National Occupational Mortality Surveillance (NOMS)

Methods

Deaths that occurred during 1984-1998 are grouped on the basis of 280 Ninth Revision, International Classification of Diseases (ICD) coding categories (ICD Categories). For deaths that occurred January 1, 1999 forward, the Tenth Revision ICD codes were used. A decedent’s usual occupation or industry was coded using 1990 U.S. Census codes between 1984 and 1993 and in 2000 U.S. Census codes from 1993 forward by most states (Industry Categories, Occupation Categories).

Data used by the PMR query system in calculating occupational mortality statistics were underlying cause of death, age, race (white, black, or other), Hispanic origin, gender, usual or lifetime occupation, industry, and state and county of residence at the time of death. Industries/occupations with less than three deaths are aggregated unless a specific industry and/or occupation is selected, or unless any sex, race, or Hispanic origin subgroups are selected. Using the query system links, PMRs by Occupation and Cause of Death or PMRs by Industry and Cause of Death, age-adjusted proportionate mortality ratio (PMR) summary statistics by race and gender are generated for industries and occupations with three or more deaths age 15-90 during the period 1984-1998. County-level queries are precluded.

For PMR Charts and Tables query system, death certificates for decedents that died between age 18 and 90 in one of the 27 U.S. states between 1984 and 1998 were the source of age, sex, race, gender, usual industry and underlying and contributing cause-of-death. For the PMR Charts and Tables, multiple cause PMRs were calculated and displayed with all race and genders combined.

Chronic disease PMR s were computed for 22 selected sites of cancer and 17 chronic disease categories for the 11 largest industries of interest within each sector. The Chronic disease PMR Charts query system displays charts and accompanying tables for the selected cancers, chronic disease, and industry categories (Chronic Disease and Industry Categories). The industries and chronic disease causes of death selected for display within each large sector were structured with assistance from the NIOSH National Occupational Research Agenda (NORA) Sector managers. The purpose of the charts was to assist sector councils to identify elevated mortality risk and to identify gaps in industrial mortality and prevention.

Proportionate mortality ratio (PMR) analysis based on the underlying cause of death was used to evaluate the mortality patterns by cause of death, occupation and industry. Race- and ethnicity-specific age–adjusted PMRs were calculated for white, and Black men and women using a computer program developed at NIOSH [Dubrow et al., 1987; Dubrow and Spaeth, 1993]. This program is designed to calculate PMRs for occupation or industry specifically for population-based data. It calculates PMRs by comparing the proportion of deaths from a specified cause within a specified occupation or industry group with the proportion of deaths due to that cause among all other decedents, and age-adjusts after stratification on race (white, Black). A PMR above 100 is considered elevated over the average for all occupations. Ninety-five percent confidence intervals (95% CI) for the observed PMRs were calculated. If the observed number of lung cancer deaths was 1000 or less, the 95% CI was computed based on the Poisson distribution [Bailar and Ederer, 1964]; otherwise, test-based CIs were calculated using the Mantel and Haenszel chi square test [Mantel and Haenszel, 1959]. Statistical significance (p<0.05 for a two-sided test) and 95% CIs should be evaluated in the context of hypothesis generation because multiple comparisons were made [Rothman, 1986]. Although adjustment can be made for the large number of statistical comparisons, other criteria, such as statistical precision, previously published studies, and biologic plausibility, should be drawn on when evaluating the observed associations.

The PMRs indicate whether the proportion of deaths due to a specific cause appears to be high or low for a particular occupation, compared to all other occupations. PMRs are usually computed when data for the population at risk are not available and rates of death or standardized mortality ratios (SMR) cannot be calculated. The population at risk for this study includes all men and women employed usually in a specified occupation or industry, ages 15-90, who were at risk of dying at any time between January 1, 1984 through December 31, 1998. Because data by occupation and industry were not available for the entire population of women at risk of death in the occupations and industries reported on the death certificates, we evaluated proportionate mortality based on cumulative deaths over a 15-year period. The unemployed, part-time workers, those in unknown occupations or industries (about three percent), were excluded from the analysis.

Limitations: Limitations in the PMR method may bias risk estimates toward the null. Misclassification may be a source of bias due to inaccurate reporting of usual occupation and industry, the causes of death, and lack of occupational exposure information. Although the NOMS database lacks information on length of employment, specificity of job description, or estimates of workplace exposures, its advantages over recent studies include its size and its broad geographic coverage, and the recent date of death of the cases. While the degree of misclassification of cause of death varies by disease, fatal chronic disease such as lung cancer is more accurately classified than many other causes of death. (Kircher 1985.)

A statistically significantly elevated PMR cannot be interpreted directly as indicating a causal relationship between the industry or occupation and the cause of death. When a very large number of PMRs are tested for statistical significance, many of the elevated or decreased PMRs will occur due to chance. Other elevated PMRs will be influenced by confounding factors. A lack of significantly increased PMRs may represent the selection of healthy workers for particular occupations or industries. However, recent studies suggest that PMR analysis used for population-based studies may be less biased than cohort study analysis because comparison with other workers lessens the impact of the healthy worker effect.