Age standardization is a statistical method for comparing rates or proportions between populations with different age distributions. It replaces, or otherwise accounts for, the populations’ actual age compositions using a reference distribution or a reference set of age-specific rates. Widely used in demography, epidemiology, and public health, it helps distinguish differences in measured outcomes from differences attributable to population age structure. An age-standardized rate is a comparative summary measure, not necessarily the rate actually observed in any population.
Purpose and statistical basis
Many outcomes vary strongly with age. For example, a population with a larger proportion of older people may have a higher crude death rate even when its death rate within every age group is lower than that of a younger population. Comparisons based only on crude rates can therefore be misleading.
Suppose a population is divided into mutually exclusive, collectively exhaustive age groups. Let be the number of events in group , and let be its denominator, such as person-time at risk. Its age-specific rate is
The crude rate is
Thus, the crude rate depends both on the age-specific rates and on the relative sizes of the age-group denominators. Standardization separates these components by replacing the population-specific weights with common reference weights.
For mortality and incidence, denominators commonly represent person-time or population counts used to approximate person-time. For prevalence, the denominator is the number of people assessed, and the age-specific measures are proportions rather than event rates. The same weighting principle applies, but the resulting quantities retain these different interpretations.
Age standardization addresses age-related confounding in descriptive comparisons. It does not automatically account for other population differences or establish causal conclusions.
Direct standardization
In direct age standardization, each population’s age-specific rates are applied to the same standard age distribution.
Let denote the standard population in age group , with weights
The directly standardized rate is
This is a convex combination of the age-specific rates. It answers the question: what overall rate would result if the population had the standard age distribution while retaining its observed age-specific rates?
Rates are often multiplied by a reporting constant, such as . The scaling does not alter the underlying calculation.
Numerical example
Consider two hypothetical populations with the following annual event rates:
| Age group | Population A: rate per 100,000 | Population B: rate per 100,000 | A: population share | B: population share |
|---|---|---|---|---|
| Younger | 100 | 150 | 60% | 90% |
| Older | 1,000 | 1,200 | 40% | 10% |
Their crude rates per are
Population A has the higher crude rate, although its rate is lower in both age groups.
Using a standard population with equal shares in the two groups gives
The standardized comparison reflects A’s lower age-specific rates under a common age composition. These standardized values do not replace the crude rates as descriptions of the populations’ actual event frequency.
Indirect standardization
Indirect age standardization applies reference age-specific rates to the age distribution of the population being studied. It is useful when local age-specific event counts are sparse or unavailable, provided that the population’s age structure and total event count are known.
Let be the reference rate for age group , and let be the corresponding exposure in the study population. The expected number of events under the reference rates is
If is the observed number of events, the observed-to-expected ratio is
For deaths, this is the standardized mortality ratio. A value of indicates that observed deaths equal those expected under the reference rates; indicates 20% more observed deaths than expected. Some publications multiply the ratio by , so that equality is represented by . The equivalent measure for disease incidence is commonly called a standardized incidence ratio.
An indirectly standardized rate may also be calculated as
where is the crude rate in the reference population.
Comparability of indirect measures
Indirectly standardized ratios for different study populations are not generally equivalent to directly standardized rates calculated with common weights. In fact,
where the reference rates are positive. The ratio is therefore a weighted average of age-specific rate ratios, with weights depending on each study population’s age structure.
Consequently, two populations’ SMRs can differ partly because their weights differ. Direct comparisons are more straightforward when age-specific rate ratios are approximately constant across age groups; otherwise, common-weight direct standardization or explicit age-specific comparisons may be more informative.
Choice of standard population
A standard population is a weighting convention, not a population regarded as biologically ideal or especially healthy. It may be based on a particular census, a regional population, a combined study population, or an internationally adopted reference distribution.
Different standards can produce different standardized rates. If one population has lower rates at every age, any common set of nonnegative weights preserves that ordering. If the populations’ age-specific rates cross, however, changing the standard can change their ranking.
Comparable results require consistency in:
- The standard population and its age weights.
- Age-group boundaries and the age range included.
- Outcome definitions and observation periods.
- Denominator definitions and reporting units.
A rate standardized only for adults, or only for a specified age range, is not directly comparable with an all-age standardized rate. When an official standard is revised, apparent discontinuities can arise from the new weights rather than from changed age-specific rates.
Statistical uncertainty
Standardized rates are estimates and may have substantial sampling uncertainty. Treating the standard weights as fixed, and assuming independent age-group estimates,
Under a Poisson event-count model with fixed exposures,
so
The square root gives an approximate standard error. A normal-approximation confidence interval can be inadequate for sparse counts, particularly when it produces a negative lower bound. Gamma-based interval methods are among the alternatives used for directly standardized rates.
For an SMR, uncertainty is often assessed by treating as Poisson-distributed and dividing confidence limits for its mean by , conditional on the reference rates and exposures. When reference rates themselves are estimated imprecisely, treating as fixed omits that additional uncertainty.
Survey-derived estimates may require calculations accounting for sampling weights, clustering, and stratification rather than independent Poisson counts.
Applications and development
Age standardization developed within mortality statistics as a way to compare populations whose age structures differed. It became a routine component of national vital statistics and international health comparisons.
Its principal applications include comparisons of mortality, cancer incidence, disease prevalence, and other age-dependent outcomes across places, population groups, and time periods. In time-series comparisons, keeping the standard fixed helps distinguish changes in age-specific rates from changes associated with population ageing.
The same general approach can standardize jointly for age and other characteristics. For example, age–sex standardization applies common weights to age–sex strata. More detailed stratification, however, increases the number of rates that must be estimated and can create sparse cells.
Limitations and related methods
Standardized rates do not measure actual event totals. Event counts depend on population size, actual age composition, and age-specific rates. A population can experience falling age-standardized mortality while its number of deaths rises because it grows or ages.
Adjustment can remain incomplete. Broad age bands leave differences in age composition within each band. Missing ages, age misreporting, inconsistent classifications, and incomplete event registration can also affect comparisons.
A single summary can conceal heterogeneous patterns. Opposing trends at younger and older ages may cancel in the standardized rate. Differences between populations may also vary substantially by age, a form of statistical interaction that a single weighted average does not display.
Direct standardization requires usable age-specific estimates. Small populations may yield unstable rates, especially in age groups receiving appreciable standard weight. Indirect standardization avoids requiring all local age-specific rates, but introduces the comparability limitations described above.
Standardization is not a causal explanation. Removing differences in age composition does not identify why age-specific rates differ. Other confounding factors, selection processes, and measurement differences may remain.
Regression-based methods, including generalized linear models, can estimate rates as functions of age and other variables. Averaging model predictions over a specified reference population gives a model-based standardized measure. Such estimates depend additionally on model specification and on adequate data for the reference population’s characteristics.
Age-standardized mortality is also distinct from life expectancy. A life table combines age-specific mortality probabilities through a survival calculation; an age-standardized mortality rate is a weighted average of age-specific rates. The two measures summarize mortality differently and are not interchangeable.