Life Expectancy

Life Expectancy: Concept Architecture

Interpret an average remaining lifetime correctly

Life expectancy at age $x$ is the expected number of additional years lived by someone who has survived to age $x$, under a specified set of mortality rates. Life expectancy at birth is the same quantity evaluated at $x=0$. It is a population summary of a survival distribution, not the age at which everyone dies, not the median age at death, and not a promise of an individual lifespan. A person who has reached age 65 should use remaining life expectancy conditional on survival to 65, rather than subtracting 65 from an unrelated life expectancy at birth.

The phrase needs a population, calendar period or birth cohort, age, sex or other relevant stratification, mortality assumptions, and data source. National period estimates summarize current age-specific mortality in a hypothetical population; a forecast for a real birth cohort requires assumptions about how future mortality will change. Individual prognosis can differ substantially from a population average owing to conditions, risk factors, and treatment, and should not be silently inferred from a national table.

Connect survival to expected years

Let $S(a)$ be the probability of surviving from birth to exact age $a$ under the stated mortality schedule. For a person alive at age $x$, conditional survival to age $x+t$ is $S(x+t)/S(x)$ when $S(x)>0$. Remaining life expectancy is the area under that conditional survival curve:

$$e_x=\int_0^{\infty}\frac{S(x+t)}{S(x)}\,dt.$$

The integral adds survival probabilities at every future duration to obtain expected years. It requires a defined distribution and a sufficiently long age range. A median is instead the duration by which conditional survival falls to one half, if that point is reached; mean and median need not coincide, especially with a long survival tail. For a finite horizon $\tau$, the restricted mean remaining lifetime $e_x(\tau)=\int_0^{\tau}S(x+t)/S(x)\,dt$ counts expected years within the next $\tau$ years and is not a complete life expectancy unless the tail beyond $\tau$ contributes nothing.

In a discrete life table, $l_x$ denotes a hypothetical number alive at exact age $x$, $L_x$ the total person-years lived by that cohort between exact ages $x$ and $x+1$, and $T_x$ the person-years from age $x$ onward. The conventional summary is:

$$T_x=\sum_{a=x}^{\omega}L_a,\qquad e_x=\frac{T_x}{l_x}.$$

The final open age interval needs an appropriate mortality assumption, represented schematically by $\omega$ in this expression. Life-table $l_x$ is often based on an arbitrary radix, such as 100,000 hypothetical births; changing that radix scales $l_x$, $L_x$, and $T_x$ together but leaves $e_x$ unchanged. The rule used to estimate person-years within each age interval matters, particularly for infancy and the oldest ages.

A small restricted-life example

Suppose an illustrative life table has 1,000 survivors at exact age 65, then 900, 800, and 700 at exact ages 66, 67, and 68. Assume for this teaching example that survivorship changes linearly within each one-year interval, so each interval's person-years equal the average of the counts at its start and end. The numbers are invented, not a population estimate.

The 1,000 people alive at 65 contribute $950+850+750=2{,}550$ person-years during those three years. Their expected restricted remaining lifetime through age 68 is:

$$e_{65}(3)=\frac{2{,}550}{1{,}000}=2.55\ \mathrm{years}.$$

This is not total remaining life expectancy at 65: 700 people survive to 68, and their later years have not been included. Adding a credible continuation of mortality and person-years after age 68 would be necessary for a full $e_{65}$. The example also shows why one cannot simply average the three annual survivor counts and call the result years; each interval supplies person-years, then the total is divided by those alive at the starting age.

Distinguish period and cohort measures

A period life table applies the age-specific mortality rates observed or estimated for a given period to a hypothetical cohort passing through all ages. Its life expectancy is a snapshot of the mortality conditions of that period, not a literal prediction that a newborn will experience unchanged rates for an entire life. It is useful for comparing mortality conditions across places or time, subject to differences in data quality and population composition.

A cohort life table follows people born in a particular period. Mortality at ages they have already reached may be observed; mortality at future ages must be projected until the whole cohort's experience is known. Its estimated life expectancy therefore depends on future-improvement assumptions and can differ from a period value. Even retrospective cohort comparisons may be affected by migration, incomplete follow-up, and changes in cause coding or population denominators. State which type is being reported before comparing numbers.

Mortality shocks can sharply change a period estimate even though most currently living people's individual remaining lifetimes will depend on rates in many later years. A change in life expectancy from one period to another summarizes the entire age-specific mortality schedule, not the simple average of actual ages at death recorded in those calendar years. Mortality at young ages has a particularly large effect on life expectancy at birth because it removes many potential years from the hypothetical cohort.

Account for estimation and subgroup limits

Life tables use age-specific deaths and populations, with methods for converting mortality rates to probabilities of dying, estimating person-years within intervals, and closing the oldest age group. Small-area or subgroup estimates can be unstable when deaths are few. Birth registration, death registration, migration, age misreporting, and choice of smoothing or projection method can change results. An apparently precise figure should be accompanied by its period, population, methodology, and uncertainty where available.

Comparisons across groups can describe inequality in experienced or modeled mortality conditions, but an observed gap is not by itself a causal effect of any one characteristic. The group boundaries, exposure histories, and age-specific rates matter. Life expectancy is based on all-cause survival unless the estimand explicitly models a cause-deleted or other hypothetical scenario; simply subtracting deaths attributed to one cause without an appropriate competing-risk model is not a valid causal estimate of years gained.

Use life expectancy in economic evaluation

Expected life-years in a decision model are obtained by integrating each strategy's survival curve over the relevant horizon. The incremental life-years are the difference between those areas under consistently defined survival functions. For a population starting at age $x$, general-population life tables can inform background mortality, but disease-specific hazards, treatment effects, and competing mortality need coherent combination. Applying a period table as though it were an individualized forecast or ignoring future mortality changes can distort long-horizon results.

Life-years and quality-adjusted life-years (QALYs) are different. QALYs weight time in health states by preference-based utility and can reflect morbidity as well as survival; an extra life-year is not automatically one QALY. Healthy life expectancy is another population summary based on specified health-state information and is not interchangeable with either all-cause life expectancy or a decision-model QALY total. Costs, discounting, and patient experience also matter for cost-effectiveness.

When trial survival is censored, the observed area through the last adequately supported follow-up gives a restricted mean. Estimating full lifetime years requires extrapolation of survival beyond the evidence, with validated model structure and explicit tail assumptions. Show observed and extrapolated contributions separately and test how alternative plausible tail and background-mortality assumptions affect incremental results.

Report a usable figure

Specify age, sex or subgroup, location, calendar period or birth cohort, time horizon, all-cause mortality source, life-table method, whether rates are observed or projected, and uncertainty. Distinguish remaining years from expected age at death: under a specified conditional distribution, expected age at death for someone alive at age $x$ is $x+e_x$, but $e_x$ itself is years after age $x$. If reporting a restricted mean, state the restriction point prominently. These details let readers understand what the number estimates and where it can support a health economic decision.

Sources and further reading

• World Health Organization: Life expectancy indicator metadata. Official definition of remaining years under current age-specific death rates.
• UK Office for National Statistics: Period and cohort life expectancy explained. Official explanation of the two life-table interpretations.
• US National Center for Health Statistics: Life expectancy and life tables. Official overview of period life-table estimates at different ages.
• US National Center for Health Statistics: Method for constructing complete annual US life tables. Official calculation methods for life-table quantities.
• Han K, Jung I. Restricted Mean Survival Time for Survival Analysis: A Quick Guide for Clinical Researchers. Korean Journal of Radiology (2022). Methodological explanation of the finite-horizon area under the survival curve.