a long oneWhat the data cannot see

Thirteen published series supply every chart, map and table on this site. Each has a record, and each record carries one sentence stating what the series does not count. This page collects those thirteen sentences. Not one of the series counts a person who crossed a border to be treated, and the pattern in how they miss it is more useful than the omission itself.

Begin with the money. The World Bank's current health expenditure per person covers government schemes, compulsory and voluntary insurance and household out-of-pocket payments, divided by population — and excludes capital spending, so a new hospital building does not appear, and excludes anything a foreign patient pays for treatment in that country. In the record's own words, a country can run a large export trade in surgery and show nothing for it there. The out-of-pocket share does the same from the other side: it is a ratio rather than an amount, it does not separate a co-payment inside an insured system from a whole bill paid in cash, and it counts nothing a visiting patient pays, which belongs to the export half of the account. The World Health Organization's Global Health Observatory answered with 3,099 indicator records when its record was written, and no indicator in it counts a foreign patient, a treatment bought abroad, or a payment made in another country's health system.

The arrivals series is the one readers assume must hold the answer, and it is the furthest from it. The World Bank republishes UN Tourism's international arrivals for 223 country codes. Each trip by the same person counts again, so the total is crossings rather than travellers. Nothing in the series records why anybody came: there is no treatment field and no purpose field, so a patient cannot be separated from a visiting relative. And it counts nobody who crossed by land where a country reports air arrivals only — which removes most of the Mexican border dental trade, one of the largest flows in this directory.

The two clinical registers that would settle the fertility half of the subject are defined to exclude it. The Human Fertilisation and Embryology Authority has held every IVF and donor insemination cycle performed in a UK licensed clinic since 1 August 1991, 107 clinics in 2023/24. A UK resident who has IVF in Spain, Czechia or northern Cyprus leaves no trace in it, because the clinic is not licensed by the Authority: the one flow this directory exists to describe is the one the register is defined to exclude. The American series has the same shape and states more of its own limits. The CDC counted 435,426 ART cycles on 251,542 unique patients at 457 reporting clinics in 2022; it counts nothing done outside a United States clinic, no intrauterine insemination and no ovulation induction because neither is ART by its definition, nothing from the clinics listed as nonreporters — about 2 per cent of cycles on its own estimate — and its clinic rates describe cycles more than a year old, which the agency says plainly should not be read as any individual's chance of success.

The capacity series describe a health system rather than a market. Physicians per 1,000 people cannot separate the capital from the province, the private hospital from the district clinic, or the surgeon who operates on visitors from the physician who never sees one; it counts no nurse, records no specialty, and holds no measure of what any of them do. Hospital beds per 1,000 cannot say whether a bed is staffed, equipped or empty, counts no operating theatre, no intensive-care capability, no dialysis station and no day-case chair, and draws no line between the public system a resident queues for and the private wing a foreign patient is admitted to. Forty-seven of the 220 country outlines this site draws carry no bed value at all — Nigeria, the Democratic Republic of the Congo, Côte d'Ivoire, Namibia and South Sudan among them — and absence there is absence of a return rather than absence of beds.

The series closest to the motive is the one that removes the most. OECD waiting times cover seven elective procedures in the countries that publish them: 6,609 observations from 24 countries, 2000 to 2024. The clock starts after the specialist assessment, so the months between a family doctor's referral and a consultant's opinion sit outside the series by definition. It counts no privately funded patient, which removes the domestic private route competing directly with travelling abroad. It counts no country that does not publish — Austria, Belgium, Czechia, France, Germany, Japan, Korea, Latvia, Luxembourg, Mexico, the Slovak Republic, Switzerland, Türkiye and the United States are all marked data not available, and several of those are among the largest sources of patients travelling for planned surgery. And it counts nobody who left the list: died, gave up, went private, or flew somewhere else. Inference — the last exclusion is the whole subject of this directory, removed from the only series that measures the reason for it.

Two of the thirteen are registers of attention rather than of activity. The Joint Commission International directory counts organisations that applied, paid for the survey, met the standards on the survey date and are inside a three-year cycle: no infection rate, no mortality, no revision rate, no price, no patient volume, no complaint, nobody in it involuntarily, and no public history of the lapsed. The OpenStreetMap harvest holds every named hospital, clinic and dental surgery OpenStreetMap carried within a set radius of each hub on the date it was fetched, and what it does not hold is anything nobody mapped — absence there is absence of mapping, not of hospitals. It carries no accreditation, no ownership, no price, no quality measure, no patient volume and no indication that a place treats foreigners; a clinic that closed last year sits in the file until somebody edits it out; and a row proves only that a point on a map once had that name typed on it.

The last two are the infrastructure, and they fail quietly. Natural Earth supplies 220 country outlines that join to everything else here through one country code, so a place with no ISO alpha-2 code, or one too small to carry a polygon at this scale, drops silently out of every map and every chart: Northern Cyprus, Kosovo and Somaliland sit here under codes this project invented for them, and Monaco, Macau and the Vatican are not present at all. The exchange-rate table holds 166 currencies against the dollar on one stated date and carries no card spread, no foreign-transaction fee, no cash-advance charge and no airport-counter margin; it holds no history, so a price published in 2019 and converted here is converted at a 2026 rate; and several of its rates are administered pegs rather than traded prices — the dirham at 3.6725, the riyal at 3.75 — with nothing in the file marking which is which.

Inference — the omission is structural rather than careless. Every one of these thirteen was built by an institution measuring its own population, its own licensees or its own members, and a patient who crossed a border is by construction somebody else's. What is left to count this trade with is made by the parties holding an interest in the answer: a hospital counting its own patients, a trade council counting member returns, a ministry counting registrations, a hospital association counting what it billed. This site therefore carries those figures with their unit and their collector attached, prints a null as a gap rather than a zero, and publishes a coverage layer on its own map so the distance between what the world reports and what this directory has written up stays visible.

The sentence lives in one field. Every dataset record carries dataset.counts, in the publisher's own definition, and dataset.does_not_count beside it; the validator warns when either is empty, because that sentence is the point of the record. Licences travel with the data: CC BY 4.0 on the World Bank series, ODbL 1.0 on the OpenStreetMap harvest, public domain on Natural Earth, WHO terms of use on the Global Health Observatory, and an attribution-required open endpoint for the rates. Where a publisher refused an automated request, the record says so and states the gap rather than filling it.

As of 17 September 2026 this site holds thirteen dataset records: five World Bank series, the WHO repository, the OECD waiting-times extract, the UK and US fertility registers, the JCI directory, the OpenStreetMap harvest, the Natural Earth outlines and one dated exchange-rate table. None of them counts a cross-border patient. The site's own coverage figures are published at /coverage/ and /numbers/, and the gap between the two is drawn on the map.