Encyclopedia of Environmental Science and Engineering, Volume I and II

386 EPIDEMIOLOGY

are found with the leukemias and lymphomas, and the longer

with solid tumors. Often this long period of latency has led

to the failure to recognize an occupational cancer because its

manifestation has been relatively late in a man’s working life

and has been regarded therefore as a spontaneous incidence

with age; this has also been true of postretirement cases.

COHORT ETIOLOGICAL STUDIES

In its basic design, the cohort etiological study is prospec-

tive, since it notes the essential details of all those exposed

to the suspected hazard and awaits the possible development

of cancers. It is possible to estimate the number of cancers

of various sites that would be expected to occur in a group

(cohort) constituted in the same way by sex and age and

observed for the same period of time, from rates of incidence

obtained from a cancer registry in the same area. This would

of course compare morbidity, though it would be equally

possible to use mortality: in general, rather more informa-

tion becomes available from the morbidity study, largely

because of differential survival rates. The same design of

study (i.e., a cohort etiological study) can be conducted

in a retrospective way by the use of records, if well main-

tained by the factory. If, for instance, a full list is available

of those employed, say, 30 years ago, when a new process

or chemical (substance) was fi rst used, and all the personnel

changes since then to present time are also available, then

the situation is almost precisely equivalent to that postulated

above. We set ourselves 30 years back in time, collect all the

original personnel data, and continue to do so until we are

back to the present day. For this reason the method is often

called the “historic cohort method.” The only other infor-

mation required is the present health status of each worker

employed over that period, whether still employed, retired,

or dead. From this information, which is seldom entirely

complete, a comparison is made with the expectations of

deaths and cases of disease, obtained by applying morbidity

and mortality rates to the exposure data.

CALCULATION OF EXPECTED DEATHS

Since both mortality and morbidity rates change rapidly

with age, and also to a lesser extent with calendar years, the

calculation of expectations requires a systematization of the

exposure data by sex, race, age, and calendar years. Usually

it is suffi cient to use 5-year age groups and also groups of

5 calendar years, classifying the data into “person-years”

within the groups. Thus, if in the age group 55–59 in the

early 1970s, the incidence rate of stomach cancer in white

males was 75 per 100,000 (note that this is a rate), and there

were 1,000 person-years (that is, men who were within that

age group and in the early 1970s, for up to a maximum of

5 years each), the total contribution to the expectation of

stomach cancer from this group would be 0.75. The overall

expectation for white-male stomach-cancer cases could be

a result from the summation of similar fi gures to cover the

full range of ages and calendar years required for the known

workforce. In all these computations it is usual to include

only their observed lifetimes, so that a man who dies during

the period of observation contributes to person-years only

until the time of his death. An alternative method ignores the

actual durations of life, substituting the life-table expecta-

tions from their sex and age at the time of entry, including

only of course up to the endpoint of the study. If one of the

effects of exposure at the factory were to cause a shortening

of the normal lifespan, this method would serve to reveal

it. In general, however, it is best to use either method with

circumspection, since other features may also contribute

to distortion of the expectations, such as the HWE or SPE,

both described earlier in this chapter. The appropriate rates,

whether of mortality or morbidity, need to also be carefully

chosen. There may be no choice, if only the rates for the

country as a whole are available: this is often true for mor-

tality rates, though there may be some regional variants.

Cancer registry data may be more regionalized and there-

fore more appropriate to the location of the factory or study

population.

COMPARISON: THE SMR

The comparison of the observed numbers of deaths (or cases

of disease) with the number expected is usually expressed in

the form of an SMR, since the aim of the method of calcula-

tion of expectations is to obtain fi gures that have allowed for

all the pertinent factors that distinguish those at risk, such

as sex, age, race, calendar-year period, and region, amount-

ing to a similar process to that of standardization described

earlier. In general the comparisons will be made separately

for different hazards, and also for different disease groups.

Frequently they are evaluated for their statistical signifi cance

by use of the Poisson test, where the difference between the

observed and expected numbers is set against the square root

of the expected to provide a ratio that, if the expected is suffi -

ciently large (12 or more), can be regarded approximately as

a t -test, but otherwise needs specifi c calculation or recourse

to tables of the function. In interpreting the contrast between

observed and expected, it is important to keep in mind the

source of the expectations, and its relevance to the compari-

son: if it based on mortality rates for a whole country, it will

include those physically handicapped or otherwise unable to

work, which favors the factory population and leads to the

HWE, and also will include other groups of the population

(e.g., social classes, other industries and occupations) that

may serve to distort the comparison. SMR values are com-

monly used to evaluate risks associated with occupational

and industrial situations.

THE PROPORTIONAL-HAZARDS MODEL

A different approach to the evaluation of specifi c hazards has

been devised that makes use in effect of a series of internal

comparisons. It is known as the method of regression models

C005_011_r03.indd 386C005_011_r03.indd 386 11/18/2005 10:25:44 AM11/18/2005 10:25:44 AM