Standardization

of Rates
26 May 2015

Aims

Concepts and basic methods for

deriving measures that are
comparable across populations that
differ in age and other demographic
variables.

Introduction

Virtually every large population is

heterogeneous in regard to
sociodemographic (e.g., age, gender,
education, religion),
geographic,
genetic,
occupational, dietary, medical history,
and other personal attributes and
environmental factors related to health.

Introduction

An overall measure that does not

take explicit account of the
composition of the population is
called crude.
It is an average of the values for the
individual subgroups (strata),
weighted by their relative sizes.

The larger the subgroup, the more

influence it will have on the crude
measure.

The death rate for a population is a

Crude rate

Consider a population of size N

consisting of five age strata. Each
age stratum has ni (i=1 to 5) people.

During the following year, each

stratum will experience some
number of deaths di.

The total population size, N = ni,

the total number of deaths, D = di,
and the crude mortality rate is D/N.

Crude rate

The crude rate can also be written

as a weighted average of the
stratum-specific mortality rates,
di/ni, as follows:

Note wi=1.

Crude rate

The crude rate is the simplest and

most straightforward summary of
the population.
But mortality is strongly related to
age, so the stratum-specific
mortality rates will differ greatly
from one another.
Crude rate glosses over this
heterogeneity of stratum-specific
mortality rates.

Crude rate

If the populations differ in

composition some of what we
observe may be attributable to these
differences.

Need a method to average the

stratum-specific rates to allow a
comparison between populations.

Standardization

Standardization refers to methods of

adjustment based on weighted
averages in which the weights are
chosen to provide an appropriate basis
for the comparison.
The number of persons in various strata of
one of the populations in the comparison.
An aggregate of these populations, or
Some external relevant population.

Motivating example

Remarks

The difference in crude death rates

between Alaska and Miami results
from differences in age distributions
rather than differential age-specific
death rates.

It follows intuitively that if Miami

and Alaska had the same age
distribution (regardless of what that
distribution might be) their crude
death rates would be similar, since

Direct Standardization

In direct standardization the

stratum-specific rates of study
populations are applied to the age
distribution of a standard
population.

In the example above, each age group is

a stratum.

The directly age-standardized death

rates are equivalent to the crude
death rates which Miami and Alaska

Direct Standardization

Computation of direct standardized

rates is straightforward:

Direct Standardization

Example

The standardized death rate for

white Miami women (1970 U.S.
population of white women as the
standard) is:

Example, contd

The corresponding standardized rate

for Alaska is:

Some points to consider

The directly standardized rate is a

weighted average.

The crude death rate in a population

can be regarded as an average of
the population's stratum-specific
death rates weighted by its own age
distribution.

Motivation for
standardization

Summary rates from two or more

populations are more easily
compared than multiple strataspecific rates.

Important when comparing rates from

several populations or when each
population has a large number of strata.

Small numbers in some strata may

lead to unstable stratum-specific

Standardized ratios or
differences

Rates that have been standardized

by the direct method, using the
same standard population, may be
compared in relative or absolute
terms

as a ratio or as a difference

We can obtain a "Standardized Rate

Ratio" (SRR) by dividing the
standardized rate for Miami by that

Standardized ratios or
differences

Similarly, the difference of the two

rates would be a "standardized rate
difference"
SRD = 6.926.71=0.21

per 1,000 the ratio has no need for the

scaling factor, but the difference does.

Standardized ratios or
differences

Since the rates are virtually

identical, the SRR is close to 1, and
the SRD is close to zero, all give the
same message:

The mortality experience in Alaska,

Miami, and the total U.S. are all
about the same when the differences
due to age structure are eliminated.

Standardized ratios or
differences

A directly standardized rate can be

compared to the crude rate in the
standard population.

The crude rate in the standard

population and the directly
standardized rates are all weighted
averages based on the same set of
weights.

The proportional age distribution in the

Standardized ratios or
differences

So

Sparse data

Standardized rates are not always

meaningful for sparse data.
For the comparisons to be meaningful
there must be large enough numbers in
all important strata.

In strata constituting substantial weight in the

standardization procedure.

There are various rules of thumb for what

constitutes "large enough.

At least 10 or 20 events (e.g., deaths, cases)

and a denominator of at least 100.

Indirect Standardization

Used when data sparse.

An "indirect" standardization
procedure is often used and a
"standardized mortality ratio"
("SMR") computed.

Mirror image of the direct

standardization.

Indirect Standardization

Indirect standardization takes

stratum-specific rates from a
standard population.

These rates are then averaged using

as weights the stratum sizes of the
study population.

In direct standardization, the study

population provides the rates and the
standard population provides the
weights.

Indirect Standardization

Comparison of indirectlystandardized rates can be

problematic.

Each study population's standardized

rate is based on its own set of weights.

Only comparison that is always

permissible: the study population vs
the standard population.

Indirect Standardization

Calculating SMR

Take the observed number of deaths

or events in the study population

Compare to an "expected" number of

deaths.

The number of deaths that would be

expected in the study population if
its stratum specific rates were the
same as for the standard population.

Calculating SMR

The ratio of observed to expected

deaths is termed the Standardized
Mortality (Morbidity) Ratio (SMR).

The expected number of deaths is

obtained as follows:

Calculating SMR

Calculating SMR

Calculating SMR

Calculating SMR

The SMR indicates the relative

excess or decrement in the actual
mortality experience in the study
population with respect to what
might have been expected had it
experienced the force of mortality in
the standard population.

Comparison of SMR:
example

Comparison of SMR:
example

Though both occupations have

exactly the same stratum-specific
rates, their SMR's differ, due to the
substantially different age
distributions for the two
occupations.
However, the directly standardized
rates for both occupations are the
same:

Notes

The apparent equivalence of the

directly standardized rates is
misleading.

Few deaths in the younger age stratum

in Occupation A and in the older age
stratum in Occupation B.

The only conclusion we can be

confident of is that both occupations
have elevated mortality rates
compared to the standard
population.
No evidence from standardization to

Calculating ISR

Indirectly standardized rate (ISR)

can be obtained from an SMR as
follows:

If the study population has twice the

mortality rate of the standard
population, the standardized rate for
the study population should be twice
the observed death rate in the

Mortality example

Miami

Alaska

Summary of rates

Key messages

Populations are heterogeneous

they contain disparate subgroups.
So any overall measure is a
summary of values for subgroups.
The observed ("crude") rate is in fact
a weighted average of
subgroup-"specific" rates, weighted
by the size of the subgroups.
Comparability of weighted averages
depends on similarity of weights.

Key messages

"Standardized" (and other kinds of

adjusted) measures are also
weighted averages, with weights
chosen to improve comparability.
Crude rates are "real", standardized
rates are hypothetical.
The "direct" method (weights taken
from an external standard
population) gives greater
comparability but requires more

Key messages

The "indirect" method (weights

taken from the internal study
population) requires fewer data but
provides less comparability.
Choice of weights can affect both
rates, comparisons of rates, and
comparability to other populations.
Any summary conceals information;
if there is substantial heterogeneity,
the usefulness of a summary is open

Study designs

Study types

1. Case reports
2. Case series
3. Ecologic
4. Cross-sectional
5. Case-control
6. Follow-up/cohort
7. Intervention trials/controlled
trials

Study types

Designs 1 and 2 are employed in

clinical studies.
Designs 3 and 4 are regarded as
primarily descriptive.
Designs 5 and 6 can be employed in
analytic (hypothesis testing) or
descriptive modes.
Design 7 is primarily analytic.

3. Ecologic

Ecologic studies obtain data at the

level of a group, community or
political entity, often by making use
of routinely collected data.
Ecologic studies may also be the
only way to study the effects of
group-level constructs, for example,
laws (e.g., impact of a seatbelt law),
services (availability of a suicide
prevention hotline), or community

4. Cross-sectional

A cross-sectional study is one in which

subjects are sampled without respect to
disease status and are studied at a
particular point in time.
The current or historical status of
individuals is assessed and may be
examined in relation to some current or
past exposure.
Most useful for conditions that are not
rapidly fatal, not terribly rare (e.g.,
elevated blood pressure/cholesterol, many
psychiatric disorders, diet etc.).

4. Cross-sectional:
strengths

Can study entire populations or a

representative sample.
Provide estimates of prevalence of
all factors measured.
Greater generalizability.

4. Cross-sectional:
weaknesses

Susceptible to selection bias (e.g.

selective survival)
Susceptible to misclassification (e.g.
recall)
Information on all factors is
collected simultaneously, so it can be
difficult to establish cause and effect
Not good for rare diseases or rare
exposures

5. Case-control

A case-control study is one in which

persons with a condition ("cases") are
identified, suitable comparison subjects
("controls") are identified, and the two
groups are compared with respect to prior
exposure.
Because case-control studies select
participants on the basis of whether or
not they have the disease, the casecontrol design does not provide an
estimate of incidence or prevalence of the
disease.

5. Case-control

Strengths
Good for rare diseases
Efficient in resources and time
Weaknesses
Susceptible to selection bias (e.g., cases
or controls may not be appropriately
"representative")
Susceptible to misclassification bias (e.g.
selective recall)
May be difficult to establish that "cause"
preceded "effect".

6. Follow-up/cohort

In a follow-up study, people without the

disease are followed up to see who
develops it, and disease incidence in
persons with a characteristic is compared
with incidence in persons without the
characteristic.
If the population followed is a defined
group of people (a "cohort"), then the
study is referred to as a cohort study.
Alternatively, the population under study
may be dynamic (e.g., the population of a
geographical region).

6. Follow-up/cohort

Strengths
Better for rare exposures
Less confusion over relative timing of
exposure and disease.
Weaknesses
Costly and time consuming if disease is
rare and/or slow to develop.
Loss to follow-up may lead to selection
bias.
Relatively statistically inefficient unless
disease is common.

7. Intervention
trials/controlled trials

An intervention trial is a follow-up

study in which the primary exposure
under study is applied by the
investigator.
In an intervention trial, the
investigator decides which subjects
are to be "exposed" and which are
not.

7. Intervention
trials/controlled trials

Strengths
Most like an experiment
Provides strongest evidence for
causality in relation to temporality
and control for unknown
"confounders"
Fulfills the basic assumption of
statistical hypothesis tests

7. Intervention
trials/controlled trials

Weaknesses
Expensive, time consuming,
sometimes ethically questionable.
Subjects are often a highly selected
group (selected for willingness to
comply with treatment regimen,
level of health, etc.) and may not be
representative of all people who
might be put on the treatment (i.e.,
generalizability may suffer).