Good Years shows two figures: how long people who answered like you live on average, and how many of those years are typically spent in good health. This page describes every data source, every effect size and every modelling decision behind them, including the ones we are least sure of.
These are population estimates, not a prediction about you. They are averages across large groups of people, shown as ranges. Your own health and lifespan depend on many things no app can know, including genetics, medical history and chance. Good Years is for education and motivation only. It does not provide medical advice, diagnosis or treatment. For decisions about your health, talk to a doctor or qualified health professional.
Describes Good Years 1.1. Last updated 15 September 2026. Everything described here runs on your phone; nothing you answer is sent anywhere.
The baseline is the United Nations World Population Prospects 2024 complete life tables: mortality by single year of age, 0 to 100+, for 237 countries and territories, for men and women separately. The most recent observed year is 2023.
A published life table holds today's death rates fixed for the rest of a life. That is the standard summary of current conditions, but a 40-year-old will live through the death rates of the coming decades, not today's. Using the fixed table understated life expectancy by roughly 4–5 years at age 40 and 6 years at 30. So Good Years projects each country's death rates forward using improvement rates fitted to the UN's own 2024–2070 projections, separately for every country, sex and age. The fit reproduces the UN's projected figures to within about a tenth of a year.
This is a moderately optimistic choice, and we say so. The UN's medium scenario assumes mortality at midlife improves by roughly 1.4–2.1% a year. Across the 2010s the UK and US actually saw 0.5–1.1%. Assuming no improvement at all would not have been neutral either, just less visible. The uncertainty this creates is carried in the range (section 5).
The WHO publishes healthy life expectancy (HALE): the years a population can expect to live in full health, with any loss of health weighted by how severe it is. That includes serious illness, but most of it is common, milder problems like back pain, low mood, hearing loss and eyesight.
Good Years does not use WHO's absolute HALE figures, because WHO's life expectancy and the UN's differ by several years for some countries. Instead it takes the healthy share of life (HALE divided by WHO life expectancy, both from WHO's 2021 data) and applies that share to the UN baseline. For example, the share is 0.878 for Denmark and 0.852 for the United States.
WHO publishes that share at birth. The share of remaining life that is healthy is lower at 60, and falls further with age. Good Years estimates the age-60 share from the birth share with one constant per sex (checked against WHO's own age-60 figures across all 185 countries, where it is within about 0.2 healthy years on average). After 60, the healthy share of remaining life declines by 0.4 percentage points a year. That rate was fitted to WHO's figures and the life tables, not chosen, and it sits in the same direction as, but less steep than, US survey data on disability-free years.
WHO publishes healthy-life figures for 185 of the 237 countries and territories.
Long-term studies usually report a hazard ratio: how much higher or lower the death rate is, at any given age, for people in one group compared with a reference group. A hazard ratio of 1.30 means a 30% higher rate; 0.80 means 20% lower. Good Years multiplies the baseline death rate at every age from yours onwards by the hazard ratio for your answer, then recalculates life expectancy from the adjusted table. This is the standard way actuaries and epidemiologists turn a relative risk into years.
Your country's life table already reflects a population that includes smokers, non-smokers, heavy drinkers and people who never drink. Applying hazard ratios measured against the best group would put almost everyone below their country's figure. So every answer's hazard ratio is divided by the population average for that question, weighted by how common each answer is. Someone who answers like the average person lands on their country's published figure. The shares of people giving each answer come from national surveys (NHANES, NHIS, BRFSS, Eurostat's EHIS and others). They describe how adults in high-income countries answer these questions, which is the right comparison for a questionnaire.
People who smoke also tend to drink more; people who are inactive tend to weigh more. Multiplying each habit's full hazard ratio would count the same risk several times. Good Years adds up the (logarithmic) effects of your answers and then shrinks the total according to how many factors are really contributing. With one answer, nothing is shrunk, so a single factor reproduces its own study. With several, the combined effect is held near what Li et al. 2018 measured for five habits together. The strength of that shrinking was set by testing it against Li's figures, not by assumption.
This is why changes in the app can look smaller than the headlines you may have read. "Smokers lose about ten years" compares smokers with non-smokers who also differ in many other ways. Once you have told Good Years about those other ways, what it shows for quitting is the part that belongs to smoking alone.
Better answers are associated with more years of life and, in the research, with a larger share of those years in good health. Good Years moves your healthy share of remaining life up or down from your country's figure according to how your answers compare with the average, weighted by how much each factor affects mortality.
How far the share can move is the least certain number in the model, and we set it conservatively. The best-known study (Nyberg et al. 2020) counts years free of six major diagnosed conditions between ages 40 and 75, which makes the effect look larger than it would over a whole life. Measured to the end of life instead, Li et al. 2020 found a much smaller difference. HALE also counts every kind of ill health, much of which none of these questions touch. Taking both into account, the best and worst possible answers are allowed to differ by about 14 percentage points in healthy share.
Every life figure is shown as a range. It combines two kinds of uncertainty:
The range never starts below your current age.
Each hazard ratio is marked with where it comes from:
Hazard ratios are listed in the same order as the answers. Checked against the papers on 15 September 2026.
| Question and answers | Hazard ratio | Source and notes |
|---|---|---|
| Smoking Never · quit before 40 · quit in 40s · quit at 50+ · smokes now |
1.00 · 1.05 · 1.35 · 1.65 · 2.80 reference · derived · published (men) |
Jha et al. 2013, New England Journal of Medicine. 113,752 women and 88,496 men aged 25 and over in the US National Health Interview Survey. Current smokers had 2.8 times the death rate of never-smokers for men and 3.0 for women, and lost more than ten years of life expectancy. People who quit at 25–34, 35–44 or 45–54 gained about 10, 9 and 6 years compared with those who kept smoking. For people who had quit, the paper gives the death rate by the decade they quit in: 1.0 before 35, 1.2 at 35–44, 1.5 at 45–54 and 1.7 at 55–64. The app's answers cross those decades, so each is an average of the decades it covers; the paper has no figure for quitting after 64, so 1.7 is used rather than guessing higher. |
| Exercise Hardly any · under 1½ h/week · 1½–5 h · over 5 h |
1.00 · 0.80 · 0.69 · 0.65 reference · published · published · derived |
Arem et al. 2015, JAMA Internal Medicine. 661,137 adults in six cohorts, 116,686 deaths. Compared with no leisure-time activity: 0.80 below the recommended minimum, 0.69 at one to two times it, 0.63 at two to three times, 0.61 at three to five times, and 0.69 at ten times or more. "More than 5 hours" covers all of those higher bands, so the app uses 0.65. The study measures activity in MET-hours; the app's hours assume moderate intensity, such as brisk walking. |
| Alcohol (drinks per week) 0–7 · 8–14 · 15–25 · over 25 |
1.00 · 1.06 · 1.20 · 1.45 reference · derived |
Wood et al. 2018, The Lancet. 599,912 current drinkers in 83 studies; lowest risk at or below about 100 g of alcohol a week. Compared with up to 100 g, drinking 100–200 g, 200–350 g and more than 350 g a week was associated with about 6 months, 1–2 years and 4–5 years lower life expectancy at 40. The paper shows its hazard ratio for each band only as a chart, so ours can't be matched to a printed figure. The study included only current drinkers, so it cannot show whether drinking nothing is better than drinking a little. We don't claim either. |
| Weight (BMI, from height and weight) Under 18.5 · 18.5–25 · 25–30 · 30–35 · 35–40 · 40+ |
1.51 · 1.00 · 1.13 · 1.45 · 1.94 · 2.76 published · derived · derived · published · published · published |
Global BMI Mortality Collaboration 2016, The Lancet. 10.6 million adults in 239 studies; main analysis in 3,951,455 never-smokers without known chronic disease. The paper uses finer bands than the app asks about: 18.5–20 is 1.13, 20–25 is 1.00, 25–27.5 is 1.07 and 27.5–30 is 1.20. The app merges these into 18.5–25 at 1.00 and 25–30 at 1.13, so the lower end of the "healthy" band carries somewhat more risk in the study than the app shows. The effect weakens with age (section 3). Being underweight is mostly a sign of existing illness rather than a cause, so the app never offers weight gain as a change and suggests talking to a doctor instead. Risk at a given BMI starts rising at lower values in Asian populations; the app does not model this. |
| Sleep Under 5 h · 5–6 · 6–7 · 7–8 · 8–9 · over 9 |
1.10 · 1.04 · 1.01 · 1.00 · 1.04 · 1.08 derived · published · published · reference · published · our choice |
Yin et al. 2017, Journal of the American Heart Association. 3,582,016 participants, 241,107 deaths, lowest risk near 7 hours. The paper gives the curve at whole hours: 1.12 at 3 hours, 1.08 at 4, 1.04 at 5, 1.01 at 6, 1.00 at 7, 1.04 at 8, 1.15 at 9 and 1.32 at 10. The app uses the value at 5, 6 and 8 hours for the bands starting there, and 1.10 for under 5. For over 9 hours it uses 1.08 instead of 1.15–1.32, because long sleep is largely a sign of illness rather than a cause: using the full figure would have told long sleepers to sleep less. Above 9 hours the app suggests talking to a doctor. |
| Blood pressure (in the person's own words, no numbers) Normal · high but under control · high and not under control · not checked in a couple of years |
1.00 · 1.15 · 1.50 · 1.20 reference · published (1.16, rounded) · derived · our choice |
Zhou et al. 2018, Scientific Reports. 13,947 US adults, median follow-up 19.1 years. Compared with normal blood pressure: treated and controlled 1.16 (95% CI 0.95–1.42), untreated 1.40, treated but not controlled 1.62. The app's single "not under control" answer covers both of the last two, so it uses 1.50. "Not checked" is our estimate for a group that mixes people with normal and undetected high blood pressure, and the app suggests getting it checked rather than offering a number. The SPRINT trial (2015, 2021, NEJM) is the only randomised evidence on death from any cause behind any factor here. |
| Diet (self-rated) Mostly whole foods · fairly balanced · a mix · often processed or sugary · mostly processed and sugary |
0.887 · 0.953 · 1.00 · 1.075 · 1.162 our choice, from sources |
Built from Sofi et al. 2008 (BMJ), Seidelmann et al. 2018 (Lancet Public Health) and studies of specific foods. Scored as +2.0 to −2.5 years at age 50, a range we chose to keep diet inside the whole-lifestyle effect in Li et al. 2018. It is not a figure from any one paper. Reduced to 60% for being one self-rated question (section 3). The shares of each answer are how US adults rate their own diet (Thomson et al. 2023, NHANES 2015–2018, n=9,757). |
These are asked later, only if you choose. They describe your situation rather than a habit, so the app never offers changing them as something to do. They make the estimate more accurate, and some answers come with a suggestion to talk to a doctor.
| Question and answers | Hazard ratio | Source and notes |
|---|---|---|
| Diabetes No · told close to it · diagnosed 55+ · diagnosed 40s–early 50s · diagnosed before 40 |
1.00 · 1.13 · 1.54 · 2.00 · 2.64 reference · published · derived · derived · published |
Emerging Risk Factors Collaboration 2023, Lancet Diabetes & Endocrinology. 1,515,718 people in 96 cohorts and UK Biobank, 23.1 million person-years. How early the diagnosis came matters more than the diagnosis alone, so the answers are split by age at diagnosis. Compared with no diabetes, and adjusted for age, sex, smoking, BMI, blood pressure and cholesterol, the paper gives 2.64 for a diagnosis in the 30s, 2.24 in the 40s, 1.79 in the 50s, 1.55 in the 60s and 1.41 at 70 or later. The app's answers span those decades, so "40s or early 50s" and "55 or later" combine them, weighted by how many people in the study were diagnosed at each age. "Told I'm close to it" is from Cai et al. 2020, BMJ: 1.13 (1.10–1.17) across 129 studies and 10 million people, where "prediabetes" was defined by a blood test. Good Years 1.1 used lower figures (1.15, 1.50, 1.80 and 2.20) that placed each band one decade too young; see Corrections. |
| Hearing (with a hearing aid, if used) No difficulty · a little · quite a lot · a great deal |
1.00 · 1.10 · 1.28 · 1.45 reference · our choice · published · our choice |
Jia et al. 2025, PLoS One. Meta-analysis of 36 cohort studies, 6,364,914 participants. Hearing loss overall: 1.21 (1.13–1.31); hearing loss measured by a hearing test: 1.28 (1.10–1.49), and 1.13 (1.06–1.20) when people reported it themselves. The app asks about day-to-day difficulty, so we checked it against that lower figure: counting how many people give each answer, the app's three "difficulty" answers average 1.12 against "no difficulty", which matches. The measured 1.28 is used for "quite a lot", an answer about 1 in 70 adults give; the other two bands are spaced around it, since the study has no severity levels. We chose it over a single-cohort figure of 1.40 (Choi et al. 2024, 9,885 US adults) because it pools far more people. |
| Eyesight (with glasses or contacts) Good · some difficulty · a lot · severe |
1.00 · 1.29 · 1.43 · 1.89 reference · published · published · published |
Ehrlich et al. 2021, Lancet Global Health. A systematic review of 28 studies (446,088 participants); the hazard ratios come from its meta-analysis of 17 studies (47,998 participants). They are for eyesight measured on a chart: 1.29 for worse than 6/12, 1.43 for worse than 6/18, and 1.89 for worse than 6/60, the last from a single study. The app asks about day-to-day difficulty instead, so we checked whether self-reported eyesight shows a weaker link. It does not: in an earlier meta-analysis quoted by Ehrlich (Zhang et al. 2016), the worst self-reported eyesight came to 1.44 against 1.36 for the worst measured on a chart. The closest match to the app's question, the Norwegian HUNT study (Brunes et al. 2017, 65,236 adults), found 1.22 for people aged 60–84 who said their eyesight was impaired. |
Every change to the model runs against a set of automated checks. The main ones hold it to studies that measured the result directly:
When a proposed change to an effect size breaks one of these, the change is treated as too large, not the check as too strict.
It would be simple, and appealing, to say that healthier habits shrink the years lived in poor health. The evidence doesn't support it. Living longer tends to add years at both ends. A review of 20 studies covering 26 cohorts (Dieteren et al. 2021) found that about half showed more years in poor health for people who never smoked than for smokers, while nearly all found a smaller share of life in poor health. Li et al. 2020 found the time in poor health got shorter for women and longer for men in the same study. In Good Years' own model, the gap between the two figures gets slightly wider for about half of 1,200 changes we tested.
What held for every change we tested is that a better answer is associated with more healthy years and a larger share of life in good health. That is what the app says, and it does not promise to close the gap.
Being underweight, sleeping more than nine hours, and not having had your blood pressure checked all carry a real statistical association. In each case the honest next step is to talk to a doctor, not to chase a number, so the app counts them in the estimate but never offers them as a change. The same goes for more serious hearing and eyesight difficulty.
An early version treated blood pressure under control as equivalent to normal, because the study's result for that group wasn't statistically significant. But not being able to show a difference is not showing there is none, and both studies we found put the figure above 1. It is now 1.15. Getting blood pressure under control is still associated with one of the largest differences the app shows.
Many things are associated with longer life in the research. A factor gets a question only if a large study supports it, the question can be answered honestly without a device or a test, and it doesn't just repeat something already asked. Some that didn't make it:
We check citations against the papers themselves. These are the changes that checking has produced. If you find another, please tell us.
| Date | What changed |
|---|---|
| 2026-09-15 | Quit smoking in the 40s: 1.30 → 1.35; at 50 or later: 1.70 → 1.65. The paper gives figures by the decade people quit in, and these two had not been averaged across the decades each answer covers. Corrected in the version after 1.1. |
| 2026-09-15 | Diabetes: 2.20, 1.80 and 1.50 → 2.64, 2.00 and 1.54 for a diagnosis before 40, in the 40s or early 50s, and at 55 or later. Checking this page against the paper showed each band had been given the figure for the decade after it, which understated the association. "Told I'm close to it": 1.15 → 1.13, now from a named source. Corrected in the version after 1.1. |
| 2026-08-28 | Blood pressure under control: 1.05 → 1.15. The earlier figure treated a non-significant result as equivalence (section 8). |
| 2026-08-28 | A blood-pressure study was cited with the wrong year and with figures from its secondary analysis. The paper is Foti et al. 2025, Hypertension 82:1095–1105; its primary analysis gives +0.9 years of life and +2.7 healthy years for lifelong prevention. The model did not depend on the wrong figures. |
| 2026-08-28 | A tooth-loss study was attributed to the wrong journal. It is Peng et al. 2019, Bioscience Reports. (The factor was later removed.) |
| 2026-08-28 | Hearing: 1.40 → 1.28, replacing a single-cohort figure with a meta-analysis of 6.4 million people. |
| 2026-08-28 | Long sleep: 1.15 and 1.30 → 1.04 and 1.08, because long sleep is mostly a sign of illness rather than a cause. Underweight and over-nine-hours sleep are no longer offered as changes. |
| 2026-08-28 | Fixed life tables replaced with projected ones, which raised estimates by about 4–5 years at 40 and made the range wider. |
| 2026-08-28 | How far healthy share can move was halved, after finding that the original target came from a study counting only ages 40–75. |
| 2026-08-30 | Every answer's population share replaced with figures from named national surveys; several earlier values had been estimates. |
| 2026-08-31 | BMI 40 and above restored as its own band (2.76). It had been grouped with 35–40 (1.94), understating the effect for the group it matters most to. |
| 2026-08-31 | Diet: answer shares taken from how people actually rate their diet (they were assumed equal), and the effect reduced to 60% for being one self-rated question. |
Life tables are © United Nations, World Population Prospects 2024, licensed under CC BY 3.0 IGO.
Healthy-life proportions are derived by Good Years from World Health Organization data (data.who.int, healthy life expectancy and life expectancy indicators), licensed under CC BY 4.0. The proportions are calculated by Good Years and are not WHO-published figures. WHO does not endorse this app.
Titles, volumes, pages and DOIs taken from each paper's PubMed record on 15 September 2026.
Good Years shows statistical estimates based on population data — averages across large groups of people, not a prediction about you. Your actual health and lifespan depend on many factors no app can know, including genetics, medical history, and chance. This app is for education and motivation only; it does not provide medical advice, diagnosis, or treatment. For decisions about your health, talk to a doctor or qualified health professional.