Article 5: George Hripcsak
By Rita Charon
"George’s move toward the theoretical enables him to see beyond either data science or human physiology toward a synthesis of the two as necessary in studying not only the charting of human disease but the nature of disease itself."
- Rita Charon
Narratives of Discovery
All non-referenced quotes are the words of Dr. Hripcsak
Imagine if we could gather and analyze all the data being stored worldwide in electronic health records as a universe of evidence for bioscientific query. And imagine if such data—while protecting the privacy of patients and clinicians—could be followed temporally backwards and forwards to reveal the antecedents of illness, their emergent phenomena and disease course, and the outcomes of treatment? Such a resource would generate now-elusive knowledge enabling biomedicine to classify, predict, understand, and intervene in disease.
A visionary pioneer in the increasingly pivotal field of biomedical informatics, George Hripcsak has not only been foretelling all this but making it happen, as if a chimera of Isaiah and Steve Jobs. Trained in internal medicine, philosophy, chemistry, and computational sciences, he integrates the empirical with the analytical and the physical with the metaphysical.
George is the rare lucky academic who early found out exactly what he is built for. He joined Columbia’s newly-inaugurated Department of Biomedical Informatics as an internal medicine resident in 1988 and never left. He became the department’s third chair in 2007 as the Vivian Beaumont Allen Professor of Columbia University’s Department of Biomedical Informatics and Director of Medical Informatics Services for NewYork-Presbyterian Hospital/Columbia Campus. A member of the National Academy of Medicine and a leader in HHS’s Office of the National Coordinator of Health Information Technology, Dr. Hripcsak has had profound influence on the development of medical informatics globally through his discoveries, scholarship, and leadership of large-scale international collaborations.
George was building computers in his dorm room in 1977, starting with assembly kits and ordering additional parts from electronic catalogues pre-internet. His first computer—which he proudly showed off to me on my first zoom into his office—provided 256 bytes of memory, which seemed to him at the time infinite.
I was interested in computers and mathematics and in astronomy at the time. In college I took the one computer course they had, but I took 12 philosophy classes. . . Kant is very analytic. Understanding Kant is kind of like programming a computer; it takes some very analytical [training] to understand what he's talking about and try to get it. . . As you get to Hegel, then it's all ambiguity and paradox.
By either luck or design, George equipped his mind to handle both the analytic and the ambiguous, pairing the perfection required in programming a computer with the tolerance for singularity and uncertainty required to comprehend illness. He early became a major architect of Columbia’s electronic health record (EHR), both its structure and its clinical use. He reports that...
in recent years, I've had the freedom to actually become a little more theoretical. . . [Since] 2008, physicist David Albers [and I] worked together on using physics in healthcare, mathematical physics. . . We've been working on the glucose insulin system. . . It's very hard to control people's glucose in the ICU. And we're wondering if it isn't the case that because of various factors in that system, that it doesn't become a mathematical chaotic system and that trying harder won't work. If it's a chaotic system, you can't predict it. . . Learning that it's a chaotic system would be a good thing to prove.
When they published their paper in Chaos naming glucose control in the ICU an instance of delay-induced uncertainty, which “causes oscillatory dynamical systems to exhibit sustained temporal chaos,” biomedicine was opened up to the vast intellectual territories of non-linear time series analyses, complexity science, and complex network approaches into the analyses of ordinary health care processes and outcomes.
It is perhaps not an accident that George is a devoted fan of the music of Philip Glass. “Sustained temporal chaos” is what one meets on first listening to Glass’s compositions, called “minimalist” or, as Glass prefers to put it, “music with repetitive structures.” Glass’s opera Satyagraha narrates Gandhi’s non-violent movement for India’s independence. The opera is staged in three acts corresponding to Gandhi’s philosophy: personal conviction, working as a community, and the need for action. George appreciated the opera’s dense orchestral and vocal music and recognized how the repeated tonal strands circle around one another, slowly evolving to expose hidden complex patterns. The chaos gives way to order and meaning.
George and David Albers have been studying the “physics of the medical record.” An astonishing object that chronicles not disease events in a patient but the acts of the health care process that document those events, the chart represents a mystifying reversal in which hidden events occur prior to the opening chapter of the chart and come into only retrospectively speculative view. The patient enters the hospital with non-observed action underway. Once the chart authors start to write things down, they are in the “future” of the phenomenon, now picking up traces of events long underway, intervening before they fully understand the unfolding disease process. It is as if the authors of the chart are consigned to treating the ghost of the patient from hours or days past. “A disease appears to follow its effect because it is the process of diagnosis, not the disease itself, that is measured.”
George’s move toward the theoretical enables him to see beyond either data science or human physiology toward a synthesis of the two as necessary in studying not only the charting of human disease but the nature of disease itself. Maybe he has launched precision informatics:
I think we're going to want to do deeper biological understanding of things and not just be purely empirical. . . Some of [our] work uses mechanistic models, using 36 ordinary differential equations: Here's what the kidney does. . . Here's what the pancreatic beta cells do. The [work is] actually embodied. . . in physiology. And then we align that with the empirical information we have about the patient. . . to come up with a patient-specific physiologic model. . . Then we're able to do good prediction.
Forays into the physics and the metaphysics of the medical chart have seasoned and deepened this informaticist. In recent university presentations, George describes data science’s deconvolution of care that removes complication or artifact to better observe the thing itself. As noted in Physics Reports in 2019, “Advanced interdisciplinary data analytics techniques help to capture the hidden structures amidst otherwise chaotic data points, including approaches from machine learning, data mining, statistics, natural language and text processing.” George advises the young investigators in the audience:
Study the EHR as if it were a natural object. Use the EHR to learn about EHR. Borrow methods from non-linear time series. It’s how you cast the problem. Asking the right questions and answering them in a simple way will let you invent something really neat as opposed to “Here’s something 50 people tried already and I’m trying the newest algorithm on it.” Pair up with someone who understands the data. The advanced researcher looking for what to do next [should] not necessarily [be] doing data assimilation but pulling together mechanistic models, physiological models, with the empirical studies. That will produce the real results that matter. 
George’s bilingual fluency in computer science and metaphysics reveals cross-overs between the biosciences and the human sciences. Deconvolution, for example, translates into deconstruction for the humanities scholar. He continues:
In the same way that a CT scan is a 3D image created by taking 2D images and deconvolving them, can we do that by looking at the record in different ways, kind of metaphorically, and figure it out? 
George is right: literary scholars have identified metaphor and other literary tropes and genres like allegory, parable, synecdoche, temporal dysjunction, and stream of consciousness in medical charts, both hand-written and electronic. Identifying these non-medical uses of languages in the chart expands what we can learn about both the nature of illness and the practice of medicine itself.
In addition to being aesthetically-minded, George is psychologically-minded. He is able to recognize and mediate conflict within a group. I’ve seen him cede last-author position to his post-doc and refuse to accept credit when it is not his alone. When I asked him how he learned his collaborativeness and deep respect and generosity toward others as department chair, he said:
My family has taught me everything about relationships. I am a much better chair because of my family. With respect to work, my family is supportive when all seems lost and also illuminating when such illumination is needed.
Later in our conversation, he recalled other roots of his leadership style:
My mentor in college, my philosophy mentor, was an American but he studied a lot in China [about] Tao philosophy. My close friend gave me the book, The Tao of Leadership, when I became chair. . . That was actually. . . a fit.
A retelling of Lao Tzu’s 4th century BC Tao Te Ching, The Tao of Leadership encourages a leadership of service, not selfishness, a responsive leadership that does not take sides but is fluid and responsive. Lao Tzu asks, “Can you become open and receptive, quiet and without desires or the need to do something? Imagine there is a pond in this valley. When no fears or desires stir the surface of the pond, the water forms a perfect mirror. In this mirror, you can see creation.”
There are many things to be seen in the mirror of the pond for George.
* * *
In 2014, George with colleagues Patrick Ryan, David Madigan, and others started the OHDSI (Observational Health Data Science & Informatics, pronounced “Odyssey”) project as an evolution of an earlier public-private partnership called the Observational Medical Outcomes Partnership. Both projects were founded to learn about the capacity of observational healthcare databases to demonstrate outcomes of medical interventions.
We have 74 countries and 2000 researchers and 810 million unique patients in our federated database. It's more than 10% of the world's populations in the database now. . . . We know that we're influencing the European Medicines Agency. . . The FDA, I think we've influenced, but it's hard [to tell].
OHDSI was created to use this massive global data to attempt to achieve causal inference reliably on the basis of observational data. They are developing capacity for both population-level estimation and patient-level prediction. The project is indeed built on the scale of the twenty-year perilous journey of Odysseus back to his wife Penelope after the Trojan War.
We're getting to the point now where we trust our evidence. If you look at the LEGEND paper, the 10 principles. . . are kind of like our 10 commandments of how to do reliable research. [We are] turning it into two commandments, which are verifiability and openness. . . And that is how you produce reliable evidence.
Taken together, the principles aim to prevent publication bias and minimize P hacking, or data dredging. They minimize the impact of biases associated with observational studies, enhance the transparency and generalizability of pooled results, and address data security and patient and site privacy. When George opened OHDSI’s 2021 annual conference with his “State of the Community” plenary, he told the audience of 1200 about Satyagraha, suggesting that the truth-force of Glass’s opera is reflected in OHDSI’s work. Like Gandhi’s mission, he proposed, OHDSI seeks truth through collaboration, encouragement, tolerance, and generosity. Both the opera and OHDSI are built on the triple foundation of personal commitment, working as a community, and taking action.
* * *
For close to a year, George has been constructing a coronagraph, which is a telescope fitted to filter out all but the sun’s corona in daylight. George and his wife drive three hours upstate on a clear day in the late fall for its first trial. George sets up the coronagraph and attaches his camera. Here is our conversation as he tells me how it went and shares with me the first images from his coronagraph:
--I just set it up. I looked in, I’m like, there it is. It was like, there was no straining or anything. Like we got it. Here, so the image on the left in here is what I could see. The image on the right is from a satellite.
--Oh, my God.
--So this shows, this is the corona. This thing here was this thing here (pointing in turn to the image on the left and the right). So far as I know, I am the second amateur astronomer to see the corona.
--It’s a deep longitudinal commitment to do something. It is not doggedness. It’s the vision of it. It’s stunning, George.
--Well, it’s been fun.
--It’s like breaking the sound barrier.
--Well, that’s how it felt to us. Whether it is or not. That’s how it felt to us.
--It’ll never end, George. The work goes on at the intellectual, cognitive, analytic level. The work goes on at the interpersonal level, which is the most delicately demanding and unselfish part. And then there’s the philosophical level where you are contributing to the Hegelian frame of mind as you swing back and forth between chaos and the knowable. Something in you allows you to do these Odyssean efforts toward a kind of certainty that might not be accessible elsewhere. And, you know, Odysseus finally came home.
 Karamched B, Hripcsak G, Albers D, Ott W. Delay-induced uncertainty for a paradigmatic glucose-insulin model. Chaos: An Interdisciplinary Journal of Nonlinear Science. 2021(31):023142 [p. 023142-1] doi: 10.1063/5.0027682
 Hripcsak G, Albers D. Next-generation phenotyping of electronic health records. J Am Med Inform Assoc. 2013;20:117–121. [p. 120] doi:10.1136/amiajnl-2012-001145
 Hripcsak G, Albers DJ, Perotte A. Exploiting time in electronic health record correlations. J Am Med Inform Assoc. 2011;18:i109ei115. [p. i111-2]. doi:10.1136/amiajnl-2011-000463
 Zou Y, Donner RV, Marwan N, Donges JF, Kurths J. Complex network approaches to nonlinear time series analysis. Physics Reports. 2019;787:1-97 [p.3].
 Charon R. Medicine, the novel, and the passage of time. Ann Intern Med 2000;132:63-8.
 Heider J. The Tao of Leadership. Palm Beach, FL: Green Dragon Books: 11.
 Hripcsak G, Duke JD, Shah NH, Reich CG, Huser V, Schuemie MJ, Suchard MA, Park RW, Ian Chi Kei Wong ICK, Rijnbeek PR, van der Lei J, Pratt N, Norén GN, Li Y-C, Stang PE, Madigan D, Ryan PB. Observational Health Data Sciences and Informatics (OHDSI): opportunities for observational researchers. Stud Health Technol Inform. 2015;216:574-8.
 Schuemie MJ, Ryan PB, Pratt N, Chen RJ, You SC, Krumholz HM, Madigan D, Hripcsak, and Suchard MA. Principles of Large-scale Evidence Generation and Evaluation across a Network of Databases (LEGEND). JAMIA 2020;27(8):1331–1337. doi: 10.1093/jamia/ocaa103