================================================================================ PEDIATRICSUICIDES.CA — THE FULL DATABASE Assembled and maintained by Tyler Black, MD, FRCPC © 2026 ASARI Holdings, Inc. All rights reserved. Terms of use: https://pediatricsuicides.ca/terms ================================================================================ VERSION 2026-09-11 curated-and-vetted.csv 91 jurisdictions, 2,754 rows censored.csv 48 jurisdictions, 801 rows years covered 1952 to 2025 suicide deaths, 0-19 480,426 across both files Files in this download: curated-and-vetted.csv every jurisdiction I can read at ages 0-19 censored.csv every jurisdiction I cannot, and the reason notes.txt this file -------------------------------------------------------------------------------- 1. WHAT THIS IS, AND WHY I BUILT IT -------------------------------------------------------------------------------- Pediatric suicide, despite its importance, is often reported badly... in a specific, repeatable way. One year's number, with no interval around it. Compared to one earlier year. Set against another country that counts something else entirely. Attributing causes to year over year changes. So I put it in one place, with the caveats attached to the numbers, and a proper way to interpret them. My goal? I would really like it if people looked at suicide numbers properly (per 100,000 per year whenever possible), understood child suicide dynamics properly (increasing rates from 10 to 24) and compared them properly (using previous established trends rather than economics-style "YEAR OVER YEAR" reporting). I am one person, and this took a long time. If you find an error, I want to hear about it. If you know a source for a country I am missing, I especially want to hear about it. dr.tylerblack@gmail.com -------------------------------------------------------------------------------- 2. HOW TO CITE IT -------------------------------------------------------------------------------- Please cite the database, and please also cite the national source for any country you rely on heavily. The `source` and `source_url` columns give you that country's own citation, and those offices did the actual counting. This database has a DOI. Use it -- it is the whole point of having one. Cite the database, any version: 10.5281/zenodo.22672942 Cite exactly this version: 10.5281/zenodo.22672943 The first always resolves to whatever the newest release is. The second pins the files you actually used, which is the one a methods section wants: I update this as countries publish, and a reader two years from now should be able to get back the bytes you had. Suggested form: Black T. pediatricsuicides.ca: an international database of pediatric suicide counts and person-time. Version . https://doi.org/10.5281/zenodo.22672942 The author identifier, if your reference manager wants one: Tyler R. Black, MD, FRCPC — ORCID 0000-0002-5681-4930 https://orcid.org/0000-0002-5681-4930 Replace with the `retrieved` date range of the rows you used, or the date you downloaded these files. I update the database as countries publish, so a version date matters. If you use a specific country's series, add its source. For example: ... Belize data from the WHO Mortality Database (ICD-10), Belize, retrieved 2026-09-04, via pediatricsuicides.ca. LICENCE: CC BY 4.0. USE IT COMMERCIALLY IF YOU WANT TO. These files are the Dataset as my Terms of Use define it (section 8.2): the death counts, the person-years, the jurisdiction, year, age-band and sex they are assigned to, the derived statistics, the case-definition and data-quality flags, the source registry, and this codebook. All of it is licensed under the Creative Commons Attribution 4.0 International licence, at creativecommons.org/licenses/by/4.0. You may copy, redistribute and adapt it for any purpose, including commercially. Give appropriate credit, link to the licence, and say if you changed anything. That licence is irrevocable and nothing else in my terms adds conditions to it. For the Dataset, please attribute in this form: Black, T.R. (2026). pediatricsuicides.ca pediatric suicide mortality dataset (Version ) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.22672942. Licensed under CC BY 4.0. A REQUEST, NOT A CONDITION. If you republish these numbers, please keep crisis resource information and the health disclaimers with them, or swap in local equivalents. I cannot require this and I am not trying to — the CC BY 4.0 grant above stands whatever you decide. I am asking because these are counts of dead children and somebody in trouble may be reading whatever you make. THE REST OF THE SITE IS NOT UNDER THIS LICENCE. The narrative text, the country commentary, the interpretive notes, the graphics and the design are Site Content, licensed CC BY-NC 4.0 — non-commercial only. If you want to quote the writing rather than use the data, that is section 8.3, and news and commercial media use is section 8.4. Full terms: https://pediatricsuicides.ca/terms THE UNDERLYING RECORDS ARE FACTS AND I CLAIM NO COPYRIGHT IN THEM. What is licensed above is my compilation, coding and presentation of them. The records themselves were reported by the national statistical offices, coronial services and international bodies named in the `source` column. If you go and get the data from one of those sources directly, that source's terms are yours to follow, not mine. -------------------------------------------------------------------------------- 3. THE FOUR RULES I DO NOT BREAK -------------------------------------------------------------------------------- When compiling every different jurisdictions numbers, I followed very important rules that minimize error. (1) COUNTS AND PERSON-TIME. NEVER RATES. There is no rate column in these files, and that is deliberate. A rate column is very alluring, but you can't really do much with it mathematically. However, you can always compute your own if you have the deaths and the person_years (whatever those totals are!): rate per 100,000 per year = deaths / person_years * 100000 Pool by summing, never by averaging: pooled rate = sum(deaths) / sum(person_years) * 100000 For an interval on a single rate I use Byar's approximation to the exact Poisson interval, which behaves at the small counts most of these countries have, where a normal approximation does not: lo = d(1 - 1/(9d) - 1.96/(3*sqrt(d)))^3 / E * 1e5 hi = (d+1)(1 - 1/(9(d+1)) + 1.96/(3*sqrt(d+1)))^3 / E * 1e5 For a rate ratio between a and b: RR = (d_a/E_a) / (d_b/E_b) 95% CI = RR * exp(+/- 1.96 * sqrt(1/d_a + 1/d_b)) For trend I fit Poisson regression of deaths on year with log(person_years) as offset. I report significance as "does the 95% interval include 1", never as a p-value. P VALUES ARE NOT PARTICULARLY HELPFUL FOR WHOLE POPULATION COMPARISONS. Differences are differences! the confidence interval is help you to decide whether or not a difference is important, not whether it occurred. (2) NEVER POOL DIFFERENT CASE DEFINITIONS. The `definition` column tells you what was counted. ICD-coded vital statistics, coronial suspected-suicide findings and police administrative counts are different measurements entirely. The most important distinction inside the ICD group is undetermined intent. Some countries count intentional self-harm alone (X60-X84); some add injury of undetermined intent (Y10-Y34). That is a difference in where an inquest sets its standard, not a different event ... but it is not small. In England, ages 15-19, deaths registered 2013-2024, undetermined intent is 19.4% of the total, and it is NOT constant: 29% in 2013, 13% in 2024. (3) NEVER INTERPOLATE AN AGE BAND. Every band column is filled ONLY where the bands that country published tile the target exactly — contiguous, no gap, no overlap, endpoints matching. That is aggregation, and it is fine. Anything else would be invention, and the cell is left EMPTY instead. The reason for this is pediatric suicide increases with age; interpolation is complicated and error-prone. (4) THIS IS NOT THE WORLD. Good vital registration covers roughly a third of global deaths, and the missing portion is where most child deaths occur. Anything you pool from this file is a summary of what has been counted, not a world rate. I say this on every page of the site and I mean it here too. -------------------------------------------------------------------------------- 4. WHAT IS IN CENSORED.CSV, AND WHY IT IS STILL HERE -------------------------------------------------------------------------------- I publish these rather than dropping them, because an absent country and a country with almost no recorded deaths are different facts and a reader deserves both. Read `censored_reason`: rate_only The source publishes a rate with no count behind it. It cannot be pooled. band_not_0_19 The published bands cannot be assembled into 0-19 without interpolating. See rule 3. all_zero No suicide deaths recorded at any age 0-19 across the whole series. A zero is a finding — but unless the denominator is very low, its an unreliable finding! too_few_deaths Too few deaths across the record for a statistic on them to mean anything. rate_below_threshold The rate sits far enough below the distribution that it almost certainly describes registration rather than children. I don't recommending merging censored.csv into curated-and-vetted.csv and pool the result. If you want them together for a coverage or completeness analysis, that is exactly what they are for — but they are not comparable series. -------------------------------------------------------------------------------- 5. THE COLUMNS -------------------------------------------------------------------------------- One row is one jurisdiction-year. IDENTITY iso3 ISO 3166-1 alpha-3. BLANK for historical states that no longer have one (Czechoslovakia, Yugoslavia, the former Federal Republic of Germany, Serbia and Montenegro). country Name as the source publishes it. slug My stable key. USE THIS AS THE PRIMARY KEY, not iso3 — it is the only identifier that is unique and always present. year Calendar year of death registration or occurrence, as that country reports it. COUNTS — for each band 0-4, 5-9, 10-14, 15-19, 20-24, 0-19, 10-19 deaths___m / _f / _all suicide deaths py___m / _f / _all person-years of exposure _all is male plus female. The source files carry only male and female rows; I have never seen a combined row I did not compute myself. 20-24 is included because most sources publish it and some of you will want it. It is NOT pediatric. Do not fold it into a 0-19 comparison. WHAT THE COUNTS ARE definition icd_suicide, undetermined_included, coronial_suspected, etc. CARRIED PER ROW, not per country, because a country can change basis part-way. New Zealand is ICD to 2008 and coronial from 2009; Northern Ireland is mixed too. If you are pooling, group by this column, not by country. icd_codes e.g. X60-X84. Empty where the basis is not ICD-coded at all, which is not the same as unknown. bands_published what that country actually published that year, before any aggregation of mine deaths_estimated TRUE where a year's value was filled by straight line between the years either side, because the source holds no submission for it. Drop these rows if you are fitting anything. provisional TRUE where the country calls the figure provisional and expects it to move year_complete FALSE for a partial trailing year. My charts stop at the last complete calendar year; a CSV cannot stop, so it says so instead. If you sort by year and read the last row, read this column first. any_suppressed TRUE where a contributing cell was suppressed for small numbers. Aggregates containing a suppressed cell are left EMPTY. HOW WELL IT WAS COUNTED who_grade WHO's own grade for that country's cause-of-death data: high, medium, low, very low, or blank where ungraded. registration_completeness_pct Share of deaths registered, where known. A country can look immaculate on a chart and be a quarter registered. These two columns are how you tell. THE TRAP I MOST WANT YOU TO AVOID pool_eligible FALSE where a series must not go into a pooled total. pool_note why. The important case: predecessor states. Czechoslovakia, Yugoslavia and the former Federal Republic of Germany overlap their successor states around 1985-1990. Pool the whole file without filtering on pool_eligible and you will count the same children twice. This is the most expensive mistake this file makes available, which is why it gets a column instead of a footnote. PROVENANCE source, source_url, retrieved That country's own citation, and when I last pulled it. notes Anything specific to that jurisdiction. -------------------------------------------------------------------------------- 6. HOW THE DATA WAS GATHERED -------------------------------------------------------------------------------- Most countries here come from the WHO Mortality Database — each country's own submission, ICD-coded, which is the closest thing to a common instrument that exists. Where a national statistics office publishes something better, longer or more current, I use that instead: the United States (CDC WONDER), Canada (Statistics Canada), Australia (ABS), the United Kingdom's three separate registries (ONS, National Records of Scotland, NISRA — they are not one series), Japan, China, Mexico and others. Denominators are UN World Population Prospects single-year-of-age estimates, mid-year, so one population vintage sits under every country built that way. Where a national office publishes its own denominator alongside its counts, I use theirs. WHAT "VETTED" MEANS IN THE FILENAME. It means the data has been through the pipeline I built to catch my own mistakes, not that a committee blessed it: - a schema validator on every file, which warns when a file mixes case definitions; - arithmetic checks on every build — a rate equals its deaths over its person-time, every interval contains the rate it belongs to, the sexes add to the combined figure, no year appears twice or goes missing; - a cross-check that the two independent implementations of my statistics (one in Python, one in JavaScript) produce identical output, character for character, across 248 comparisons; - seven reference figures checked against an independent implementation on every single build; - the age-band tiling rule enforced in code, so no cell can be filled by interpolation even by accident; - and me, reading each country's page one at a time, against its source. WHAT IT DOES NOT MEAN. I did not re-obtain the primary source documents or re-key them. Where a national office publishes a figure, this database carries that figure. This is not a re-analysis: I have not corrected, adjusted or modelled around any country's counting. I have shown what was published and said what I know about how it was counted. And being in the curated file does not mean a series is comparable with any other — it means it can be read at 0-19 and I have said what it is. -------------------------------------------------------------------------------- 7. IF YOU ARE WRITING ABOUT THIS -------------------------------------------------------------------------------- Please follow safe messaging guidance. reportingonsuicide.org is the standard and it applies to data stories as much as to any other kind. In short: do not frame a rise as inexplicable or as having a single cause, do not describe method or location, and carry crisis resources. If you are in crisis: in the US and Canada, call or text 988. Elsewhere, findahelpline.com will show you the services where you are. -------------------------------------------------------------------------------- 8. CORRECTIONS -------------------------------------------------------------------------------- Errors, omissions, and better sources for any country — especially the ones I have nothing for — to dr.tylerblack@gmail.com. The grey countries on the map are a list of things I still have to find, and I would rather be told than guess. -------------------------------------------------------------------------------- 9. VERSIONS -------------------------------------------------------------------------------- A version is a state of the data you can cite, not a release of software. Every version and what came with it is also at https://pediatricsuicides.ca/new 1.1 World Suicide Prevention Day (opened 2026-09-10) - India arrives, from the National Crime Records Bureau's police count at ages 0–17, 2020–2024 — the first police series on this site. - Kenya arrives at ages 0–14, from the two years of community death registration that can be read. - The Philippines reaches 2024, four years past the WHO submission, drawn as two runs because no year is covered by both sources. - Canada goes back to 1974 at ages 15–19, half of it recovered from rates Statistics Canada published to fifteen significant figures and never published counts for. - Nine European series extended from Eurostat, Norway by eight years and Germany by three, each tested against the years both sources cover. France failed that test and was not extended. - Forty-four years of phantom zeros removed across six countries, where the WHO file records a year a country never sent as a row of zeros. - An age range on every page: twelve ranges from 0–14 to 20–24, wherever a country's own file supports them, assembled from published bands and never interpolated. - A cookieless page counter that records no address, no user agent and no cookie, and honours Global Privacy Control and Do Not Track in the browser. - A review reads a page's own age range and the words around it, not every variant of the sentence under the chart, which the engine recomputes for each range, sex and window. 1.0 Launched (2026-09-09) - 113 countries and territories published, from 78 with a rate at ages 0-19 to 115 with a page of their own. - 405,756 suicides over 20.91 billion person-years of observation at ages 0-19. - Twelve age ranges, from 0-14 to 20-24, each assembled from published bands and never interpolated. - A world map with quartile bins, a tile grid of the same data, and a page for every jurisdiction. - 43 pages for countries whose published figures are not usable, each saying which of the two reasons applies and showing what was published. - Guernsey and Alderney as prose, because six to nine deaths in twenty years is a paragraph and not a rate. - China from urban and rural surveillance points, weighted by the 2020 census split, with the two lines shown so the weighting can be judged. - School days and youth suicide: US ages 5-17 by weekday, month and school term. - bands.csv and world.json: every jurisdiction at every range it can be read at, as one table. — Tyler Black, MD, FRCPC ================================================================================