Myths of the Bitsphere and Biosphere
A brief reflection on the past year and some spurious predictions for hc/bio in 2023
I will be in San Francisco for JPM and hope we cross paths! If you’d like to meet, reach me directly at singareddynm@gmail.com.
2022 was one of those strange and frenetic years, marked as much by crisis and absurdity as it was by remarkable progress.
Nuclear energy companies made significant advances toward deploying small modular reactors. The Lawrence Livermore National Laboratory made history, performing the first fusion reaction that generated more energy than it used to start it. We marveled at the generative paintbrushes of Stable Diffusion and Midjourney. Developers deployed GPT and large language models to make music, write fanfiction, build websites, interpret medical records, and determine whether subordinates would follow orders in high-risk military scenarios. We seem to be in a golden age of biochemistry - just look at cellular reprogramming! Researchers successfully targeted previously undruggable genetic mutations and released tools like AlphaFold2 to predict protein structures in silico. Even a mixed year for crypto could not hinder the Ethereum merge.
But this good news commingles with a tremendous amount of noise. Unyielding social feeds. The synthetic high of announcement culture (“some personal news…”, another PR Businesswire, overpriced fundraising rounds, Forbes 30,000 under 30). Too much information - to the point where data fatigue is actively impeding science.
"What are the roots that clutch, what branches grow. Out of this stony rubbish?"1 Where are we supposed to look? Which shiny toy is worth paying attention to?
We’re encroaching upon the information scaling threshold. Civilizations rely on systems to record, organize, and understand events and transactions. These systems vary from belief systems to administrative methods and technology instruments. Shin et al. theorize that societies producing dizzying amounts of data stall and decline unless they “invent new ways… of coping with the complexity of the information environment.”2
If you know me, it is perhaps unsurprising that I take a keen interest in the myth as a system of information processing. Myths played an important role in ancient civilizations to order and explain an unknowable world from the Norse Yggdrasil, a sacred tree that unifies life, to antiquity’s four humors of human health. Myths prefer to work with incomplete, uncertain, complex milieus. In healthtech and biotech, myths take a didactic tone. They’re extrapolative hymns of progress — promises of unproven, complex systems and technologies.
Leo Marx calls this intoxicating rhetoric the technological sublime. New innovations need to be countervailed by myths that legitimize and spread their adoption. We can think of this installation period (borrowing from economist Carlota Perez) like a “turnstile” pushed by our chosen mythologists: the experts, insiders, innovators, investors, tastemakers, and the C-suite.3 Over a long enough time frame, mythologists compound their rhetoric with proof points… or their myths risk disintegration.
If you’ll indulge me in a review of healthcare and bio’s innovations:
Value-based care myth: divined by CMMI and the big consulting cavalry, VBC promised to reduce costs by shifting care delivery from volume to value. But 10 years, $20B+, and 54 models later… VBC has yielded mixed and discouraging results. Providers complain of complexity and taxing reporting requirements. We now question the association between care quality and overall VBC payments. And this is before we even consider widespread belief that VBC metrics don’t actually achieve outcomes.
An anonymous provider: “Outcomes as metrics cease to be good metrics once you mandate them because organizations will rapidly find ways to technically do well. Value-based care metrics are no different. By creating an unassailable outcome (financial penalties) for not "gaming the system" in an optimal manner, all sorts of perverse outcomes occur. Case in point, please see not finding blood infections and UTIs by not looking for them anymore.”
Champions of the computational biology myth tell us their models engineer the future. Putting aside these mythologists playing fast and loose with deep learning metaphors, they beckon with troves of impressive interaction and structure predictions while side-stepping the problem of high-quality training data in short supply. Remember, this is a field known for contaminated cell lines and fabricated imaging.
Sandeep Chakraborty of Nvidia: “It is not an isolated problem - the wrong application of computation to biology "big data" is rampant. Given a sample set, an algo will give an answer - which will have no bio significance.”
Richard Law of Exscentia: “A useful way to see AI is to understand that it can only do well what we can build really reliable, big data sets on that really encompass the problem, meaning likely humans are already good at it; e.g. driving & chemistry. With biology we are still fumbling in the dark!”
Outside of a few key examples like Omada and Virta, the digital health industrial complex has largely failed to demonstrate meaningful ROI. Adoption/utilization continues to be low (~6%), weighted towards prescription filling and therapy. Many prefer hybrid or in-person services. By primarily treating acute episodes, should we expect much more of digital health’s impact?
We still regard these innovations as the next big thing, the myths of the bitsphere and biosphere.4 But what happens when lofty promises of value-based care, of comp biology, of digital health repeatedly over-promise and underdeliver? They hollow out. It’s normal people - workers, patients, consumers, clinicians, admins / techs - who bear the cost in disillusion and wasted time/resources.
In an era of cognitive overload, unstable myths risk losing their downstream believers. “They don’t have patience for nebulous claims.”5
Linguist Roland Barthes is a rich resource on myths and their methods of proliferation and adaptation. In 1957, he wrote that impoverished myths demand reconciliation, ways to cohere imagination with reality. I foresee this reconciliation in healthtech and biotech as maps.
The World Atlas of Wine. The Nuclear War Atlas. A Guide to North Carolina Mineral Resource Development and Land Use. In just these titles, we can appreciate the map for doing two things: first, organizing data of past / present phenomena, and second, manifesting myth in what it values, represents, indexes, and ignores. It persuades and constructs a useful worldview. As cartographer Denis Wood muses in The Power of Maps, “the cartographic claim is to be a system of facts but it carries an exhausting burden of myth… it is not that the map is right or wrong but that it takes a stand [by] transmuting interests into being.”6
In the coming year, I hope those maps take the form of…
Financial operating systems that give clinicians and admins visibility into the financial health of a practice (starting with simple expenses and accounting). Imagine if practices could use past / present financial data to easily unlock working capital, realizing practice growth and easing into VBC.
Functionally-annotated molecule atlases. Commercially available molecules grew from 1B to 60B+ in the last 5 years.7 Researchers need new opinionated, tagged libraries to keep up. Some examples include Basecamp’s protein atlas, the Human Cell Atlas for gene expression, and Enveda’s small molecule search platform. In the energizing tone of Tess von Stekelenburg, “sequence the world!”
An institute for longitudinal studies in endocrinology. I agree with my colleagues who write the TechBio substack: we lack datasets and insights into most diseases and their interactions. I’m primarily interested in endocrine biomarkers given their impact and limited range of mechanisms and chemical classification. Think GWAS cataloging for hormones.
Universal atlas of clinical diagnoses. There are already thousands of diagnostic textbooks dedicated to each specialty but we now have GPT tools to train on our collective clinical knowledge. Instead of the gate-kept UpToDate or hogwash that is WebMD, clinicians and patients alike can prompt / search an atlas that topologizes pathways and explains its decision results. Clinical organizations could also use GPT for internal semantic search on EMRs. Wharton professor Ethan Mollick shared a relevant, promising study here.
Pricing transparency datasets hit the mainstream. Most people have no idea that they can search hospital prices. We can quibble about the specifics of GTM but I hope consumers find Turquoise Health or open-source provider Dolt to empower themselves with visit cost estimates. The next phase is enterprises leveraging these datasets for better payer-provider contracting and insurance network design.
Genetic counseling atlases. Expecting families began using genetic counselors like Counsyl a decade ago. As the cost of genomic sequencing decreases dramatically (free for NICU parents in the UK), consumers should be able to securely upload and interpret their own genetic data and family health history.
Lab operating systems that connect, harmonize, display, and enable native manipulation of instrument and software data. This is even more critical as talent in lab tech market tightens, companies outsource components of their research, and the FDA heightens data scrutiny. Some examples include Ganymede Bio and modular Genoma.
These health tech and biotech mapping technologies (linking one-to-one and one-to-many) hold an unveiling power for its users suspended between doubt and faith. “Suddenly the things represented… are opened to discussion and debate.”8 Maps are myths reconciled, making progress real in how they guide myth receivers to see, measure, and take action. Through them I hope our mythologists inch closer to materializing their grand promises of the bitsphere and biosphere.
—
Bonus section! Other things I’m thinking about in 2023:
The first year of the rural emergency hospital, a new site of care. PE firms snapped up a number of small rural hospitals at the end of 2022 that will likely re-open under improved margins and relaxed reporting requirements. REH rules are very permissive, even giving sites the ability to determine and grant medical privileges to staff. This is going to be a mixed bag in its results but temporarily stops the bloodletting of rural hospital closures and patient access issues.
In 2020, we discovered vivores (organisms that feed on viruses) and in 2021, we discovered borgs (archae that metabolize methane). I bet we discover another breakthrough organism in 2023 especially as microscopy imaging improves.
I mentioned last year that provider-payer arbitration was going to get crazy with the No Surprises Act. I was right, and it’ll probably get worse.
We’ll see the explosion of supplemental benefits. There are 5,000 insurance plans that can offer SBs to 30M+ eligible members. There aren’t many entities that have the scale to cover transportation, nutrition, OTC, healthy rewards, etc. but expect Optum, Walmart, and CVS to grow their offerings.
Andrew White, Head of AI at CRO provider Vial, graphed the last 60 years of drug candidates to show their eerily similar molecular structures. Does this imply our drug discovery methods have converged on an ideal structure type… is structure design overfitted?
This is the year everyone will learn about and get on GLP agonists. Just yesterday, the FDA approved Wegovy for 12-year-olds and older. We’ll also see concerning off-label GLP use by people with eating disorders (semaglutide / Ozempic / Mounjaro chatter has popped up on pro-anorexia and bulimia Instagram and Tik Tok).
Biomanufacturing will be the bottleneck companies rush to solve.
HCA, the nation’s biggest private health system, owns a number of nursing schools. It’s smart; the schools serve as a pipeline back into their hospitals and purportedly reduce HCA’s labor cost, one of their biggest line item. Smaller, regional systems will adopt the HCA model as clinical labor costs skyrocket post-Covid.
~10 Humira biosimilars hit the market this year! Not all of them require step therapy (a method of prior authorization for expensive prescriptions). This is great for patients suffering from inflammatory diseases, and an unlock for people in the gastroenterology space and those exploring off-label prescribing.
HuggingFace recruited its first engineer to work directly on biomedical and healthcare applications. The company is already a sponsor of the Big Science Initiative which open-sources 140 biomedical datasets. I consider Katie’s hiring as a harbinger of AI/ML companies plucking from industry over hc and bio companies landing engineering talent.
We’ll see the rise of preventative prescribing spun as a longevity strategy. I’ve already heard of clinicians prescribing sauna treatment (with a note of medical necessity, they’re often HSA eligible). Conventionally healthy people will take baby aspirin, statins, or the more experimental rapamycin.
📖 I’m reading Lionel Shriver’s We Need to Talk about Kevin and Nick Lane’s Transformer: The Deep Chemistry of Life and Death
👂I’m listening to Pablo’s Eye
👀 I’m watching reruns of Australia’s Curiosity Show (1972-1990)
I’ve waited years to cite one of my favorite poems T.S. Eliot’s The Waste Land.
Stumbled onto this paper through Gordon Brander’s wonderful Substack, Subconscious, and his writing on collective intelligence.
Barthes, Roland. Mythologies. Hill and Wang, 1972. I bought this book at a topsy-turvy bookstore in Berlin.
I fully expect someone to read this and go... Nikita, that’s a lot of words to say healthtech and biotech are overhyped. I don’t consider myths to be the same as hype even if they share common elements. Hype is exaggerated expression (of specific products, practices, events). Modern myths are collective cultural beliefs. Because their concepts are fuzzier, they tend to be longer-standing, stretched in many directions by mythologists with ever-changing signifiers (symbols, slogans, institutional goals). See: myths of defense tech / American Dynamism.
Evan Feinberg of Genesis Therapeutics via Vineeta Agarwala.
Wood, Denis. "Maps and Society." The Power of Maps. New York: Guilford Press, 1992. 108-116.
https://practicalcheminformatics.blogspot.com/2023/01/ai-in-drug-discovery-2022-highly.html
Wood, Denis. "Maps and Society." The Power of Maps. New York: Guilford Press, 1992. 22.
Opinions my own and do not reflect the views of any affiliate organizations. I ran these footnotes through Easybib.