An afternoon observation
On 2026-05-07, Chairman Chang Hung-jen — former Vice Director-General of Taiwan's Department of Health — visited me in Taichung together with Dr. Yang Chi-hang. We were meant to talk about iRehab, the orthopedic recovery platform I founded. But the conversation kept gravitating toward a larger question: "Can data actually follow the patient?"
They mentioned three platforms:
- The Taiwan Health Network (TwHealth Nexus), led by Academician Yang Pan-chyr (former president of National Taiwan University), capitalized at NT$1 billion by Wistron, Catcher, Delta, Elan, and Realtek — positioned as national-level health-data governance infrastructure.
- H2U Health, expected to list on Taiwan's Innovation Board in 2026 Q3 (ticker 7835), commanding ~70% market share of Taiwan's health-checkup centers.
- Far Eastern MedTech (FEMET) — eSIM + cross-border telemedicine + maritime mixed-reality consultation.
Plus my own pending conversation with AlleyPin (翔評互動) — a Taiwan PRM platform serving 2,000+ clinics with 19 HIS connectors, A-round funding NT$75M. That makes four.
Each describes itself with the same line: "data follows the patient."
But when I look at what each is actually doing, the world they describe hasn't quite arrived yet.
"Personally-Owned Data" Is a Half-Truth
Let me put the problem plainly.
"Patients should own their own health data" — in Taiwan's legal context — has three structural blind spots:
First, the Taiwan Medical Care Act mandates that medical records be held by medical institutions. Patients have a right to request a copy, but not ownership.
Second, Article 6 of Taiwan's Personal Data Protection Act (PDPA) treats medical data as special personal data — collection, processing, and use require specific purpose plus the data subject's written consent. The patient is a data subject, not a data owner.
Third, the payment structure dictates bargaining power. The NHIA pays for covered services; employers pay for occupational health; pharma pays for research access. Whoever pays sets the data terms.
So when the four platforms say "data follows the patient," what they're actually doing is:
| Company | What "follows the patient" really means |
|---|---|
| AlleyPin | Patient LINE messages + appointment data, institution-held, patient-viewable |
| TwHealth Nexus | GUID-based cross-hospital linking + Dynamic Consent — data stays in hospital, patient authorizes use |
| H2U | Health-checkup centers + H2U platform jointly hold data; the personal app surfaces it |
| FEMET | SIM card as identity key; medical data sits in cloud backends |
These are all variations of "institution-held + patient-authorized," not the reverse-ownership the rhetoric implies.
The four platforms are doing important work — they're solving custody, access, and consent. But "data follows the patient" — if it means a patient can hold their own data across institutions and across time — Taiwan isn't there yet.
This isn't the platforms doing wrong. It's the promise running ahead of the reality.
Splitting "Ownership" into Five Concepts
To clear the fog, "ownership" — a word too vague to be useful — should be split into five independently decidable concepts:
| Concept | Definition | Who Decides |
|---|---|---|
| Custody | Where the bytes physically reside | The system holding it |
| Access | Who can view / read / write | Authorization matrix (≠ Custody) |
| Agency | Who can trigger share / revoke / re-use | Governance layer |
| Continuity | Whether data stays coherent across time, institution, role | Market gap |
| Provenance | Who, where, and how the data was originally generated | Factual layer (immutable) |
Among these five, Provenance is the real anchor of the governance debate: most platform data has institutional provenance (created by clinicians during care), so governance authority naturally tilts toward the institution. iRehab's data has patient-generated provenance (created by the patient's own body, in their own home) — and this is what makes the other four concepts even arguable. Before debating who controls the data, ask who originated it.
Once you split this way, the story changes.
"Institutions hold the data, patients have access" — this won't flip in Taiwan anytime soon. But who holds Agency is an open battlefield, and who can deliver Continuity has no incumbent at all.
This is what I've slowly come to see across a year of building iRehab: the real paradigm shift isn't ownership — it's agency × continuity.
To put it differently:
Patients don't need to physically own every health record. What they need is durable, revocable, portable agency over their care journey.
This formulation upgrades "personally-owned data" from a half-truth to something that holds legally. I'd call it patient-mediated continuity.
A Land Where Governance Naturally Tilts to the Patient
Putting the four platforms and iRehab on the same timeline:
| Stage | Subhealth | Checkup | Chronic | Acute | Surgery | Discharge | Home recovery | Return visit | Long-term |
|---|---|---|---|---|---|---|---|---|---|
| Primary actor | H2U | H2U | — | — | Hospital | Hospital / Brief | iRehab | Brief | — |
The "home recovery" column is the only stage iRehab uniquely occupies — the period between discharge and the next visit, when the patient is most anxious, the family most uncertain, and the clinician most blind.
The legal status of data created during this stage is subtler than it appears:
- ROM measurements, self-reported PROM scores, exercise check-ins, wound photos, daily pain ratings — these aren't records the institution generated; the patient generates them, one entry at a time
- The Taiwan Medical Care Act defines "medical records" primarily as documents created and maintained by medical institutions through the care process. Whether home-entered patient data counts as "records" depends on who collects it, whether it enters the care workflow, and whether medical professionals use it — case-by-case
- Article 6 of the PDPA is broader than "medical records": it covers special-category data including medical, healthcare, and health-checkup data. So even if home-entered data isn't a "medical record," it remains highly protected
- The defensible conclusion isn't "the data legally belongs to the patient" (overstated) — it's: the share / revoke / continuity design for this data category most naturally originates with the patient
So iRehab occupies a strange position: we're not competing with the four platforms over "who owns the data." We sit on a land that all four cannot easily reach — post-discharge dynamic behavioral data.
This land has two properties:
- Legal attribution naturally favors the patient
- No institution has generated it, so no institution can claim ownership
What iRehab has done over the past 18 months is build slowly on this land.
Seven Shipped Pieces of Evidence
If patient-mediated continuity is a thesis, what does it look like in iRehab? The past 18 months have produced seven implementation proofs — four are direct patient-governed designs, three are adjacent governance primitives (e.g., cross-clinician handoff, edge-case authority boundaries):
| Date | Shipped Feature | What It Demonstrates |
|---|---|---|
| 2026-03-18 | iRehab Open Ecosystem DS-01~DS-05 (Walkaway Test) | "Make leaving easy and people don't leave" — clinician-side data sovereignty |
| 2026-04-13 | Patient AI Prep (patient exports a PHI-filtered summary to their own subscribed ChatGPT/Claude/Gemini) | Patient governs the export decision; iRehab owns PHI filtering and the safety boundary; zero LLM cost expansion to upstream AI |
| 2026-04-14 | Family Link (family members get independent identities, invite-able and revocable by the patient) | Multi-actor first-class identity; observer-as-governor in concrete form |
| 2026-04-16 | Pre-consult push, not pull (patient fills the form, then pushes to the hospital — hospital does not pull from iRehab) | The patient is the courier, not the source being scraped |
| 2026-04-17 | Phase 4 Handoff/Handshake (13 ADRs, D1–D13) | Cross-clinician / cross-institution continuity infrastructure with consent atoms |
| 2026-04-26 | Five immovable governance boundaries for an athlete-medical system across institutions (anti-doping, tiered break-glass, retention hold, etc.) | A reference design that pushes patient-governed under stress: multi-stakeholder + high-risk + cross-institution |
| 2026-05-05 | Brief Open Schema (5-stage journey × U-curve) | Cross-episode structured handoff format; an open-schema candidate for a national standard |
Two observations about this inventory.
First, patient-mediated continuity wasn't a thesis born on 2026-05-07. It's an 18-month phenomenon I'm only now putting one sentence around.
Second, while the four platforms are starting to ask "can data follow the patient?" — iRehab has been shipping answers. Not "a year ahead" — that comparison is the wrong frame. We've simply built some houses in this narrow stage of post-discharge to return-visit, and they work.
Observation Is Intervention
There's a deeper root to this thesis.
I keep a cross-domain note in my own knowledge base recording three completely unrelated fields:
- Personal knowledge management (cyberbrain / Heptabase: externalize chaotic thought into text → self-calibration)
- Implantable sensor physics (LC resonance: convert invisible bone-healing mechanics into a frequency signal → measurable SaMD)
- Disaster-zone operations (xGrid Event Sourcing: each operation recorded as an event → resilience)
Their shared logic is: none of them improve outcomes through control — they do it through observation.
Observation is itself intervention. Faithful recording is the most powerful improving force.
iRehab's Recovery Loop is the fourth case of this logic. We don't ask the patient to change, the clinician to change, or the institution to change. We just make the period between discharge and the next visit — previously invisible — visible.
Once it's visible, all three sides self-correct:
- The patient sees their own progress trajectory → self-calibration
- The family sees the patient is doing the work → reassurance + timely nudges
- The clinician sees the trend → return visits don't waste five minutes asking "how's it been"
There's a name for this kind of logic. Academically it's called Safety-II — a safety paradigm that begins by understanding why things go right most of the time. Traditional Safety-I asks "why did it fail?" Safety-II asks "why does it succeed?"
My wife happens to be doing Safety-II research on postpartum hemorrhage for her master's thesis, using a method called FRAM. We've recently been talking about an interesting reverse application: FRAM is normally used to analyze why most surgeries go right; can it be turned around to analyze why some patients drop out of home rehab?
The answer is yes — and iRehab has prepared exactly the kind of data Safety-II researchers wish they could collect: every time a clinician decides to deviate from the standard protocol mid-recovery, the system automatically captures the change, the timing, and the rationale. This kind of "what clinicians actually do, not what the textbook prescribes" record is exactly what Safety-II calls Work-as-Done (WAD), and it's notoriously hard to collect — clinicians move on without leaving structured traces. iRehab, by design, requires every protocol deviation to be written down, so this record is a free byproduct.
That's another story for another post. But at the deepest layer, patient-mediated continuity is Safety-II made concrete on the patient side.
Why This Matters Now
Back to that afternoon on 2026-05-07.
When Chairman Chang mentioned the four platforms, my first instinct was "how can we cooperate?" An hour later, something different had crystallized:
These four aren't four competitors — they're four pieces of a puzzle, each at a different layer. The layer iRehab occupies — "post-discharge dynamic behavior" — is currently underserved, and not easily reduced into a feature inside any one platform.- AlleyPin is the clinic distribution layer
- TwHealth Nexus is the national RWD governance layer
- H2U is the subhealth / checkup / employer-health layer
- FEMET is the cross-border access layer
- iRehab is the recovery continuity layer (the missing middle)
Put differently: the four platforms are building macro-institutional infrastructure — connecting hospitals, employers, borders. Necessary work. iRehab is building the micro-behavioral application layer — making what patients do at home visible, shareable, portable. Infrastructure without an application layer is scaffolding; an application layer without infrastructure is an island. We can plug into all four simultaneously, let each legitimately reference patient-originated recovery data, without slotting ourselves under any one — and without framing them as opponents.
This thesis carries three larger implications:
First, for regulators: Taiwan's PDPA and Medical Care Act have a legislative gap when it comes to patient-generated home health data. Whoever first establishes governance norms for patient-mediated data shapes how the eventual amendments converge.
What about Taiwan's My Health Bank (健康存摺), then? This deserves a direct answer. The NHIA's My Health Bank is impressive infrastructure, but its core is medication, lab, and visit records — claims-based continuity — meaning the patient can retrieve, retrospectively, what already happened inside institutions. iRehab fills a different layer: clinical behavioral continuity — what the patient is doing at home, day by day, between visits. One is retrospective; the other is dynamic. They're not in competition. They're complementary.
Second, for patients: you don't need to wait for the government, the hospital, or any platform to grant you data sovereignty. You're already generating your own data. iRehab simply turns it into something you can see, share, and take with you.
Third, for Taiwan's medtech industry: this frontier has no leader in East Asia. Japan (the MHLW + METI joint PHR push, integrated with My Number Portal), Korea (MyHealthWay, integrated with 600+ hospitals), Singapore (Synapxe-managed HealthHub + NEHR) — each is experimenting with personal health-data platforms, but most still sit at "institution-holds + patient-authorizes." If Taiwan can, in 1–2 years, package patient-mediated continuity into an exportable reference vertical, that's a national-level industrial position.
In Closing: A Provocation, Not a Call to Action
This piece isn't trying to recruit you to use iRehab.
It's an invitation to pause and re-examine the four platforms — and the fifth and sixth that will follow. The next time you hear "our data follows the patient," ask three questions:
- Where is the Custody? (Who physically holds it?)
- Whose hands hold the Agency? (Who can share / revoke?)
- What does the Continuity span? (Cross-institution? Cross-time? Cross-device?)
If the answer is "we hold it, we authorize, continuity within our walls" — that's progress, but it isn't yet patient-mediated.
What does true patient-mediated continuity look like? Something like this:
I switch hospitals, switch clinicians, switch countries, switch phones — and my recovery trajectory comes with me. I can show it to my family, to a new doctor, to my own AI subscription, anytime. I can revoke it just as easily. I don't need permission from any platform first.
That world hasn't arrived.
But what iRehab has done over the past 18 months is ship a small piece of it — post-discharge dynamic recovery behavior. If we can hold this piece, the next is chronic disease management. After that, longitudinal health checkups. After that — the next-generation health-data infrastructure for Asia.
This thesis emerged from the afternoon meeting on 2026-05-07 with Chairman Chang Hung-jen and Dr. Yang Chi-hang, due diligence on four platforms, multi-round review by three AI reviewers (Gemini Deep Think + ChatGPT + Codex), and the seven reference implementations iRehab has shipped over the past 18 months. The full strategic brief and technical ADR remain internal.
Next post: Safety-II on the patient side — when iRehab's protocol-deviation corpus becomes Work-as-Done evidence.