13 Research methods, statistics, critical review and evidence-based practice13.2 Epidemiological studies
13.2.1.3 Mortality statistics
3 min read1 figure
Written by Resident Doctor Committee · Published
Crude and specific rates
A mortality rate relates deaths to the population at risk over a stated period, so always check the denominator, the time and the multiplier.
Rate | Numerator | Denominator |
|---|---|---|
Crude | All deaths, at all ages, from all causes | Total population, with no adjustment |
Specific | Deaths in one subgroup, such as an age band, a sex or a cause | That subgroup |
Cause-specific | Deaths from one named cause | Whole population of interest over a period |
Case fatality | Deaths from the disease | People with the disease |
Proportional mortality | Deaths from one cause | All deaths |
Crude rates describe burden, but comparisons between populations are distorted by differences in age structure.
Age-specific rates compare the risk of death across age bands, and between populations, without confounding by age structure.
Two cohorts of 2,000 with identical age-specific rates had crude rates of 3.5 and 11.0 per 1,000, because one had 900 patients aged 55 to 74 and the other 100.
Infant and perinatal mortality
Rate | Numerator | Denominator |
|---|---|---|
Infant mortality | Deaths of live-born children before their first birthday, excluding stillbirths | Per 1,000 live births |
Perinatal mortality | Stillbirths plus early neonatal deaths | Per 1,000 births |
Unlike infant mortality, the perinatal numerator includes babies born dead.
Direct and indirect standardisation
Feature | Direct | Indirect |
|---|---|---|
Age structure used | Standard population's | Study population's own |
Rates used | Study population's own | Reference population's |
Output | Standardised rate | Expected deaths, compared with observed |
Preferred when | Study stratum rates are dependable | Stratum rates are sparse, as in most of psychiatry |
A directly standardised rate is hypothetical, so multiplying it by the real population does not recover actual deaths.
Directly standardised rates are comparable only if they use the same standard; the Office for National Statistics uses the 2013 European Standard Population, and rates on the 1976 standard are not interchangeable.
Standardised mortality ratio
The standardised mortality ratio (SMR) is observed deaths divided by the deaths expected if a standard population's age- and sex-specific rates applied to the study group.
The SMR is the measure that corrects for differences in age structure.
On a reference of 1, the excess is the SMR minus 1, so 1.4 means 40% more deaths than expected.
On a 100-based scale, 100 means deaths as expected and 75 means 25% fewer.
Expected deaths may be fractional because they are calculated, not counted: 22 observed against 10.5 expected gives an SMR of 2.1, or 210.
Arcelus and colleagues reported SMRs of 5.86 for anorexia nervosa and 1.93 for bulimia nervosa.
An SMR shows relative excess, not absolute burden: a high SMR in a young population may mean fewer extra deaths than a lower SMR in an older one.
Limits of the SMR
With few deaths the SMR is unstable, so report observed and expected counts with a confidence interval (CI): an SMR of 150 with a 95% CI of 31 to 438 does not establish an excess.
Two cohorts' SMRs can differ even with identical age-specific mortality, because each cohort supplies its own age weights.
Cause-specific SMRs depend on death certification, since one death may be coded as poisoning, liver disease or undetermined intent.