Introduction

If we are serious about better vitiligo outcomes, we need to stop treating registries, biobanks, trials, patient support, and AI tools as separate planets. They are one system.

And in that system, foundational infrastructure cannot be built as a collection of separate entries. A registry here, a biobank there, an app somewhere else, and a pilot AI companion bolted on later is not an ecosystem. It is a patchwork.

Vitiligo needs something more coherent: an AI-powered ecosystem in which registries, biosamples, clinical workflows, patient support, and real-world evidence are designed to inform one another continuously.

That is the real infrastructure challenge.

A vitiligo registry, then, is not merely a nice extra. It is one foundational layer of an ecosystem that must be built as an integrated whole. The same is true of biobanks, patient-facing digital tools, and the “last mile” systems that determine whether patients remain engaged long enough for care and research to mean anything.

This matters for a simple reason: even the most powerful AI can make the problem worse.

AI is a hyper-efficient amplifier. It will optimize whatever target the system gives it. If the target is enrollment speed, it will accelerate enrollment. If the target is completed forms, it will collect forms beautifully. If the target is short-term, publishable, easy-to-measure proxies, it will optimize those too.

But none of that guarantees better continuity, better adherence, better quality of life, or better long-term outcomes for patients with vitiligo.

So the challenge is not merely to deploy AI. The challenge is to build the right system for AI to serve.

Why infrastructure matters now

Future progress in vitiligo will increasingly depend on high-quality, longitudinal, real-world data: who gets the disease, how fast it progresses, what treatments patients actually receive, what works in routine practice, what happens after the first few months, what happens after relapse, and what happens long after the neat endpoints of a clinical trial have come and gone.

Without that, we are flying half-blind.

Vitiligo has long suffered from a fragmented data ecosystem. Clinical trials can tell us important things, but they do so within a narrow frame. They capture selected populations, fixed time windows, and protocol-defined outcomes. They rarely capture the full patient journey, the burden of maintenance, the reasons for discontinuation, the psychosocial wear and tear, or the difference between early response and durable control.

Registries can do that. Biobanks can add translational depth. Patient-facing digital layers can keep the system alive between visits. Real-world evidence can connect what happens in the clinic to what happens in actual life.

But only if these pieces are designed to work together.

This is why the field needs to move beyond isolated tools and toward connected architecture. Not because “ecosystem” is a fashionable word, but because the underlying disease reality demands it. The VITAL initiative was launched precisely to define what should be measured not only in vitiligo clinical trials, but also in registries and clinical practice. In other words, the field itself has already acknowledged that these domains cannot remain siloed forever.

Vitiligo is visible, chronic, psychologically loaded, slow to improve, and prone to relapse. It is exactly the kind of condition where fragmented infrastructure quietly sabotages good science.

CloudBank: a good idea, too early

More than a decade ago, the Vitiligo Research Foundation began building unusually ambitious infrastructure: a federated Vitiligo BioBank linked to the Vitiligo CloudBank, an electronic health record concept designed (PDF 1, PDF 2) specifically for the condition (read more).

At its peak, this effort operated across 11 locations in 9 countries, with a central biorepository overseas and distributed collection and storage sites connected through cloud-based infrastructure.

In retrospect, CloudBank was both well-meant and well-designed. It was also, in practical terms, a false start.

Not because the concept was wrong. The concept was ahead of its time.

The deeper problem was that the surrounding system was not ready. Disease-specific registry infrastructure was still treated as a niche exercise rather than core translational machinery. Scaling pathways were weak. Governance models were immature. Long-term operational support was fragile. Then came COVID-19, geopolitics, and the forced hibernation of physical biobanking operations.

A biobank without durable clinical integration, patient engagement, operating pathways, and long-horizon support is always vulnerable. Science tends to admire the first mile. The last mile is where elegant ideas often quietly run into a wall.

Still, it would be wrong to call CloudBank a failure.

The broader effort contributed clinical data and biosamples that supported major international work and helped prove that vitiligo could be studied through serious, networked infrastructure. The 2016 Nature Genetics paper led by Ying Jin and colleagues identified 23 new vitiligo risk loci and helped clarify the disease’s genetic architecture. In that sense, CloudBank was less a dead end than an early signal. It showed what the field would eventually need, even if the field was not yet capable of sustaining it.

The global registry gap

The registry landscape is becoming more concrete, but it remains uneven.

In the UK, serious progress is being made. VIRTUAL-UK was set up in 2025 as a long-term multicentre observational registry designed to collect real-world safety and effectiveness data in adults and children with vitiligo. In parallel, the VOICE registry and bioresource is emerging as a high-fidelity specialist dataset. These are important developments, and they suggest that the UK is beginning to assemble visible backbone infrastructure.

But they are still early. They are not yet a broad, mature, fully integrated patient ecosystem.

In the US, the gap is more obvious. There is still no comparably visible public-facing national vitiligo registry backbone serving as an operational platform for long-term follow-up, AI-enabled support, and bidirectional patient-science learning.

This is the registry gap. Not a total absence of activity. Not a total absence of research. But a shortage of mature, connected, longitudinal infrastructure.

And that shortage matters because AI always lands inside whatever system already exists. If the underlying infrastructure is fragmented, AI inherits fragmentation. If the workflows are shallow, AI scales shallowness. If the incentives are misaligned, AI optimizes the wrong things with frightening efficiency.

AI and the “last mile” problem

Artificial intelligence is already changing medicine. In trials and registries, it can help parse records, summarize histories, support eligibility review, organize data, and reduce certain kinds of administrative friction.

Fine. Useful. Welcome.

But too much of the AI conversation still focuses on the front end: faster matching, faster recruitment, faster startup, faster form handling. That is not where the real struggle lies.

A system does not fail only because it cannot find patients. It fails because real people struggle to remain engaged, understand what is happening, tolerate the treatment burden, manage the logistics, report outcomes over time, and stay connected long enough for care and research to mean something.

That is the last mile problem. And in vitiligo, it is not a side issue. It is central.

The disease is visible. The emotional burden can be heavy even when the clinical surface area looks modest. Improvement is often slow. Maintenance is often tedious. Relapse is common. Frustration is baked into the experience. The issue is not simply getting patients into the system. It is building a system worthy of keeping them there.

The AI translation layer

This is where AI can be genuinely useful.

Used responsibly, large language models can function as translation layers between formal medicine and lived patient reality.

• They can convert raw records into understandable timelines.
• They can explain treatment pathways in plain language.
• They can support symptom logging and help patients prepare for visits.
• They can improve continuity between appointments.
• They can collect patient-reported outcomes between visits.
• They can help bridge the gap between “patients have access to their records” and “patients can actually make sense of their records.”

That is not glamorous. It is also exactly where many real benefits live.

The biggest opportunity may not be that AI can help recruit the patient.

It may be that AI can help not lose them.

The hidden frictions AI runs into

This is where the cheerful AI story usually starts to wobble.

AI does not enter a clean system. It enters a messy one. Then it amplifies whatever that system already rewards, tolerates, and neglects.

1. Proliferation of pilots without scaling pathways

Vitiligo does not lack promising pilots. It lacks repeatable pathways for expansion. Small digital efforts can work beautifully in one clinic, one study, or one highly motivated site. Then they hit the wall: inconsistent workflows, uneven staffing, different data models, unclear ownership, weak funding logic, and no operational backbone to support multi-site growth.

This is the pilot trap.

The UK efforts matter precisely because they are trying to move beyond that trap. But even there, the field is still early. VIRTUAL-UK is a serious start, not a finished national machine. VOICE is valuable, but specialist excellence is not the same thing as broad ecosystem integration. In the US, where no unified vitiligo registry backbone is publicly visible, the scaling challenge is even more obvious.

2. Productivity gap and re-absorption of gains

AI can save time on matching, chart review, or data parsing. Wonderful. But if the surrounding workflow is not redesigned, those gains get swallowed almost immediately by manual follow-up, missing patient-reported outcomes, adherence problems, dropout management, and the same old treatment-fatigue issues.

The machine gets faster. The process stays broken.

Vitiligo is especially vulnerable here. Repigmentation often takes months. Maintenance matters. Side effects matter. Frustration accumulates. Patients lose steam. Front-end efficiency and real-world continuity are not the same thing, and pretending otherwise is one of the lazier sins of modern health tech.

Real-world treatment data make the point rather brutally: in one US claims analysis of 19,335 newly diagnosed patients, 49.9% received no vitiligo-related treatment during the first year after diagnosis. That is not a mere intake problem. It is a continuity problem.

3. Burden of process debt and legacy fragmentation

Vitiligo care still runs through patchy pathways. PRO capture is inconsistent. Phototherapy practices vary. Documentation varies. Handover between sites is often weak. Some services are structured. Others are held together with memory, goodwill, and mild panic.

AI can expose this mess in high definition. It cannot, by itself, resolve it.

Without standardized workflows and shared outcome logic, an AI companion added on top of fragmented care is like putting a very smart dashboard on a car with missing wheels. It may look impressive right before it goes nowhere.

4. The identity problem of tribal knowledge

Some of the most important knowledge in vitiligo is tacit.

Experienced clinicians know how to judge subtle repigmentation, when lesion location matters more than body surface area, when a patient is losing patience, and when a technically mild case is psychologically devastating.

Patients carry another kind of expertise: what concealment feels like, how stigma alters behavior, how treatment burden competes with ordinary life, and what “worth it” means outside a research form.

This knowledge does not externalize neatly. It does not fit naturally into a dropdown menu or form-completion metric.

If AI systems ignore it, they become shallow. If they flatten it into tidy but tone-deaf prompts, patients may disengage. In a disease with such visible psychosocial burden, that is not a side effect. That is system failure wearing a polite digital smile.

5. Governance in an agentic world

A one-off summarization tool is one thing. An always-on AI companion that interprets records, nudges adherence, collects outcomes, and feeds a registry is something else entirely.

Now the questions multiply.

• Who is accountable for errors?
• How is bias audited?
• How is privacy protected?
• How does the system perform across different skin types, literacy levels, languages, and digital access conditions?
• Where does human review sit when psychological burden is high and nuance matters?

These questions cannot be treated as annoying legal garnish. Governance has to be part of the architecture from the start. Once AI becomes an active participant in the ecosystem rather than a passive utility, the governance burden becomes heavier, not lighter.

6. Architectural complexity and vendor patchwork

Everybody says “ecosystem” until it is time to connect the actual machinery.

Then suddenly we have EHRs, registry platforms, patient apps, sponsor dashboards, image repositories, clinic records, spreadsheets lurking in corners, and one more vendor promising to “unify the journey.”

In the US, where no clear vitiligo backbone exists, any serious AI layer would have to bridge highly disparate systems. In the UK, the pieces are more promising but still not seamlessly unified.

AI often arrives after years of institutional patchwork and is then asked to behave like a universal solvent. It is not.

7. The efficiency trap versus value re-creation

If AI is framed mainly as a speed tool or cost-cutting tool, its ambition stays small.

Faster recruitment is useful. Faster forms are useful. Better summaries are useful. But these are not the true prize.

The true prize is value re-creation: turning patients from passive subjects into active contributors, turning registries into living infrastructure, and turning clinicians from overloaded data clerks into interpreters and orchestrators of a continuous learning loop.

In vitiligo, that broader redesign matters far more than shaving a few minutes off onboarding.

Why these frictions are particularly acute in vitiligo

These frictions exist in many chronic diseases.

They are particularly acute in vitiligo for four reasons.

1. Vitiligo is chronic and treatment is often slow. That makes continuity and retention unusually important.
2. Vitiligo is visible. Psychosocial burden can be intense even when traditional clinical metrics look modest.
3. Treatment fatigue is real. Maintenance, uncertainty, relapse, and perceived inefficacy all wear people down.
4. Mature registry infrastructure is still limited. That means the field is trying to build advanced AI layers on top of a foundation that, in many places, is still being poured.

This is also why earlier efforts such as CloudBank stalled. The issue was not simply technology. It was the lack of a mature operating ecosystem around the technology.

Goodhart’s trap in vitiligo

There is, however, a deeper and darker issue.

AI may not merely fail to save the system. It may become the perfect accelerator of its degradation.

This is Goodhart’s law in its most dangerous form: when a measure becomes a target, it stops being a good measure. AI does not create that problem. It industrializes it.

In vitiligo, the risk is especially visible.

• Scientists can optimize publishability and short-term statistical neatness instead of durable outcomes such as stable repigmentation and better quality of life.
• Startups and sponsors can optimize investment appeal and enrollment velocity.
• Clinics can optimize procedure volume, workflow throughput, and reimbursement logic.
• Payers and regulators can optimize the kind of real-world evidence that is easiest to count rather than the outcomes that matter most.
• AI teams can optimize model accuracy, throughput, bias scores, and pretty dashboards while leaving the actual patient problem largely untouched.

Each local optimization looks rational. Each one is rewarded. Each one can be defended in a meeting.

Taken together, they move the whole system further from the true goal.

That is what makes AI dangerous in a proxy-driven environment. It does not merely automate activity. It hyper-optimizes the wrong target.

The machine is not confused. It is doing exactly what it was asked to do. The confusion belongs to us.

How Goodhart’s trap manifests in vitiligo

We can already see the outline. A registry can collect excellent safety data, quality-of-life data, and longitudinal variables. But if the operative KPI becomes number of enrolled patients, percentage of forms completed, or speed of data capture, then an LLM companion will simply make those proxies faster and prettier.

Meanwhile, true adherence, durable repigmentation, reduced relapse, restored normalcy, and actual life impact remain off-stage.

CloudBank can also be read through this lens. The system succeeded in organizing data and biosamples, which were real achievements. But the harder task was never collection alone. The harder task was creating a genuinely bidirectional loop between patient and science, in which participation returned practical value to the patient and learning flowed back into care.

The proxies were easier to build than the loop. That is why the next chapter must be sharper in its design logic.

AI will not solve the vitiligo problem on its own. It will solve the problem we ask it to solve. If we give it proxies, it will optimize proxies to perfection.

If we give it the real goal, and structure the system around that goal, then it can become a tool rather than an accelerator of drift.

What realignment would actually look like

This has practical implications. Do not start with “Where can we add AI?” Start with “What is the real outcome we refuse to lose sight of?”

In vitiligo, that outcome is not more dashboards, more forms, more site activity, more pilot slides, or more biosamples.

It is closer to what patients actually care about: stable repigmentation where possible, lower burden, less confusion, better continuity, lower relapse, more trust, and a life less organized around disease.

From there, the design logic becomes clearer.

• Start with clean-sheet registry design that is patient-facing from day one.
• Use AI as a translation and continuity layer, not merely as a recruitment trick.
• Externalize tacit knowledge where possible through structured PROs, contextual data, and carefully supervised patient companions.
• Build centralized governance for bias audits, privacy, accountability, and equity.
• Design the architecture around bidirectional value: data should improve care, and care should improve data.
• Measure success not only by extraction and throughput, but by retention, comprehension, continuity, trust, and meaningful long-term outcomes.

This is less flashy than another “AI-powered trial acceleration” slogan. It is also far more likely to matter.

The next chapter

The future is not just smarter recruitment.

It is not just better registries. It is not just more AI.

It is a bidirectional ecosystem in which registries, biobanks, clinical care, patient support, and real-world evidence continuously inform one another. It is a system in which patient participation is not treated as a one-way donation of data, but as an ongoing relationship that returns clarity, support, and practical value. It is a system in which AI is neither worshipped as a cure-all nor dismissed as a gimmick, but placed where it is genuinely useful: translating complexity, reducing friction, supporting continuity, and holding together the parts of medicine that tend to fall apart between visits.

Most of all, it is a system designed around the real goal rather than the easiest proxy.

Without that realignment, even the most powerful AI will simply make the current vitiligo system faster, prettier, and more efficient at missing the point.

That would be quite an achievement. Just not the one patients need.

Yan Valle
Prof. h.c., CEO
Vitiligo Research Foundation
New York, March 2026