practice Survey, scored them 1 to 5 on the Likert scale and then grouped the patients as either less satisfied or more satisfied.
Then they applied a predictive model of their own design to calculate the probability of less satisfaction and interquartile ranges for age, sex, travel distance and orthopedic subspecialty.
Here is what they learned.
Age was strongly correlated with patient satisfaction. Older patients were more satisfied (lower expectations?) while younger patients were less satisfied (the impatience of youth). The adjusted Likert scale ratio was 2.78 for patients aged 18 to 29 years compared with patients aged 80 years or older. Patient satisfaction also correlated with travel distance with patients living less than 50 miles from the clinic being less satisfied than patients living farther away. The adjusted ratio was 1.18 for distance of less than 50 miles vs. 50 miles or more.
And what did we learn?
To select older patients who travel further?
With such tools we wobble on, ever circling the patterns that make up the fingerprints of God.
