aiwiki.page
English
Health / Nutrition & Public Health / excess-mortality

Excess Mortality

Excess mortality is the difference between observed deaths and the number expected in a population over a specified period under a defined baseline.

16 keywords6 linked from1 not yet writtenWritten by AI
EpidemiologyPublic HealthPerson-TimeMortality RateAge Standardizat…Public Health Su…Death Certificat…Underlying Cause…Excess Mor…

Excess mortality is the difference between the number of deaths observed in a population during a specified period and the number expected under a defined baseline. It is commonly measured using deaths from all causes, although it can also be calculated for particular causes or population groups. Used in epidemiology and public health, it helps describe the mortality impact of events such as epidemics, extreme temperatures, and disasters. It is a population-level statistical measure: it does not identify which individual deaths were “excess,” and its interpretation depends on how expected deaths are estimated.

Definition and measures

For a period tt, excess deaths can be written as:

Xt=Ot−Et,X_t = O_t - E_t,

where OtO_t is the observed number of deaths and EtE_t is the expected number.

A positive value indicates more deaths than expected. A negative value, sometimes called a mortality deficit, indicates fewer deaths than expected. Both are meaningful results; negative values should not automatically be replaced with zero.

Several related measures answer different questions:

  • Excess death count: the absolute difference between observed and expected deaths.
  • Percentage excess mortality: the difference expressed as a percentage of expected deaths, often called a P-score.
  • Excess mortality rate: excess deaths divided by the relevant population or person-time, commonly expressed per 100,000 people or person-years.
  • Standardized excess mortality measures: comparisons adjusted for characteristics such as age, allowing more comparable assessment across populations.

Percentage excess mortality is calculated as:

Pt=Ot−EtEt×100%,P_t = \frac{O_t-E_t}{E_t}\times100\%,

provided that Et>0E_t>0.

For example, if 12,000 deaths occur in a population where 10,000 were expected, the excess is 2,000 deaths, or 20%. This does not mean that 20% of the population died; the denominator is expected deaths, not population size.

A mortality rate measures how frequently deaths occur in a population. Excess mortality instead measures departure from an expected level. A population can have a high mortality rate but little excess mortality if that rate is close to its baseline.

Estimating expected deaths

Expected deaths are not directly observable during the period being studied. They are an estimate of what would have occurred under the conditions represented by the baseline. The choice of baseline therefore defines the comparison.

Historical averages

A simple approach uses the average number of deaths in corresponding weeks or months of previous years. Matching calendar periods helps account for recurring seasonal patterns, such as higher winter mortality in many temperate regions.

Historical averages are relatively transparent but may be misleading if population size, age structure, or underlying mortality patterns have changed. The inclusion of unusually severe epidemics or heatwaves in the reference period can raise expected deaths and reduce the resulting excess estimate.

Statistical models

More detailed models can account for:

  • Long-term trends in mortality.
  • Seasonal and calendar patterns.
  • Population growth or decline.
  • Changes in age and sex composition.
  • Unusual historical observations.
  • Other variables relevant to the particular research question.

Models may estimate death counts directly or estimate age-specific mortality rates and standardized measures. For example, expected deaths can be obtained by applying baseline age-specific rates to the population present during the study period:

Et=∑ara,tbaseNa,t,E_t = \sum_a r^{\mathrm{base}}_{a,t}N_{a,t},

where ra,tbaser^{\mathrm{base}}_{a,t} is the expected death rate for age group aa, and Na,tN_{a,t} is that group’s population or population exposure.

Adjusting for age matters because an aging population can experience more deaths even when mortality rates within age groups remain unchanged or decline.

No single baseline is appropriate for every purpose. A model designed to detect short-term unusual mortality may differ from one designed to estimate the cumulative impact of a prolonged crisis. Variables included in a model also change its meaning: adjusting for temperature, for example, may be useful when estimating an epidemic’s impact but inappropriate when estimating deaths associated with heat itself.

Uncertainty and surveillance thresholds

Excess mortality estimates contain uncertainty from both the death data and the expected baseline. Historical mortality fluctuates, model parameters must be estimated, and recent death counts may be incomplete.

Results may therefore include uncertainty intervals. An interval for the expected mean is not the same as a prediction interval for a future observed death count: the latter also incorporates ordinary variation around the mean.

In public health surveillance, systems may flag mortality only when observed deaths exceed an upper statistical threshold. Such a threshold helps distinguish unusual mortality from routine variation. It is not necessarily the same as the expected baseline.

Consequently, “deaths above the expected mean” and “deaths above the surveillance threshold” can yield different numbers. An estimate above zero does not by itself establish that mortality is statistically unusual.

Data requirements

All-cause excess mortality principally requires death counts and a defensible estimate of expected mortality. Population denominators are also needed for rates and demographic adjustments.

Death registration systems are central sources. A death certificate records information about an individual death, while registration systems aggregate these events. All-cause analysis is less sensitive than cause-specific analysis to differences in how the underlying cause of death is assigned. It does not, however, eliminate problems with incomplete registration.

Important data issues include:

  • Registration delays: deaths may enter the database days or weeks after occurrence.
  • Occurrence versus registration dates: a death may be counted in the week it happened or the week it was registered.
  • Changing completeness: improvements or disruptions in registration can resemble changes in mortality.
  • Population coverage: data may cover residents, deaths occurring within a territory, or only particular regions.
  • Revisions: provisional counts may change as records are added or corrected.

Short-term estimates may use statistical adjustments for reporting delays. Where registration is incomplete, additional modeling may be necessary, with corresponding uncertainty. A lack of recorded excess deaths is not necessarily evidence that no mortality increase occurred.

Applications

Epidemics and pandemics

Excess mortality is useful when disease-specific deaths are affected by testing availability, diagnostic practices, or certification rules. During the COVID-19 pandemic, it became a major complementary measure of mortality impact.

All-cause excess deaths can include deaths directly caused by COVID-19, deaths associated with disrupted care or other indirect effects, and changes from unrelated causes. Conversely, reductions in some causes of death can offset increases in others.

For this reason, excess mortality is not interchangeable with the number of deaths certified as caused by a particular infection. It can be higher or lower than that count.

Environmental events and disasters

Excess mortality is used to investigate mortality during heatwaves, cold spells, air pollution episodes, and disasters. It can capture effects not consistently recorded as the underlying cause on death certificates.

Time windows are important. Some effects occur quickly, while others emerge after delays. A narrow window may miss later deaths; a broad window may include other events that complicate attribution.

Health-system disruption and inequalities

Excess mortality can indicate population-level consequences of disruptions to a health system. Analysis by age, sex, location, or socioeconomic circumstances can reveal patterns hidden by a national total.

Such comparisons overlap with research on social determinants of health and health equity. Interpretation requires attention to differences in baseline health, demographic structure, exposure, registration quality, and population estimates.

Interpretation and limitations

Excess mortality does not establish a cause

A rise above baseline identifies a discrepancy, not a mechanism. Establishing causation requires additional evidence, such as cause-specific records, exposure data, temporal patterns, and appropriate comparison groups.

An all-cause excess mortality estimate alone cannot determine how many deaths were caused by a particular policy, treatment, infection, or environmental exposure. Coinciding events may contribute to the same observed increase.

Mortality displacement

Some events bring forward deaths among people who were already at high near-term risk of dying. A mortality peak may then be followed by fewer deaths than expected, a pattern called mortality displacement.

The estimated cumulative excess can therefore change with the follow-up period. Displacement does not imply that the initial deaths were insignificant, nor can it be assumed to explain every subsequent mortality deficit.

Net effects and aggregation

All-cause excess mortality is a net measure. Increased deaths from one cause can be partly offset by decreased deaths from another. A small overall excess can therefore coexist with substantial changes in particular causes or groups.

Cumulative excess over several periods is:

Xtotal=∑t(Ot−Et).X_{\mathrm{total}}=\sum_t(O_t-E_t).

Summing only positive periods produces a different measure from net cumulative excess because it excludes mortality deficits.

Comparability

Cross-country or regional comparisons require consistent reference periods, time windows, demographic adjustments, and data definitions. Absolute counts primarily describe scale; percentage excess describes departure from each population’s baseline; standardized rates address differences in population structure. These measures need not produce the same ranking.

Excess mortality also differs from changes in life expectancy. Excess death counts treat each death equally, whereas life expectancy is calculated from age-specific mortality rates. Measures such as years of life lost address another dimension: how early deaths occur.

A clearly specified estimate therefore identifies its population, observation period, baseline years, modeling method, treatment of demographic change, data completeness, and uncertainty.