The Worst Road-Trip In America

Wednesday. January 15, 2025 - 24 mins

How Six Pages Built America, and Why Biomedical Research Needs Its Own Interstate

At precisely 11:15 on 7 July 1919, a military convoy of 79 vehicles with 24 expeditionary officers, 15 War Department staff, and 258 enlisted men left the White House lawns to embark on a journey across America. Their task was simple, to appraise the most developed highway in America for military use, the Lincoln Highway. The outcome was anything but. The convoy encountered 230 road incidents, delays necessitating extra encampments, they broke and rebuilt 88 wooden bridges, ultimately resulting in nine vehicles retiring, and 21 men not completing the trip due to injuries.

62 days later, on 6 September 1919, to much fanfare, the convoy finally arrived in San Francisco. The convoy logged 3,250 miles at an average speed of 5.67 mph. The roads from Illinois to Nevada were a patchwork of mud, gravel, and the road to hell was paved with good intentions. The answer back to Washington was swift. One Lt. Col wrote, and I quote, “It is clear that, for now, trans-continental road trips would remain the domain of publicity stunts and motor enthusiasts” [1].

It was, by all accounts, “the worst road trip in America”.

Lincoln Highway. From Barry Lawrence Ruderman Antique Maps Inc.

But then the years began to tick by. Administration after administration started noticing the number of motor vehicles increasing in the lives of everyday Americans. They recognised that road construction could not be left to the responsibility of states. Each state had its own standards. Sometimes the road was 20 feet wide. Sometimes 10. Sometimes it vanished entirely. Each state did build high-quality urban transportation networks; however, their interests were insular. Iowa, for example, would have no interest in building a highway to connect Illinois to Nebraska.

As Americans went off to fight in World War II in Europe, they noticed the efficiency of the German Autobahn network. But when they returned home, they were forced to navigate a patchwork of state roads of varying quality to get anywhere beyond their local area.

Sound familiar?

Today’s biomedical researchers face the same journey. Each lab, each hospital, each dataset speaks its own language, its own culture. As I write this from AAIC 2025 in Toronto [2], I overhear an Alzheimer’s researcher in Buffalo, New York, complaining that they are unable to easily share data with another researcher in Rochester, New York. (Spoiler: I told them about Synapse.) A real-world I-90 interstate journey of less than 80 miles seems nigh on impassable! Not because they don’t want to. But because the interoperability roads don’t exist yet.

The Road to Hell is Paved with Good Intentions

Pardon the minor digress for a moment. But do you want to see what good intentions look like in biomedical research?

Here’s a real data access request. For de-identified health data. In the UK.

The Data Access Marathon: 5 organisations, 40+ checkpoints, 2+ years, and counting. Each star represents another approval. Each red arrow, another delay. Source: Taylor JA et al.; BMJ Open 2021 [23].

Five organisations. Forty checkpoints. Two years minimum for a 4-year funded project!

And that’s for data that’s already been de-identified. While you have funded PhD students, post-docs, data scientists, and lab technicians twiddling their thumbs waiting for the data!

This isn’t incompetence. It’s competence, perfected.

Perfectly executed bureaucratic competence.

Every committee believes they’re protecting patient privacy. Every review board thinks it’s ensuring ethical research. Every delay feels justified.

They’re all right. And yet, they’re all wrong.

Because while they’re protecting data from hypothetical misuse, real patients are dying from actual diseases.

The cure delayed is the cure denied [24].

Look at that timeline again. See those overlapping bars? That’s four organisations reviewing the same ethics application. Simultaneously. Like four mechanics checking if your tyres have air, while your engine’s on fire.

The road to hell isn’t paved with bad intentions. It’s paved with good ones. Layer after layer of well-meaning governance, each adding just a little more friction. Until the friction stops all movement.

I know what some of the data protection/privacy terrorists will say, believe me, I suffered my fair share of them. “Susheel, these are the right checks and balances we need. It makes sure that patient privacy is paramount. If you remove them, it’s the start of a slippery slope!”

Ah, the privacy terrorists. They’re not wrong. They’re just answering the wrong question.

Privacy and confidentiality matters. Of course it does. But here’s what they won’t tell you: The current system doesn’t protect privacy. It protects the system.

That image from BMJ Open? Two years. Seven hundred thirty days. That’s how long it takes to get approval to even take a sneak peek at data that might save lives. While we’re checking boxes, people are dying. The slippery slope isn’t removing safeguards. The slippery slope is when bureaucracy becomes more sacred than outcomes.

You know what violates patient privacy? When their cancer treatment fails because the research that could have saved them is stuck with “Reviewer number 3” in committee meeting number 37.

Real privacy protection is proactive, smart, automated, dynamic, and auditable. Like credit rating and financial fraud detection, instant, invisible, and effective. Not death by a thousand paper cuts.

The privacy terrorists defend the moat while the castle burns. We don’t need fewer protections. We need better ones. Ones that protect patients, not processes.

That’s not a slippery slope. That’s a cliff we’ve already fallen off.

Two years to access data. Five minutes for cancer to metastasise.

We built these processes to protect patients. Instead, we built a system that protects itself.

Good intentions aren’t enough. Roads need to lead somewhere.

Back to 1919

Which brings us back to our muddy roads, the broken bridges, and that military convoy. Because sometimes, it takes someone who’s been stuck in the mud and muck to understand the need for highways.

Getting back to our story: One of the 297 members of the original 1919 Transcontinental Motor Convoy went on to pursue a successful career in the military and eventually became the Supreme Commander of the Allied forces in Europe during World War II. In fact, it was the very same Lt. Col who wrote the swift response back to Washington. After he returned from the war, he pursued a career in politics, culminating in his inauguration as President of the United States of America on 20 Jan 1953.

This was, of course, General Dwight D Eisenhower. Now the 34th President of the United States.

Gen. Dwight D Eisenhower.

With the groundwork laid with decades of planning and false starts, Eisenhower’s task was to corral the political will and funding to turn this concept map from 1947 into reality. After three years of political manoeuvring, he succeeded.

On June 29 1956, The Federal-AID Highway Act of 1956 was passed, dedicating $25B to construct 41,000 miles (or 66,000 kms) of highway.

Whether it is I40 in Tennessee, or I90 in Idaho. The way these roads were built was highly, highly standardised. And that was, after all, the point! Importantly, these standards were codified in a simple six-page document with a genuinely creative and enticing title called “A Policy on Design Standards Interstate System” [3].

The Interstate Highway System changed everything. Not through technology, asphalt is ancient! Through standards. 12-foot lanes. Everywhere. Same signs. Same exits. Same everything. The system was incredibly simple, and it must be, and again, that is the point. As Dan McNichol wrote in “The Roads That Built America”, “The interstate system’s standards create uniformity. Uniformity seems banal, but it’s that uniformity that keeps you alive. There’s no wondering, ‘Is the asphalt going to change to concrete or brick? Is the road going to suddenly get narrow?’” [4].

To get disparate and often conflicting groups moving forward in a shared direction of travel (pun intended). You need to make things simple, not complicated. Making things simple is actually anything but a simple endeavour.

A summary of the Interstate Highway Standards, aside from a few minor details. That’s it, all six pages.

Eisenhower’s 62-day road trip in 1919, today takes ~42 hours

Boring? Maybe.

Revolutionary? Absolutely!

National system of interstate and defense highways. Published by the American Automobile Association, June 1958. Courtesy of the Library of Congress.

McDonald’s went from 34 restaurants to 1,000 in twelve years. Not because their burgers got better. Because their supply chain trucks could deliver anywhere, predictably [5]. In such a decentralised country with a patchwork of states, each with its own governments, laws, cultures, and more. The consistency of a nationwide transportation system is unique. While not perfect by any measure, these interstate highways are more than what they seem; they are genuinely an interoperability system for America.

While you might think, “What has the interstate done for me?”, take it away and you will realise you are suddenly disconnected from society and unable to participate in productive economic activity, truly.

They are also incredibly safe compared to non-interstate highways and roads, accounting for half as many deaths per 100 miles travelled. That accounts for saving roughly 6.5k people per year, and using the Dept. of Transport’s own actuarial tables, that values a human life to roughly $11.7M per person, those six pages represent a moral and ethical impact of saving 6.5k people per year and consequently the economic impact of approximately $75B per year just by saving these lives. If you add in the trade, commute to work, study, leisure, etc., some studies estimate the annual impact of the interstate highway system to be $752B.

Not bad for just six pages! That’s $125B per page, per year!

Here’s the thing about standards: They feel like constraints, until they set you free.

The Silo Paradox: Research at the Speed of Fax Machines

But unlike the highways that now crisscross America, our digital research infrastructure remains a collection of dead ends and detours. It’s a puzzling fact that we can sequence an entire human genome in hours, but we still can’t get two research labs within the same University, let alone hospital systems, to talk to each other. Even with fax machines! But that’s a story for another time.

From XKCD:3105, Interoperability.

We’ve built AI models that can diagnose rare diseases from retinal scans; yet, 70% of researchers can’t reproduce their colleagues’ experiments because the data is locked in digital Fort Knoxes [6,7]. We spend $28B annually on research that can’t be replicated [8,9], not because the science is wrong, but because the data remains inaccessible [10,11].

This is the silo paradox of modern biomedical research. We’re drowning in oceans of data while dying of thirst for accessible knowledge. Every research institution is building its own Tower of Babel, like “warring nations” speaking its own digital dialect, protecting its culture and intellectual property cathedrals, while the real enemy, disease, is already at the gates laughing at our so-called borders [10].

The Permission Paradox

Researchers hoard data NOT because they’re selfish, but because the system rewards scarcity. First to publish wins! Share too early, lose your edge! [12] It’s academic capitalism at its finest, where data becomes currency and collaboration becomes a form of competition. We’ve created a permission paradox, where nobody, in practice, has any incentive nor permission to share, even though it is federally mandated [16].

Each research domain has its own tribe; Genomics speaks FASTQ, Clinical Trials speak CDISC, drug discovery speaks SMILES. They’re all describing the same human biology, but we might as well be speaking Klingon to each other. These aren’t just technical standards; they’re tribal and cultural identities. Changing your data format/model feels like betraying your scientific heritage. “We are the tribe, while all other tribes are the others”.

The Purple Cow Dilemma

Everyone wants their research to be remarkable, to stand out in the crowd. So they create bespoke databases, proprietary formats, and unique identifiers. And if you don’t use their language/dialect, they bash you over the head with it until you either succumb or, worse, are forced to create a more elaborate language to stand out in the crowd. They paint their data a different shade of purple. But when everyone’s data is purple, nobody can see the bigger picture. The very uniqueness that makes individual research remarkable makes collective progress impossible. Seth Godin refers to this as the “purple cow dilemma” [13].

From Emma Brooks Design.

Right now, we’re in that difficult period where the old ways aren’t working, but the new ways aren’t built yet. It’s easier to stick with spreadsheets and CSVs than learn new systems. It’s safer to keep data siloed than risk being scooped [12,14]. But those who push through the dip, who become the linchpins (borrowing an idea from Seth Godin, again), connecting disconnected systems, will lead the next revolution in medical discovery.

But Wait… Haven’t We Had FAIR For A Decade Now?

Doesn’t everyone love the FAIR data principles? Findable, Accessible, Interoperable, Reusable [15]. It’s like motherhood and apple pie; who could be against it?

Everyone, apparently.

Because while we preach FAIR, we actually practise FAKE:

Findable (if you know where to look), Accessible (if you know the right person, security-by-obscurity), Kinda-sorta-interoperable (let’s make up a new identifier, format, model, oh I forgot to tell you we have an undocumented API, that you can’t use), and Exclusive (for our lab use only).

The FAIR principles are like a diet everyone starts on January 1st. Great intentions, mediocre execution, ultimately abandoned by February [14]. We love to put the “FAIR compliant” sticker on our grants like “organic” on processed food, technically true, spiritually false. A bit like the AI buzzword, these days.

Here’s the uncomfortable truth: Real FAIR is expensive. Not just in dollars, but in change. It means redesigning processes. Retraining staff. Revealing weaknesses. Admitting that your revolutionary database is actually just a glorified Excel sheet with some elaborate and cumbersome extra steps that only you understand.

So we do this elaborate FAIR musical theatre dance instead. Jazz hands included for good measure. We create metadata that nobody reads. We deposit data in repositories that nobody searches [14]. We complete complicated Data Management Plans [16] not because we believe in them, but because we can check the compliance boxes while keeping our real data as unfindable, inaccessible, incompatible, and unusable as ever.

The FAKE FAIR is worse than NO FAIR at all. At least with no standards, everyone knows they’re in the wild west. With fake FAIR, we pretend we live in a civilised world while still shooting from the hip. And the FAIR cabal gets to beat us over it, not because we are not FAIR, but because we are not their version of FAIR. I was one of them, still am. I still believe in the principles. I just don’t hold on to them too militantly these days. But principles are just that, principles! [17].

FAIR Capability Maturity Model. From twitter.com/susheelvarma.

But here’s the thing about principles: They only work if you’re willing to pay for them. Organic food costs more. Democracy is inefficient. And real FAIR data requires investment in unsexy, hard graft and unglamorous infrastructure, rather than sexy discoveries.

The question isn’t whether we believe in FAIR principles. Everyone believes in them. The question is whether we’re willing to live with them when it makes things inconvenient, expensive, and exposes our data and the ecosystem’s dirty laundry.

Especially in the age of AI. For foundational AI to truly transform science, we must first establish the right foundations. Otherwise as Neil Lawrence tweeted a while back:

People often don’t understand the shocking state of our data ecosystems and how challenging it is to fix. AI looks like a magic wand, but it’s like installing a high-tech Japanese commode in a Victorian house that has no plumbing or sewage; the brown stuff is just going to hit the fan!

Real FAIR starts with admitting our current data ecosystem is ugly. Only then can we make it findable, accessible, interoperable, and reusable.

Anything else is just FAKE.

The Dark Data Problem

Here’s what nobody talks about: Dark Data.

It’s like dark energy in the universe; it’s invisible. Unmeasurable. But it warps and pushes everything around it.

Dark Data is FAKE in its purest form.

Unfindable: buried in filing cabinets, defunct hard drives, and forgotten folders. Inaccessible: locked behind expired passwords and retired professors. Incompatible: stored in formats that software stopped reading in 2003. Unusable: even if you found it, accessed it, and converted it, the context and metadata is gone.

But here’s the thing: Dark Data still exerts a gravitational force.

That clinical trial that “failed”? The one nobody published? It’s Dark Data now. That “negative result” you got? The one that you couldn’t convince a journal to publish? It’s Dark Data now. That dataset sitting on your grad student’s laptop who left three years ago? You guessed it; it’s Dark Data now.

But researchers still feel its pull. They avoid that research direction. They tell students, “Don’t go there.” They shape grants around their invisible boundaries. And worse, they let their research colleagues in other institutions (cough, cough, competitors) make the same mistakes they made.

We orbit around these black holes of knowledge, while we try our best to absorb enough matter to gain angular momentum to get escape velocity. In the meantime, these black holes just keep getting bigger, and there are more of them. Everywhere!

Conservative estimates suggest that 50% of all clinical trial data never sees the light of day [22]. That’s not just data in the dark. That’s half our scientific universe we’re pretending doesn’t exist.

The gravitational bubble is real. Every time we design a new study, we’re unconsciously influenced by all the studies we can’t see. Every hypothesis is shaped by the Dark Data we dance around but never acknowledge.

NASA can map dark energy. We can’t even admit Dark Data exists.

Until we do, we’re not doing science. We’re doing mythology.

Building the Biomedical Interstate Highway

When every medical device speaks an interoperable language, when every research database uses an interoperable protocol, when every clinical trial follows an interoperable data structure, and when people can cooperate despite their differences (notice I didn’t use the overused buzzword, collaborate), that’s when breakthroughs accelerate.

But what would this actually enable? Let me paint you a picture:

For Patients: Imagine arriving at an emergency room where doctors instantly access your complete medical history, not just from their hospital, but from every provider you’ve ever seen. No more repeating your allergies for the hundredth time. No more redundant tests because your results are trapped in another system. In the US, without an NHS to coordinate care, this is literally a matter of life and death.

For Researchers: Picture building a cohort of all 7 million Alzheimer’s patients in minutes, not years. Further, having access to rich, annotated datasets from diverse international populations provides the statistical power to detect rare genetic variants. The same data that helped develop an Alzheimer’s treatment in Boston is immediately available to validate results in Berlin. We could finally move from “this worked in 200 patients at our institution” to “this works in 2 million patients globally.”

For Health Economists: The interstate highway system generates $752 billion annually. What’s the ROI on biomedical data highways? McKinsey estimates that interoperability could save the US healthcare system $300B annually through reduced redundancy alone [21]. Add accelerated drug discovery, reduced clinical trial costs, and prevented medical errors, and we’re talking trillions.

For Research Breakthroughs: The mRNA vaccines that saved millions during COVID? Built on decades of “failed” cancer research. With true interoperability, we wouldn’t stumble upon these connections by accident; we’d systematically mine them. Every “failed”, “null hypothesis” experiment becomes a data point. Every negative result contributes to the map.

“But what about intellectual property?”, I hear you ask!

Fair point. I’m not suggesting we ought to live in some kumbaya data commune where everyone shares everything. Industry needs to protect legitimate IP. The next blockbuster drug won’t develop itself.

But here’s the thing: we can build highways with toll roads. We can create data trusts with access controls in place. We can enable pre-competitive collaboration while protecting competitive advantage.

The interstate highway system didn’t eliminate private property; it just made it more valuable by connecting it. Your land is worth more with highway access than without it. Same with data. Shared standards don’t mean shared data; they suggest that when you choose to share, it works. Private Big Tech works because of years of public investment that went into the creation of the Internet.

The cure for Alzheimer’s might already exist. It’s just stuck in a decade-long traffic jam between a lab in Tokyo and a hospital in Toronto.

How did this happen? Who’s to blame? Well, certainly some are more responsible than others, and they need to be held accountable. But again, truth be told… if you’re looking for the guilty, you need only look into a mirror. ~ V for Vendetta

The factories of the industrial age gave us assembly lines and mass production. The factories of the information age gave us data silos and mass confusion.

It’s time for post-industrial research, where connection trumps protection, where sharing multiplies value instead of dividing it, where the real competitive advantage isn’t in what you hide, but in how well you interconnect.

If you’re an adventure-seeker and you’d like to go on this transcontinental interstate highway journey (see what I did there) with me (US, UK, EU at least), like or comment on this post, and we can navigate the as-of-yet undiscovered, uncharted wild west that is the interoperability highway.

Having spent the past decade building international-scale health and biomedical data infrastructure; from architecting the UK’s HDR UK and Trusted Research Environment (TRE) network to shaping the UK regulatory landscape in AI and Data Science at the ICO to leading international standards initiatives like the GA4GH to leading projects that manage over 3.5PB of biomedical data at Sage Bionetworks [18-20]. I’ve learned that the data and technical challenges are only half the battle. There is going to be another key battle in the regulatory, ethics and governance space, especially around AI. And it starts with you, speaking truth to power!

If you’re up to the task, we’ll explore each siloed waypoint, understand each tribe, and decode each dialect. We’ll meet some of the most amazing linchpins already building bridges and the purple cows creating remarkable solutions. Most importantly, we’ll have fun along the way and discover why the most significant barrier to biomedical breakthroughs isn’t data, technology, process, or science; it’s human.

Because, in the end,

Cancer doesn’t care about your cute little standard, dataset, publication record, tenure, or stock options. Disease doesn’t respect intellectual property. And the patient waiting for a cure doesn’t have time for our childish games of data hide and seek.

Welcome to the Interoperability Revolution. How long will we wait to build our interstate highways? I’m getting started now. Are you?

Opinions expressed are solely my own and do not express the views or opinions of my employer, affiliations, or memberships.

References

The Interstate Highway System. In: The Geography of Transport Systems. Jean-Paul Rodrigue; 5 Nov 2017. Available: https://transportgeography.org/contents/chapter5/road-transportation/interstate-highway-system/
AAIC, Alzheimer’s Association International Conference. Available: https://aaic.alz.org/
AASHTO. A Policy on Design Standards Interstate System. Available: http://large.stanford.edu/courses/2014/ph240/suresh2/docs/AASHTO-InterstateDesignStandards.pdf
The Roads that Built America: The Incredible Story of the U.S. Interstate System. Available: https://books.google.com/books/about/The_Roads_that_Built_America.html?id=b0qO1yCFRxYC
Interstate Highway System. In: WTTW Chicago; 9 Jul 2018. Available: https://www.wttw.com/ten/modern-marvels/interstate-highway-system
Lehne M, Sass J, Essenwanger A, Schepers J, Thun S. Why digital medicine depends on interoperability. NPJ Digit Med. 2019;2: 79.
Cobey KD, Ebrahimzadeh S, Page MJ, et al. Biomedical researchers’ perspectives on the reproducibility of research. PLoS Biol. 2024;22: e3002870.
Begley CG, Ioannidis JPA. Reproducibility in science: Improving the standard for basic and preclinical research. Circ Res. 2015;116: 116-126.
Freedman LP, Cockburn IM, Simcoe TS. The economics of reproducibility in preclinical research. PLoS Biol. 2015;13: e1002165.
Torres-Espin A, Ferguson AR. Harmonization-information trade-offs for sharing individual participant data in biomedicine. Harv Data Sci Rev. 2022;4. doi:10.1162/99608f92.a9717b34
Gomes DGE, Pottier P, Crystal-Ornelas R, et al. Why don’t we share data and code? Perceived barriers and benefits to public archiving practices. Proc Biol Sci. 2022;289: 20221113.
Institute of Medicine (US). Barriers to Data Sharing. Sharing Clinical Research Data: Workshop Summary. National Academies Press (US); 2013.
March. Purple Cow Redux. In: Seth’s Blog; 23 Mar 2007. Available: https://seths.blog/2007/03/purple_cow_redu/
Hughes LD, Tsueng G, DiGiovanna J, et al. Addressing barriers in FAIR data practices for biomedical data. Scientific Data. 2023;10: 1-7.
Wilkinson MD, Dumontier M, Aalbersberg IJ, et al. The FAIR Guiding Principles for scientific data management and stewardship. Scientific Data. 2016;3: 1-9.
NOT-OD-21-013: Final NIH Policy for Data Management and Sharing. Available: https://grants.nih.gov/grants/guide/notice-files/NOT-OD-21-013.html
Varma S. FAIR Capability Maturity Model. Available: https://x.com/susheelvarma/status/1407650573378281472
UK Health Data Research Alliance, NHSX. Building Trusted Research Environments, Principles and Best Practices; Towards TRE ecosystems. doi:10.5281/zenodo.5767586
Guidance on AI and data protection. 2025. Available: https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/artificial-intelligence/guidance-on-ai-and-data-protection/
Sage Bionetworks. Available: https://sagebionetworks.org/
Healthcare data interoperability could save $300 billion annually. McKinsey & Company; 2022.
Song F, Parekh S, Hooper L, et al. Dissemination and publication of research findings: an updated review of related biases. Health Technol Assess. 2010 Feb;14(8):iii, ix-xi, 1-193. doi: 10.3310/hta14080.
Taylor JA, Crowe S, Espuny Pujol F, et al. The road to hell is paved with good intentions: the experience of applying for national data for linkage and suggestions for improvement. BMJ Open 2021;11:e047575. doi: 10.1136/bmjopen-2020-047575
Biswas J, Drogin EY, Gutheil TG. Treatment Delayed is Treatment Denied. J Am Acad Psychiatry Law. 2018 Dec;46(4):447-453. doi: 10.29158/JAAPL.003786-18.

Originally published on LinkedIn.

Susheel Varma

Chief Data Officer at Sage Bionetworks