How Odainn calculates your risk of dying

All data comes from official statistics and covers 37 countries: Austria, Belarus, Belgium, Bulgaria, Canada, Croatia, Cyprus, Czechia, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Moldova, Netherlands, Norway, Poland, Portugal, Romania, Serbia, Slovakia, Slovenia, Spain, Sweden, Switzerland, Türkiye, United Kingdom, United States. For every country except the United Kingdom, the United States, Canada and Belarus the source is Eurostat, the statistical office of the European Union:

• Causes of death (dataset hlth_cd_aro): deaths of residents by cause (ICD-10), sex and 5-year age group.
• Deaths (demo_magec): all deaths by single year of age and sex — the all-cause total.
• Population (demo_pjan): residents on 1 January by single year of age and sex.
• Life tables (demo_mlifetable): official life expectancy, used to validate our engine.

The United Kingdom no longer reports to Eurostat, so its numbers come from its own statistics offices, with the same method and the same checks: deaths by underlying cause (ICD-10), sex and age from the Office for National Statistics (England and Wales), National Records of Scotland and the Northern Ireland Statistics and Research Agency; ONS mid-year population estimates; and ONS single-year life tables for validation.

For the United States we use every death certificate of 2023 and 2024 (the National Center for Health Statistics' public-use files: underlying cause, sex and age), Census Bureau population estimates, and the official US life tables for validation. Above 65 the official tables rely on Medicare records, because census counts of the oldest Americans are too high — so from 65 we take the level of risk from the official tables and the split between causes from the death certificates.

For Canada we use Statistics Canada's deaths by detailed cause, sex and age for 2023 and 2024, its July population estimates, and its official single-year life tables for validation. Habits come from the Canadian Community Health Survey (share of smokers, weight and activity levels).

For Belarus the source is Belstat, the National Statistical Committee of the Republic of Belarus: its Demographic Yearbook 2019, with 2018 data — deaths by sex and age, deaths by cause, population and life expectancy. Belstat publishes deaths by age and deaths by cause but not the two together, so the split by cause and age is our estimate: estimated countries.

We use 48 causes of death built from ICD-10 codes, following Eurostat's European shortlist so that every country is grouped the same way and can be compared. A few names to know: accidental poisoning (ICD-10 X40–X49) is mostly drug overdoses; suicide is intentional self-harm (X60–X84). National lists can differ — the UK's ONS counts deaths of undetermined intent with suicide, and US and Canadian overdose statistics include every intent — so our numbers won't always match a country's own “leading causes” table.

We use the most recent post-pandemic years (2023, and 2024 where published). Pandemic years 2020–2022 are excluded because excess deaths would distort the picture of normal risk. Small countries (Cyprus, Estonia, Iceland, Latvia, Luxembourg, Malta and Slovenia) also include 2018 and 2019 so rates are not dominated by chance. Belarus uses 2018, the latest year Belstat has published in full.

For each sex and 5-year age group we divide deaths by person-years lived to get a death rate per cause. Stepping forward one year at a time, we combine all causes with standard life-table and competing-risks methods: the chance of dying from a cause in a year is its share of the total hazard times the chance of dying at all. This gives the 12-month and 10-year probabilities, lifetime probabilities, life expectancy and the chance of reaching a given age.

Lifetime probabilities — the chance of each cause being the one someone dies of, from their age today — use today's death rates at every age. Those rates have been falling for decades, so lifetime figures describe today's world rather than predict anyone's; the 10-year figures depend far less on that assumption. At 37 the two look very different: the next 10 years are mostly injuries, a lifetime is mostly heart disease and cancer — which is why the cause list can switch between them.

Ill-defined causes. Some deaths are recorded without a specific cause (for example about 10% in Poland). Following standard WHO practice, we redistribute them across the specific causes in proportion, within each age and sex group.

Complete totals. Cause-of-death statistics can miss some deaths — for example residents who die abroad (about 5% in Luxembourg). So we take the all-cause total from the official death statistics and use causes only for the split.

Small numbers kept private. Cyprus, Luxembourg, Malta and Iceland (under 25), and Estonia and Slovenia (under 15), keep some causes of death for a single sex and five-year age group confidential, because the numbers are small; Eurostat publishes them for wider age groups or for both sexes together. For those groups we take the split by cause from the closest published group — for children, both sexes at the same age; for 15–24-year-olds, the same sex across 15–24. The number of deaths still comes from the official totals.

Some countries publish deaths by age and deaths by cause, but never the two together. Rather than leave their people out, we estimate the missing piece — and say so wherever their numbers appear. Belarus is the first: Belstat's last full edition is the Demographic Yearbook 2019, with 2018 data.

• Official: deaths by sex and age to 69, births and infant deaths, population, life expectancy, and deaths by cause group and sex — including suicides, homicides, road deaths, falls, drownings, fires and alcohol poisonings — with the same groups at working age.
• Estimated: how each cause group spreads over ages, fitted (iterative proportional fitting) so every official total above holds exactly; the detail inside groups (cancer sites, heart attacks vs strokes, COPD vs pneumonia), in the proportions of Lithuania and Latvia for the same sex and age; and ages 70+, where the official tables stop — Lithuania and Latvia's age pattern, scaled to Belarus's official life expectancy at 70.

The result reproduces Belstat's life expectancy at birth, 35 and 65 within 0.1 years. Estimated countries are left out of every cause ranking and Europe-wide comparison, so no comparison rests on an estimate; they do appear in life-expectancy rankings, which use only official all-cause data.

A country is published only if our engine, using nothing but these counts, reproduces that country's official life expectancy at birth, at 35 and at 65 within half a year, for both sexes. All 37 countries pass (Georgia didn't, so it isn't published). These checks run automatically as tests every time the data is rebuilt — here are the results:

Life expectancy at birth in years. “Largest gap” is the biggest difference at birth, 35 or 65, for either sex, in years; the limit is 0.5 and the largest anywhere is 0.21.

Members can answer seven questions about smoking, exercise, height and weight, alcohol, and fruit and vegetables. For each habit we take relative risks by cause of death from large published cohort studies — for example, current smokers die of lung cancer about 25 times as often as people who never smoked (Carter 2015, NEJM).

Relative risks can't simply be multiplied onto national rates, because those rates already include smokers and non-smokers, active and inactive people. So we use national health surveys — the European Health Interview Survey (Eurostat EHIS 2019; 2014 for the UK, its latest wave; the EU-27 average for Switzerland, Moldova and Belarus, which it doesn't cover), the CDC's National Health Interview Survey 2024 for the US and the Canadian Community Health Survey for Canada (with European splits where it reports less detail) — to see how common each habit is for your country, sex and age, and rescale: the average over everyone reproduces the official rate exactly. Automated tests enforce this for every country, cause, sex and age.

• Smoking: Carter et al. 2015 and Thun et al. 2013 (NEJM), including how the extra risk fades in the years after quitting — scaled by cigarettes a day: in proportion for lung cancer and most causes (Pirie et al. 2013, Lancet), and much less for heart disease and stroke, where even one cigarette a day carries a large share of the risk (Hackshaw et al. 2018, BMJ).
• Weight (BMI): Prospective Studies Collaboration 2009 and Global BMI Mortality Collaboration 2016 (Lancet).
• Exercise: Arem et al. 2015 (JAMA Internal Medicine).
• Alcohol: Wood et al. 2018 (Lancet), Bagnardi et al. 2015 (British Journal of Cancer), Roerecke et al. 2019 (American Journal of Gastroenterology) and Zhao et al. 2023 (JAMA Network Open).
• Fruit and vegetables: Aune et al. 2017 (International Journal of Epidemiology) — per extra 200 g a day, 8% lower risk of coronary heart disease, 16% of stroke and 3% of cancer, but never more than the study's own estimate for the highest intakes (24% and 33% lower at 800 g a day; no further cancer benefit above 600 g). We don't apply its larger all-cause estimate. How many people eat 5 portions a day comes from EHIS 2019 (Canada: its own survey; the US: the EU-27 average, because the US survey doesn't ask).

We are deliberately conservative: where a study covers only part of one of our causes we use the lower estimate, and we don't model associations that look like confounding rather than cause and effect. As a result our estimates of what smoking and heavy drinking cost are, if anything, on the low side — for example, US studies find smokers lose more than 10 years of life; we typically show 6–9 for an average smoker (15 a day) and 9–11 for 20–29 a day.

“Expected days of life gained” for an action is the change in your life expectancy when we recompute everything with that one habit changed. It is an average over many people like you — a statistic, not a promise.

When you change an answer, the new habit counts in full straight away — as if you had always lived that way and keep it up. The studies compare people who kept a habit for years, so the real effect builds gradually: faster for injuries and liver disease, slower for cancers. Smoking is the exception: there we count the years since quitting, and the extra risk fades over decades. Members see a dated timeline of their changes.

Inside a whole-person estimate we sometimes have to use a broader estimate — for example a study's “other causes” category, or an effect measured for all heart disease — where no study measured one of our causes on its own. On our public cause pages we only state a habit's effect on a cause when a study measured that exact cause.

Two optional questions — how you get around, and activities such as swimming in the sea or hiking in the mountains — don't change the risk numbers. No official source says how many people drive, cycle, swim or climb in each country, by sex and age, so we can't rescale them the way we do for habits.

For road accidents we show who dies on the roads at your age and sex in your country — drivers and riders, passengers and pedestrians — from Eurostat's road accident statistics (2022–2024 together; where a group had fewer than 30 deaths, the all-age split for your sex). Everyone walks and rides as a passenger, so if you don't drive, ride a motorbike or cycle, the pedestrian and passenger share of the average is your part: for a 37-year-old man in Poland, about a quarter. If you do, we keep the average: deaths per driver would need how many people drive and how far, which isn't published by country and age — and your risk grows with the time you spend on the road. This covers the 31 European countries in Eurostat's road statistics; the United Kingdom, the United States, Canada, Türkiye, Moldova and Belarus aren't included yet.

For activities we show what helps, the effects an official body or study measured (for example, alcohol is involved in up to 70% of deaths during water recreation — CDC), and who to call: the emergency number, and in Poland the mountain rescue (GOPR, TOPR) and water rescue (WOPR) numbers, checked on their official pages.

The check-ups members are reminded about follow each country's official programmes: USPSTF and the CDC in the United States; the Canadian Task Force and NACI in Canada; NHS programmes in the United Kingdom; NFZ programmes in Poland; statutory screening (G-BA) and STIKO in Germany; the Assurance maladie programmes in France; RIVM screening in the Netherlands; the HSE in Ireland; the national programmes in Sweden, Denmark, Finland and Norway, Austria, Czechia, Portugal (DGS), Lithuania, Latvia, Estonia, Luxembourg, Hungary, Slovakia, Slovenia (Svit, DORA, ZORA) and Croatia; Belgium's regional programmes; Switzerland's federal insurance rules (cantonal programmes); the national ranges in Spain and Italy (where regions vary); Belarus's dispensary check (Ministry of Health, 2024); Greece's national programme (Προλαμβάνω); Serbia's national screening programmes; Malta's National Screening Centre; Cyprus's Ministry of Health programmes; Iceland's national invitations; Türkiye's national programmes (Ministry of Health); Bulgaria's preventive check-up (NZOK); and the cervical screening programmes of Romania and Moldova. Where a programme leaves a gap, we use the EU Council Recommendation on cancer screening (2022) and European guidelines. Each list is checked against the official sources and dated in our evidence notes.

Members can tell us about close family (a parent, brother, sister or child) with bowel, breast, ovarian or prostate cancer, or a heart attack or stroke at a young age. That brings checks forward as the guidelines describe — for example a colonoscopy from 40 (US Multi-Society Task Force 2017), a PSA conversation from 45 (European Association of Urology 2026), a family-risk check for BRCA (USPSTF 2019) and a heart-risk check at any adult age (ESC 2021) — always as “ask your doctor”, because national rules differ. It doesn't change the risk numbers.

The cancer screening programmes side by side, country by country: breast, bowel, cervical, lung and prostate.

Members can ask an AI guide to explain their results, answer questions or prepare a doctor's visit. It only explains: it receives the engine's numbers and may restate them, and an automatic check removes any sentence with a number that isn't in the engine's output. It never sees names or emails and doesn't give medication advice. If a message suggests someone is in crisis, they get a human-written reply with the help lines for their country instead of a model's answer.

They describe people like you: the same age, sex and country — and, if you answered the health questions, the same habits. They are not a diagnosis or a personal prediction: your real risk also depends on your genes, medical history and much else. Odainn is educational and does not give medical advice.