TL;DR — Taiwan’s 30 catastrophic illness categories map to five validity tiers (permanent, five years, three years, one year, case-by-case). In practice, the “case-by-case” tier also absorbs shorter periods such as three or six months; this article merges them for clarity. The design is not administrative convenience but reflects thirty years of NHI’s accumulated actuarial reasoning on disease prognosis and long-term cost. Each tier corresponds to a kind of “status judgment” — whether the disease is reversible, whether it needs long-term follow-up, whether cure is possible, and whether it follows an acute event. Viewed through an AI-era lens, this system looks remarkably like an early prototype of tiered reimbursement, already rooted in Taiwan’s local design.

In the previous piece, I mentioned that while preparing materials for a Japanese partner, I opened the NHIA’s “Catastrophic Illness Categories” page. It lists 30 categories, each with a field labeled “validity period of certificate.”

Look at these periods closely and a curious set of contrasts emerges.

Rare diseases: permanent. Most congenital metabolic disorders: permanent. Post-organ-transplant follow-up (except bone marrow): permanent. Motor neuron diseases: permanent. Aplastic anemia and cirrhosis with severe complications: five years. Thyroid cancer, breast cancer stage I, cervical cancer stage I, oral cancer stage I: three years. Myasthenia gravis: three years. Burns covering 20% or more of total body surface: one year. Acute cerebrovascular disease: under NHIA rules, physician determination within one month of acute onset.

They are all catastrophic illnesses under the same legal framework, yet the validity periods differ dramatically. Why?

Put simply: rare diseases are treated by the system as long-term, irreversible conditions requiring lifelong care, so they receive “permanent.” Cancers, by contrast, vary widely in prognosis across types and stages thanks to advancing treatment, and many early-stage cancers can remain stable long-term after treatment. The system therefore uses three or five years as a re-confirmation cadence.

This article tries to answer that question more fully. The design logic behind these periods runs deeper than it first appears, and is worth revisiting through an AI-era lens — because this thirty-year-old “tiered duration” system already carries the embryonic form of tiered reimbursement, mirroring exactly the actuarial and case-specific trends now emerging in the AI era.

Tiered durations are more than administrative convenience — they reflect actuarial reasoning

Most people, on first encountering this table, will instinctively read it as either administrative shorthand or a trade-off in resource allocation. In fact, this period design can be understood as one accumulated outcome of three decades of NHI’s institutional actuarial work. “Actuarial” here does not mean only financial calculation; it includes integrated judgment on disease prognosis, long-term treatment cost, and re-confirmation frequency. The same validity table simultaneously reflects whether a disease is reversible, how fast medical technology is advancing, and multiple other considerations of institutional design.

The reason this logic needs a tiered structure is that the system has to answer several questions at once: Will this diagnosis change in the future? Does the patient need lifelong medication or follow-up? Is there a possibility of leaving the catastrophic illness state? With advances in medicine, will today’s “severe disease” become curable or controllable in five years?

The answers differ across diseases. They also differ across time for the same disease. Tiered durations are the concrete materialization of these “status judgments,” matching each disease to an appropriate re-evaluation cadence.

Five duration tiers, mapped to five status judgments

If we step away from the legal text and re-sort the 30 categories by their validity periods from a clinical-course and prognosis standpoint, we see a structured distribution that falls broadly into five status judgments. NHIA’s published table also includes shorter periods such as three or six months in practice; for clarity, this article folds those into the “case-by-case” tier.

Permanent covers diseases that are “irreversible” and “require lifelong follow-up.” These include rare diseases (designated under the Rare Disease Control and Orphan Drug Act), most congenital metabolic disorders, motor neuron diseases, chromosomal abnormalities, post-organ-transplant follow-up (except bone marrow), and systemic autoimmune syndromes requiring lifelong treatment. The common feature is that once diagnosed, the patient’s condition typically will not return to “normal,” and under current medical technology, the disease is rarely understood through the lens of simple “cure.” A permanent validity period reflects the system’s recognition of the long-term and irreversible nature of these conditions.

Five years covers diseases that are “effectively treatable but still require long-term follow-up.” These include aplastic anemia, cirrhosis with severe complications, post-bone-marrow-transplant follow-up, congenital immunodeficiency, among others. The common feature is a relatively stable disease course with room for change, requiring periodic re-evaluation. The number “five years” is more than an administrative figure: it shares a comparable timescale with the clinically familiar concept of “five-year survival rate” used to observe long-term prognosis, though this does not mean the two are institutionally equivalent.

Three years mostly covers early-stage cancers with high cure rates: thyroid cancer, breast cancer stage I, cervical cancer stage I, and oral/oropharyngeal/hypopharyngeal cancer stage I. Their common feature is high curability with early intervention, and an absence of recurrence within three years of treatment generally indicates that the therapeutic outcome is stable. The design reflects how the system uses time to observe whether “treatment is durably successful.”

One year is the recovery period after an acute event. It covers burns over 20% of total body surface area and first-occurrence major trauma (Injury Severity Score, ISS ≥ 16). The common feature is that the event has a clear onset, and initial prognosis can usually be assessed after roughly one year of recovery.

Case-by-case is the highly dynamic medical situation, where the validity period adjusts with patient status. For chronic renal failure requiring dialysis, before long-term dialysis is confirmed the validity is usually only a few months, requiring frequent re-evaluation; once long-term dialysis is confirmed, it switches to permanent. Long-term ventilator dependency follows a similar logic: 42 days for first determination, three months upon recurrence, and one year from the third determination onward, gradually relaxed by duration of use. Acute cerebrovascular disease falls in this category as well — under current NHIA rules, it can be determined by the attending physician within one month of acute onset, with partial-payment exemption applying without prior issuance of a certificate. Actual validity periods and determination methods should follow the NHIA’s latest announcements and physician judgment. The common feature is that patient status changes rapidly; the system cannot rely on a single fixed period, and must preserve room for physician judgment in context, allowing short-term periods to progress into long-term or permanent ones.

Five tiers, five status judgments. Seen together, this is not merely a resource-allocation rule. It is a disease classification system that carries dynamic evaluation and actuarial logic.

Local Taiwan design layered on top of ICD-10

As I wrote in Part 1, I did not first encounter ICD-10 through Wikipedia. I studied life science in college and was a medical technologist before that. Back then, this disease classification system ran on the computers of the clinical floor.

But this same coding system later became part of daily life at home as well. A family member of mine entered the catastrophic illness system because of cancer; the certificate had to be reissued every three years, every follow-up visit meant waiting for tracking results. That rhythm — I have watched it, and I have walked through it. So later, when I opened that NHIA page, I did not just see a table of 30 categories. I saw the time behind each renewal that the table itself does not record.

But ICD-10 codes by themselves do not tell you “how long this disease’s validity period should be under Taiwan’s catastrophic illness system.” ICD-10 is the international classification of diseases published by the WHO, universally used to describe how diseases are clinically categorized. Taiwan NHI’s tiered duration design ties this international coding to Taiwan’s local reimbursement system, then assigns each diagnosis to one of the permanent, five-, three-, one-year, or case-by-case tiers based on whether the disease is reversible, whether it requires follow-up, and whether cure or stable control is possible. NHIA announcements show that each time the ICD-10 version is updated (for example, the 2023 ICD-10-CM/PCS revision), the diagnostic-code scope and crosswalk for catastrophic illnesses is updated in sync.

To make this correspondence quickly searchable, I compiled the 30 categories along with their ICD-10 codes, validity periods, and sub-item breakdowns into a four-language structured query tool. The tool is currently based on the NHIA’s “Catastrophic Illness Categories and Validity Periods” announced on 16 September 2024 (ROC 113), effective from 1 January 2025, and also notes the NHIA’s subsequent update to the rare-disease diagnostic-code crosswalk on 28 April 2025. Actual applicability should follow the NHIA’s latest announcements. The tool shows how the same ICD-10 code, once it enters Taiwan’s NHI system, is further divided into main and sub-categories with different durations, at a granularity far beyond what international coding alone can represent.

In other words, what we see in the 30 categories is not a plain ICD-10 classification, but a version of ICD-10 that has been filtered, processed, and tiered as it entered Taiwan’s NHI system. These institutional judgments have been continually updated through successive revisions. The NHIA website currently shows revision records for at least the 2015, 2019, 2022, and 2024 versions of the “Catastrophic Illness Categories and Validity Periods.” As medicine advances, some diseases that originally carried a “five-year” period may be adjusted to “permanent”; some that were “permanent” may be reset to periodic renewal because new treatments have emerged. Cancer is a case in point: the NHI once changed certain cancer categories from permanent to 3-year or 5-year renewal, reflecting how improvements in cure rates and five-year survival have led the system to recalibrate its evaluation cadence.

I tend to think of this system as a long-running classifier that has been continually updated and fine-tuned. Its inputs are diagnoses, disease courses, and case status. Its outputs are validity periods, reimbursement eligibility, and recognized scope. In other words, even though it was not built with AI, it already has several of the basic features that AI-era data systems need: well-defined inputs, traceable classification rules, version-controlled updates, and outputs tied to reimbursement results. Over three decades, this classifier has accumulated remarkably fine-grained judgment rules through repeated institutional revisions.

Why this system is an early local prototype of AI-era tiered reimbursement

If we view this system through the lens of a “classifier,” another thing becomes visible.

Health systems around the world are facing the same question: when medical resources are finite and patient needs are increasingly individualized, how should reimbursement be allocated? The traditional answer leans toward “one diagnosis, one reimbursement level.” This design worked well enough through the twentieth century, but its limits have become increasingly clear in an era of precision medicine, gene therapy, cell therapy, and AI-driven imaging.

The U.S. Medicare’s Value-Based Payment, the U.K. NHS’s NICE assessment framework, and Japan’s MHLW’s Designated Intractable Disease Medical Beneficiary Certificate system are exploring “case-specific, outcome-based, risk-tiered reimbursement” along different paths. These trends share a common technical premise: the system itself needs a sufficiently fine-grained, machine-readable, continually updatable framework for disease classification and status assessment.

Looking back from today, Taiwan’s tiered duration design already contains many of the elements that tiered reimbursement systems require: clear categories, period design, dynamic determination, version updates, and rules tied to reimbursement eligibility. It has run for three decades, is keyed to ICD-10, and has accumulated mechanisms for “case-by-case” dynamic judgment. It was not designed for AI, but it fits remarkably well into AI-era reinterpretation. Put another way, what Taiwan lacks in approaching AI-era tiered reimbursement is not the institutional logic itself, but how to re-interface and re-structure this existing logic into a data foundation that algorithms can call in real time.

This brings us back to a point from Part 1: the data is already there. The interface layer is the next evolution.

Behind one administrative table, a policy asset

This article is not trying to render a verdict on any institutional design, nor to evaluate it on a good-or-bad scale. All contemporary health systems are still evolving. Whether it is Taiwan’s 30 categories or Japan’s designated intractable disease system, none is a finished endpoint — each is still being adjusted.

What I want to do is take a piece of data that is often treated as “just an administrative table” and place it back into its proper policy context. This is an actuarial classification system that has been operating for thirty years, and from a design standpoint it can also be seen as one of the earliest local prototypes of tiered reimbursement in the AI era. If global health reimbursement is heading over the next decade toward precision, individualization, and AI assistance, then a Taiwan-local design like “tiered catastrophic illness durations” deserves to be re-understood, re-interfaced, and re-presented at international policy dialogues.

This is also what the “Regenerative Medicine Technology” series is trying to do, one piece at a time.

Part 1 turned an ODT file into a searchable tool page. This piece translates that comparison table into a more understandable policy logic. More will follow, layer by layer — institutional, industrial, technical.

Glossary

  • ICD-10-CM/PCS: The diagnostic and procedure coding basis currently used by Taiwan NHI. ICD-10 is the international disease classification published by the WHO; ICD-10-CM is a clinical-modification version based on ICD-10, primarily maintained by U.S. NCHS/CMS; ICD-10-PCS is used for inpatient procedure classification, published by U.S. CMS. Taiwan NHI has carried out a version migration and local crosswalk based on the 2023 ICD-10-CM/PCS edition. See the Part 1 glossary for details.
  • Catastrophic Illness: A set of 30 disease categories designated under Article 2 Annex 1 of the Regulations Governing the Exemption of Insured Persons from Cost Sharing under the National Health Insurance. Once a catastrophic illness certificate is obtained, related treatment is exempted from partial co-payment.
  • Rare Disease: A disease designated under Article 3, Paragraph 1 of the Rare Disease Control and Orphan Drug Act. Rare diseases are automatically included as the 30th catastrophic illness category, with a permanent validity period.
  • ISS (Injury Severity Score): An internationally used trauma assessment metric. Among the six body regions (head/neck, face, thorax, abdomen, extremity, external soft tissue), the three regions with the highest AIS (Abbreviated Injury Scale) scores are each squared and summed. Higher scores indicate more severe trauma; a score of 16 or above is defined as “major trauma.” Taiwan NHI uses ISS ≥ 16 as the threshold for catastrophic illness category 12.
  • Value-Based Payment (VBP): A newer reimbursement model compared with traditional “fee-for-service.” Reimbursement levels are determined by treatment outcomes, patient prognosis, and improvement in health. U.S. Medicare has gradually adopted VBP since the early 2010s.
  • NICE (National Institute for Health and Care Excellence): The U.K.’s national health technology assessment body. It evaluates the cost-effectiveness of new drugs, therapies, and medical technologies, serving as a key reference for NHS reimbursement decisions.
  • Designated Intractable Disease Medical Beneficiary Certificate (特定醫療費(指定難病)受給者證): A medical-cost-assistance certificate issued by Japan’s MHLW for designated intractable diseases (conceptually similar to Taiwan’s rare diseases), with tiered reimbursement based on severity and income level, generally requiring annual renewal. The previous name “Specified Disease Medical Beneficiary Certificate” was in use before the Intractable Disease Act took effect in January 2015.
  • Tiered Reimbursement: An institutional design that sets different reimbursement levels by disease severity, outcomes, individual variation, and risk assessment. It is the next-stage model relative to traditional uniform reimbursement.

Disclaimer: This article is a compilation of public data and an institutional observation. It does not constitute medical, legal, or NHI application advice. The eligibility, validity period, and scope of catastrophic illness certification should follow the latest announcements of Taiwan’s National Health Insurance Administration, physician diagnoses, and the determinations of competent authorities.

Sources: