Methods
Glossary
References
These data were prepared for the third update of Occupational Mortality in
Washington State, (1950-1971), published in 1976 by NIOSH.1
The first update published in 1983
covered the years 1950-1979.2 The second
update covered the years 1950-1989 and was published in
1997.3 The current data tables cover the period from 1950-2010
and includes 1,053,769 male deaths occurring in 1950-2010 and 349,037 female deaths
occurring in 1974-2010 (326,713 deaths to ‘housewives’ were not included in the analysis).
Deaths to Washington residents age 20 or older are included in these data tables.
About 94% of the decedents were classified as ‘white.’ The analysis includes deaths of
Washington residents who died in other states or in Canada. Deaths that occurred in
Washington to people who were residents of other states are not included.
There are two tables contained on this web site: male mortality by occupation for 1950-2010,
and female mortality by occupation for 1974-2010. The tables include the number of observed
and expected deaths, the PMR, and a p-value for eight age groups: 20-64, 20-29, 30-39, 40-49,
50-59, 60-69, 70-79, and 80+, and six time periods: 1950-1959, 1960-1969, 1970-1979,
1980-1989, 1990-1999, and 2000-2010.
Methods
Punch cards containing death record information were available for the years 1950-1971.
Death certificate numbers of all deaths of Washington State male residents, age 20 or older,
for the years 1950-1971 were listed in certificate number order within year of death. The death
record occupation statements were abstracted and coded using a modification of the 1960 U.S.
Census Bureau code. The death certificate numbers and the occupation codes were keyed into
punch cards, and then matched to the original death cards on year and death certificate number.
For matching cards, the occupation code was punched into the initial death card. Non-matches
were resolved clerically. The completed cards were read onto magnetic tape.
For male deaths 1972-1986 and female deaths 1974-1986, occupation was manually coded on a
monthly basis, and added to the routinely prepared computer record of death. These records
were added to the existing 1950-1971 file. Occupational and industry literals were keyed
into the death records from 1987-2010 and computer coded to occupation with a manual backup.
At present, over 97% of death records are successfully computer coded to occupation, with
less than 3% needing manual coding.
Since the study period includes four International Classification of
Diseases (ICD) code changes, tenth, ninth and eighth revision codes were
translated to seventh revision codes. The classification for causes of death
given in this report are according to the seventh ICD revision with one
exception; acquired immune deficiency syndrome (AIDS) is coded according
to the ninth ICD (codes 042-044) starting with 1987 deaths. During the
years 1983-1986, AIDS was coded to the ninth ICD code 279.1 (deficiency
of cell-mediated immunity).
Age adjusted PMRs were calculated for the entire set of deaths, and for
deaths to persons between 20 and 64 years old at death. Fourteen five-year
age groups were used: 20–24, 25–29, ..., 85+. In addition to these PMRs,
age-specific and calendar time period-specific PMRs were calculated in order
to allow a detailed examination of the association between occupation and
cause-of-death. To compute the PMRs, deaths were tabulated as in the
following typical table for a specific age group:
|
Cause A |
All Causes |
Occupation B |
ni |
Ni |
All Occupations |
Mi |
Ti |
where
i = ith age group
ni = observed number of deaths in Occupation B due to
cause-of-death A in the ith age group
Mi = total number of deaths due to cause A in the ith age group
Ni = total number of deaths in Occupation B in the ith age group
Ti = total number of deaths in the cohort in the ith age group
Then
E(ni) = Mi Ni/Ti is
the expected number of deaths in occupation B due to cause-of-death A, and
When both the observed number of deaths and the expected number of
deaths were greater than 5, the Mantel-Haenszel chi-square test statistic was
used to test the null hypothesis that the PMR = 100, and to calculate a
p-value for the test. The Mantel-Haenszel chi-square test statistic is
calculated as follows:4
and the p-value is taken from the chi-square distribution.
When either the observed or expected number of deaths was 5
or less, a binomial test was used where the p value was the
2-sided binomial probability, under the null hypothesis, of
obtaining the observed number or a number more extreme,
relative to the expected number of deaths.5
Glossary
Proportional mortality ratio (PMR)
A proportional mortality ratio is the proportion of deaths
in a specific occupation that are due to a specific
cause-of-death divided by the proportion of deaths in all
occupations that are due to that specific cause of death. For
example, among male school teachers there were a total of
7,778 deaths, and 332 of these were caused by lung cancer.
There were a total of 811,256 deaths to males of all
occupations, of which 46,802 were caused by lung cancer. So
the crude PMR for lung cancer among male school teachers is
(332/7778)/(46,802/811,256) X 100 = 74. (In the data reported
on this website, crude PMRs are not used, instead, the PMRs
are age-adjusted to account for different age distributions in
different occupations.) A PMR greater than 100 indicates that
members of that occupation were more likely than average to
die of that cause-of-death, while a PMR of less than 100
indicates that they were less likely than average to die of
that cause-of-death.
PMRs may also be calculated by dividing the
observed number of deaths by the expected
number of deaths in an occupation and cause-of-death group.
The greatest value of proportional mortality ratios is that
they can be used in studies where mortality rates cannot be
calculated, because of a lack of information on the population
at risk. The greatest weakness of PMRs is that they say
nothing about the overall force of mortality.6
A proportional excess in one
cause of death may mean either an excess in the rate for that
cause of death, or a deficit in the rates of other causes. In
a healthy cohort, a deficit of cardiovascular diseases can
cause a proportional excess of some cancers, even when there
is no excess in the absolute rate of those cancers. Several
researchers have compared the performance of PMRs to that of
standardized mortality ratios (SMRs), and studied the effect
of the `healthy worker effect' on PMRs. Decouflé and
colleagues show that a cause-specific PMR is equivalent to the
cause-specific SMR multiplied by the overall SMR for the
cohort.7 Therefore, the PMR will either
over-estimate or under-estimate the corresponding SMR if the
overall SMR is either less than or greater than 100.
Correct interpretation of the results of PMR studies,
therefore, requires an awareness of the weaknesses of the
technique, and a consideration of the epidemiologic factors
operating on each occupation.
Observed number of deaths
The number of deaths that actually occurred in a specific
occupation and cause-of-death group.
Expected number of deaths
The age-adjusted number of deaths that would have occurred
in a specific occupation and cause-of-death group, if that
occupation had the same mortality experience as the entire
cohort.
p-value
A measure of the statistical significance of the PMR. All the p-values in this analysis are two-sided.
References
1 Milham S (1976). Occupational Mortality in
Washington State, 1950-1971. DHEW Publication No.(NIOSH) 76-175-A,B,C.
2 Milham S (1983). Occupational Mortality in
Washington State, 1950-1979. DHEW Publication No. (NIOSH) 83-116, 1983.
3 Milham S (1997). Occupational Mortality in
Washington State 1950-1989. DHHS Publication No.96-133.
4 Breslow NE and Day NE (1980). Statistical
Methods in Cancer Research: Volume I- The Analysis of Case Control Studies,
International Agency for Research on Cancer, Lyon.
5 Ross S (1988). A First Course in Probability.
MacMillan Publishing Co. New York, third edition.
6 Milham S (1975). Methods in occupational
mortality studies. J Occ Med 17(9):581-585.
7 Decouflé P, Thomas TL, Pickle LW (1980).
Comparison of the proportionate mortality ratio and standardized mortality
ratio risk estimates. Am J Epi 111(3):263-269.
|