Huebner, Wendy W. PhD; Wojcik, Nancy C. MS; Jorgensen, Gail; Marcella, Susan P. MBA; Nicolich, Mark J. PhD
Mortality studies of North American employees in this petroleum company have included both genders,1–13 although analyses for women have been limited by low numbers. For example, the updates of several older U.S. refinery/chemical plant cohorts (Paulsboro, Baytown, Beaumont, Torrance, Baton Rouge, and Baytown facilities)6,8,11–13 each contained from 400 to 800 women, of which 10 to 90 had died.
A few larger cohort studies have provided some information about mortality among women in the company, mainly for overall patterns of deaths. For example, a U.S.-based cohort of research and engineering workers from 1964 to 1986 included 3870 women. For the period of 1964 to 1992, analysis of 94 deaths among these women found overall 35% deficits for all causes of death and for the all-cancer subset, compared with death rates in the New Jersey population.7 Breast cancer rates were slightly above expectation and statistically non-significant, based on 16 deaths.
A study of employees from 1964 to 1983 who worked in all operating segments of a Canadian affiliate included 8238 women.10 Their all-cause mortality for the period of 1964 to 1994 was 20% below the Canadian population comparison, based on 506 deaths, and the all-cancer rate was similar to expectation. For ovarian cancer, office workers had a standardized mortality ratio (SMR) of 1.76 (95% confidence interval [CI] = 0.96 to 2.94), based on 14 deaths. In a study of more recent employees in Canada (hired between 1964 and 1994), the subset of 8062 women had 109 deaths and overall mortality 31% lower than the general population comparison.14 The only other statistically significant result was a 59% deficit for all external causes of death. An SMR of 1.74 (95% CI = 0.70 to 3.58) was found for ovarian cancer based on seven deaths.
The creation of a Health Status Registry by Exxon in the late 1980s established a U.S. cohort of relatively recent employees. This cohort includes all full-time, U.S.-paid men and women working in all aspects of the company in manufacturing, office, and field settings (not including retail stores). The initial mortality report for the years 1979 to 1992 included analysis of 373 deaths among the 22,495 women, and it showed a 36% lower overall death rate than expected in the U.S. general population comparison.15 The broad category of circulatory diseases had the largest deficit, 62%, while results for all malignant neoplasms and all external causes of death had deficits of 13% (178 deaths) and 11% (67 deaths), respectively. Motor vehicle accidents (MVAs), a subset of external causes of death, had a death rate similar to expectation, based on 27 deaths. The study did not find any statistically significant elevations for specific causes of death and found statistically significant deficits for subsets of circulatory disease, such as heart disease, and for all nonmalignant diseases of the respiratory system and of the digestive system.15 These results indicated both the overall favorable health profile for these women and the relatively low power of the study to examine more specific causes of death.
U.S.-paid Mobil employees were added to the Registry after the merger of Exxon and Mobil in 1999, and ascertainment of deaths has been completed through 2000. The resulting doubling in size of the cohort has increased the number of women to almost 50,000. This, along with an expanded 22-year follow-up period, has substantially increased power to reliably study women in greater detail. The objective of the research reported here is to assess the overall and specific mortality patterns of this cohort and identify any areas of concern or for future study. This effort will augment and update information from the past U.S. study described above.15 (Results for men have been reported in a separate article.16)
MATERIALS AND METHODS
Employee information is obtained from work history and demographic records derived from human resources databases, collected, and updated in the internal Health Status Registry. The population for this study consists of all U.S.-paid, full-time women with at least one day of employment during the period January 1, 1979 (Exxon), or 1980 (Mobil), and December 31, 2000. (The study group includes women who have some part-time work in addition to their full-time work.)
Vital Status Ascertainment and Study Variables
Vital status ascertainment, determination of study variables, and analysis methods are described in the companion article on results for men16 and are summarized briefly here. We obtain death certificates from company benefits sources and through vital status tracing with the National Death Index (NDI) and the Social Security Administration (SSA). A single nosologist, trained by the National Center for Health Statistics, codes all death certificates according to the International Classification of Diseases (ICD) revision in effect at time of death (ICD 917 for 1979 to 1998 and ICD 1018 for 1999 to 2000).
Several surrogates of exposure were derived from work history records. For job type, we used classifications based on the Equal Employment Opportunity rubric. The first Equal Employment Opportunity code in each subject's work history record was used to represent a possibly more manufacturing- or field-based job in early employment. We also assigned each cohort member to a mutually exclusive operating segment category—downstream, upstream, chemicals, coal and minerals, or combined corporate and global services. One of these segments was assigned if all work records indicated a single segment or, in rare instances (<0.5%), when the vast majority of records were for that segment. All other employees were placed in a “mixed” or “missing” category. (Broadly, the upstream segment includes exploration, development, and production of petroleum resources, and the downstream segment includes refining, marketing, and distribution of products.) Additional work-related information for analysis includes work era, dichotomized as hired before 1960 versus 1960 or later, and job duration and latency (time from first hire to death or end of study), both categorized as 0 to 9.9, 10 to 19.9, and 20+ years.
Study Design and Analysis
This study uses the retrospective cohort design. Each subject's person-time was counted from January 1, 1979 (Exxon), or January 1, 1980 (Mobil), or first hire date, if it was later, to end of study (December 31, 2000) or date of death, whichever was earlier. Cohort members who could not be confirmed as deceased or alive at the end of the study period were considered lost to follow-up, and their follow-up time was truncated at last date of employment.
We used the SMR as a measure of risk. The SMR compares observed deaths in the cohort to expected deaths based on U.S. mortality rates with the same distributions of gender, race (white/nonwhite), and 5-year categories of age and calendar time of death. All SMRs, 95% CIs, and P values were calculated using a modified life table approach, with the software OCMAP-PLUS from the University of Pittsburgh (Pittsburgh, PA).19 Death rates for 95 causes of death from the National Center for Health Statistics were obtained through the University of Pittsburgh and included the 63 standard causes and others of interest to the study. We used underlying cause of death for all analyses.
As sample sizes allowed, we examined subgroups by job type and operating segment and, as appropriate and when possible, further explored patterns by work era, job duration, and latency. We consulted death certificates and additional work history information, when available, to describe specific causes of death of interest.
Tables and graphs include SMRs if there are at least five observed or five expected deaths; otherwise, we report observed and expected numbers only. Positive findings carried more weight if they had greater statistical stability (ie, a narrower CI).
Quality Control and Human Subjects
To ensure data accuracy, consistency, and completeness, we used several data quality procedures, including cross-checks of data from different record sources, double data entry for death certificate information, and independent audits of computer output and reports.
The study required no contact with study subjects or their next of kin. Strict confidentiality standards were upheld in all phases of the study according to requirements of the Good Epidemiology Practices20 and those of NDI. The protocol was approved by the Institutional Review Board at the University of Texas Health Science Center in Houston and by an internal health research ethics committee that includes two academic advisor/ethicists.
The U.S. cohort contains 49,705 women, with approximately half from each of the parent companies. Table 1 describes characteristics of the cohort, which is mainly white and relatively young, with 60% hired since 1980 and 64% under age 50 years at last observation. Sixty-one percent have fewer than 10 years of employment, and less than half have latencies of more than 20 years. At the end of the study period (December 31, 2000), the majority (71%) were alive and either retired or separated from the company, 17% were still actively employed, and 4% had died. The nearly 9% of women lost to follow-up compares with only 3% among the male segment of the cohort16 and is likely explained by difficulties in follow-up caused by surname changes among women.
The majority (61%) of deaths were ascertained from NDI, 38% from company benefits, and 1% from SSA. Corresponding with locations of company facilities in the U.S., the largest proportions of deaths occurred in Texas (30%), New York (11%), and New Jersey (9%), and approximately 5% each from California, Illinois, and Louisiana. We obtained death certificates for 98.1% of deaths (most others had death notices or confirmation from SSA) and have a specific cause of death for 98.3%.
Table 2 displays distributions of women by job type and operating segment. Seventy-five percent are office/clericals or professionals, and the three most common operating segments, in descending order, are downstream, chemicals, and upstream. We also note that among groups of blue-collar women, laborers and operators are mainly in the chemicals operating segment (81% and 48%, respectively), and 55% of craftsmen are in the downstream segment. Overall, our ability to examine women by these and other subgroups varied considerably, depending on the number of people in a subgroup and on the commonality of a particular cause of death.
SMR Results and Commentary
Results and comments are presented together in this section. The Discussion section then examines strengths and limitations of the study.
As noted above, 4% (1947) of the 49,705 women have died, and SMR calculations are based on 774,263 person-years of observation during the 22-year follow-up period (1979 to 2000). Table 3 presents results for the 95 causes of death studied.
The all-causes SMR is 0.75 (95% CI = 0.72 to 0.79). SMRs for major categories of malignant neoplasms, diseases of the circulatory system, and external causes of death also have statistically significant deficits. Low overall SMRs are expected in light of the “healthy worker effect” (HWE)—an influence on employee studies because of selection of healthier people into employment and due to health and socioeconomic advantages associated with maintaining employment. These overall results are similar to those for women in the previous U.S. study.15
Table 3 shows that no causes of death have statistically significant elevations and that there are several statistically significant deficits. Among the 1947 deaths, 815 (42%) are due to cancer, the most common being malignancies of the respiratory system and breast, as is true for the U.S. adult female population. The overall SMR for cancer is 0.91 (95% CI = 0.84 to 0.97), with SMRs for most cancer subtypes at 1.0 or slightly/moderately below. In contrast, a stronger HWE is suggested for circulatory system diseases, with 35% to 40% lower rates for heart disease and strokes, based on 341 and 91 deaths, respectively. The 203 deaths due to external causes are 13% below expectation. About half of these deaths are due to accidents, with a statistically significant SMR of 0.77, and the other half are from suicide or homicide, each with an SMR near unity.
Findings by Job Type and Operating Segment
We then examined SMRs for all-cause mortality and the main cause-of-death categories by job type, and Fig. 1 displays findings for job-type groups with at least 100 expected deaths (professionals, office/clericals, technicians, operators, and laborers).
The graphs show that white-collar groups (office/clericals and professionals) have deficits generally greater than those of blue-collar groups (operators and laborers), with results for technicians in between the two groups. This white-collar/blue-collar variation is expected and is probably explained mainly by differences related to lifestyle factors. We cautiously interpret differences or similarities between job types (and between operating segments—see next paragraph), because results may be confounded by age differences among study populations (SMRs are adjusted for age within a particular analysis, but not between separately run analyses of populations). However, we examined age distributions between job-type groups and between operating segment groups and did not see large differences.
As Table 4 shows, there are also differences in the major cause-of-death categories by operating segment. Namely, the chemicals, and coal and minerals segments have smaller deficits and some elevations compared with the others. This does not appear to be due to confounding by age. Rather, these two segments have relatively large proportions of blue-collar workers in their populations (Table 2), and presumably, this is a lifestyle-related reason for differences by operating segment.
Subgroup Findings for Cancer
Regarding common cancers in women, SMRs for breast cancer vary slightly above or below unity for the job types with sufficient numbers for study. We note similar results in other company studies that report 20 to 50 breast cancer deaths10,14,15 or report incidence cases.14 In this study, overall SMRs by job type for all cancers of the digestive system and all cancers of the respiratory system are generally unremarkable, except for an elevation of the latter among operators (discussed below). For all cancers of the blood and blood-forming organs (lymphohematopoietic malignancies), we found 60 total deaths and, where numbers were sufficient for analysis, no elevations by lymphohematopoietic subtype or job type. Women hired before 1960 account for 24 of the 60 deaths, for a slightly elevated, statistically not significant SMR of 1.27 (95% CI = 0.81 to 1.88).
Within the limited statistical power to study specific cancers among women by job type, a few cancer findings of note are shown in the first three rows in Table 5. In contrast to an SMR for lung cancer near unity for professionals and a deficit of 0.73 for office/clericals (not shown), Table 5 shows that women operators have an elevation of lung cancer, based on 14 deaths. Most of the operator decedents were hired in the 1970s and died at ages from about 55 to 65 years. No patterns in job title are seen. Nine of the 14 decedents worked in the chemicals segment, for an SMR of 1.97 (95% CI = 0.90 to 3.75). Lack of job- or time-related patterns argues against a workplace etiology. Rather, tobacco use, a lifestyle factor related to socioeconomic status (SES), is a more likely explanation, although we do not have smoking data to examine this.
In an examination of a possible relationship between smoking and job class, a study of Shell employees in several U.S. refinery and petrochemical plants reported that smoking among women varied by working group, as measured by smoking history questions during company physical examinations during 1976 to 1979, 1980 to 1989, and 1990 to 1997.21 The authors noted that while age-adjusted smoking rates had decreased over these time periods, women in production (hourly) jobs had a higher smoking prevalence compared with women in staff (salaried) jobs and to all U.S. women. When measured by cigarettes per day, the difference between hourly and salaried women was largely diminished, but both groups had considerably higher consumption than all U.S women.21
Our previous company studies do not show elevations of lung cancer in women10,11,14 and neither do results from other petroleum and chemical companies.22–24 A study from Iceland of nearly 19,000 female manual workers in many industries found an SMR for lung cancer of 1.66 (95% CI = 1.00 to 2.59), among women with 5 or fewer years' job duration and 10 or more years of latency.25
Also shown in Table 5, office/clericals have a slight elevation of ovarian cancer deaths that is borderline significant. Examination of records for these 46 decedents did not reveal any patterns by job tenure, latency, state where death occurred, parent company, age at hire, or age at death. The table indicates a small difference for all women by work era, with a slightly increased SMR for women hired before 1960 and an SMR near unity for those hired between 1960 and 2000.
As noted previously, the elevation of ovarian cancer in the company's older Canadian cohort is also among office workers (14 deaths).10 The latest update of the Baytown cohort13 had five deaths among white women compared with 1.01 expected; this is a statistically significant difference, but the CI is wide. Results for this rare cancer are not usually reported in other petroleum industry studies. Although ovarian cancer is understood to be related to age and reproductive factors, the exact cause is not known and work-related etiologies are not established.26
Bladder cancer is also elevated among office/clerical workers, whose nine deaths result in an SMR of 2.08 that is not quite statistically significant. The decedents have generally worked for 10 or more years and have no patterns regarding operating segment. Viewed by work era (Table 5), this elevation is exclusive to women hired in 1960 and later; they have a statistically significant SMR of 2.19, and 7 of the 10 decedents are office/clerical workers. Other studies of women from this company10,11,14 and others22–24 that have reported bladder cancer results had numbers of deaths too small (n = 0 to 4) for meaningful analysis.
Subgroup Findings for Circulatory, Noncancer Respiratory, and Neurological Diseases
For all heart disease combined and the subset of sudden heart attacks (acute myocardial infarction), professionals and office/clericals have statistically significant deficits of 30% to 50%. In contrast, mortality from heart disease is near unity for operators and laborers.
The last two rows in Table 5 show elevations among laborers for chronic-disease causes of death. The first is a doubling of cerebrovascular disease deaths. No patterns are evident regarding work era, length of employment, latency, or any age-related factors. Twelve of the 15 decedents are from the chemicals segment, for an SMR of 1.92 (95% CI = 0.99 to 3.35). (As mentioned earlier, 81% of laborers in the cohort are in the chemicals segment.) Regarding other job types, the SMR for technicians is at unity (6 deaths), and the other job-type groups with sufficient numbers for analysis have SMRs well below unity.
Women who work as laborers also have an elevation of deaths from chronic obstructive pulmonary disease (COPD) that is 2.5 times greater than expected. Five of the nine decedents died in Illinois and had worked as packers. As with cerebrovascular disease among laborers, most (8 of 9) decedents worked in the chemicals segment, with an SMR of 2.55 (95% CI = 1.10 to 5.02).
In contrast to female laborers, office/clericals have statistically significant deficits for all respiratory diseases combined and for COPD specifically, and professionals have deficits that are not statistically significant. Only professionals have an elevated SMR for emphysema, with an SMR of 1.87 (95% CI = 0.75 to 3.85), based on seven deaths.
This variation in chronic disease mortality by job type is consistent with a study by the National Institute for Occupational Safety and Health that analyzed mortality by socioeconomic class,27 a factor that is related to job type and lifestyles. The study examined mortality between 1984 and 1997 for employed persons in 27 U.S. states who were between the ages of 35 and 64 years. SES was determined by usual occupation on the death certificate, and denominators were based on occupation-specific U.S. Census data. A 29% higher death rate was found for all causes of death among women in the lowest SES quartile versus the highest SES quartile. The researchers found a similar elevation for lung cancer and successively higher percent increases for external causes of death, stroke, coronary heart disease, diabetes, and COPD (the latter reaching a doubling of risk in the lowest versus highest SES quartile). When looking at all four SES quartiles from lowest to highest, the trend is clear for coronary heart disease and COPD and less consistent for the others.
Amyotrophic lateral sclerosis, a progressive degenerative disease of nerve cells, has a slightly elevated SMR of 1.35 (95% CI = 0.65 to 2.48) in the cohort. Seven of the 10 decedents worked as office clericals, with an SMR of 1.52 (95% CI = 0.61 to 3.13). Among the 10, 9 had latencies of 20 years or longer, and there are no patterns for date of hire, duration of employment, age at death, or state where death occurred. All deaths occurred between 1993 and 2000, with seven between 1997 and 2000, and additional years of follow-up should help clarify patterns for this disease.
Subgroup Findings for External Causes of Death
As previously shown in Fig. 1, mortality from external causes of death is above expectation for the two blue-collar groups with sufficient numbers to study—operators and laborers. Some details about the causes of death involved are shown in Fig. 2 and described below.
Female operators and laborers who died in MVAs are mostly in their 20s and 30s and have job tenures from 0 to 13 years (many <5 years), and no patterns by state where death occurred. Eight of the 11 operators' deaths occurred before 1990 and 9 died while employed. For the 12 laborer MVA decedents, year of death is more evenly distributed throughout the 22-year study period, and seven died while employed. We were able to examine all but three of the death certificates of women in both groups who died while employed and found that none were designated as having died while at work. For laborers only, decedents were mainly from the chemicals segment, with an SMR of 2.13 (95% CI = 0.98 to 4.05), based on nine deaths.
Although numbers of women from the coal and minerals segment are generally too small for analysis of specific causes of death, we note a borderline statistically significant elevation of MVAs, based on five observed versus 1.6 expected deaths. Three of the five decedents had office/clerical jobs. Two studies of women in the chemical industry, also with small numbers of MVA deaths, reported elevations among women with hourly versus salaried jobs23 and among women with mainly production jobs versus a general population comparison.28
An elevation in suicide (eight deaths) is seen among laborers and is borderline statistically significant, although based on small numbers (Fig. 2). Decedent profiles are similar to those seen for the MVA laborer decedents, with short job tenures of 0 to 5 years, and most in their 20s or 30s at time of death. No patterns are seen for year of death. Four died while employed, with no indication on death certificates that the suicides occurred at work. Six of the eight decedents were laborers in the chemicals segment, with an SMR of 2.41 (95% CI = 0.89 to 5.25).
A company study of women in research and engineering found five deaths by suicide versus 2.7 expected (not statistically significant).7 This finding and a reported elevation among women in a chemical company23 involve numbers too small for interpretation.
Referring once more to Fig. 2, SMRs for homicide and legal intervention are at 3.0 and nearly 2.5 for operators and laborers, respectively. As noted among women who have died of other external causes, there is a tendency toward short job tenure and death at an early age. No other job- or time-related patterns are seen. All seven laborer decedents were separated from the company at time of death, whereas four of the seven operator decedents died while employed, with no indication on death certificates that the homicides occurred at work. Regarding other company studies, the latest update of the Baton Rouge cohort found four homicide deaths among women versus 1.01 expected.13
These elevations for several external causes of death, based on small numbers, offer no indication of workplace etiology. Alternatively, the results are suggestive of lifestyle factors associated with SES and are consistent with a large U.S. study conducted by National Institute for Occupational Safety and Health and others that shows relationships between SES and external causes of death among women.29 As noted earlier,27 researchers examined mortality during 1984 to 1997 for employed persons in 27 U.S. states, determining SES by usual occupation on death certificates and using occupation-specific U.S. Census data for denominators. For analysis of external causes of death, the age range studied was 20 to 64 years.29 A 60% higher death rate for all external causes was found for women in the lowest SES quartile versus the highest, with considerably smaller elevations in the middle two quartiles. The subset of MVA deaths had a similar percentage increase, and suicide and homicide had increases of about 25% for the lowest SES quartile compared with the highest SES quartile.
Study Strengths and Limitations
This study of nearly 50,000 cohort members offers the best opportunity so far to study mortality trends among women in the petroleum industry. Almost all other company studies and others in the industry lacked sufficient numbers of women to study specific causes of death by job type. Also, the cohort definition of working at least one day in the period of 1979/1980 to 2000 allowed study of employees whose experience is more relevant to today's workplace conditions.
Other strengths include the completeness of the cohort as defined from computerized payroll records, the quality of personnel information, and the high degree of capture of deaths from intensive searches of NDI and SSA information. This argues for results that are not biased downward because of incomplete ascertainment of cohort members or deaths (follow-up time for those lost to follow-up was truncated at last date of employment).
As is common with the retrospective cohort design, however, interpretability of the study is limited by potential for both underestimation and overestimation of risk. For example, risks may be underestimated because of a “dilution effect” that can potentially mask a risk in a small segment of the cohort, and because of the HWE, ie, the generally lower mortality among employed persons compared with the general population (discussed in next section).
Conversely, risks may be overestimated because of chance findings from multiple comparisons. SMRs may also be overestimated by the possibility that employed people receive more accurate diagnoses for some causes of death than others in the general population.30
Although the analysis takes into account the differences between the cohort and the general population regarding gender, age, and race, results can be biased in either direction by the inability to account for potential confounding by influences such as lifestyle factors and employment elsewhere. Another limitation is our necessary reliance on surrogates of “exposure” rather than actual exposure measurement or estimation. Among these surrogates, we focused on analyses by job type and operating segment to describe patterns of deaths among subsets of female employees. Such analyses are more informative for subgroups with larger numbers and more common causes of death. As such, office/clericals and professionals are best studied, whereas women in blue-collar jobs have smaller sample sizes that limit interpretation for many specific causes of death, including most cancers. Interpretation of subgroup results is further limited by the general nature of the job type and operating segment classifications. Finally, as noted previously, this cohort of women is relatively young, with shorter tenures and latencies, and future updates with additional follow-up time will allow more extensive assessment of their mortality patterns.
Generally Low SMRs—Impact of the HWE
The HWE is no doubt an important factor in the low overall SMRs observed in this cohort, as has been discussed previously.16 It is generally held that there are various possible selection factors that make it more likely that healthier people gain employment and that staying employed offers continuing positive effects, such as better access to health care and healthy lifestyles.31–33 There is agreement that the HWE is modified by time-related factors; attenuation of the HWE has been observed with increasing time since hire, follow-up time, duration of employment, and age at risk.31–35 The current cohort of relatively young women has not had time for this attenuation to occur. Another likely contributor to the low overall SMR in this cohort is the large proportion of white-collar workers with presumably healthier lifestyles and safer living environments associated with their socioeconomic level. It is likely that familiarity with company health and safety programs has also had an impact.
The HWE is known to vary by cause of death, such as having a stronger influence on the all-cause category than on the all-cancer category, because cancer tends to have fewer apparent symptoms before diagnosis than some other diseases.31–33,35 This is borne out in our data, in which the SMR for all cancer is 0.91 compared with the SMR of 0.75 for all causes of death. We note that a large study of employed Swedish women (compared with unemployed women) did not demonstrate any HWE for cancer incidence.36
Regarding possible gender differences in the HWE, a stronger HWE in women compared with men has been reported in an analysis of two Canadian occupational cohorts.35 A study of various aspects of the HWE among workers in the synthetic vitreous fiber industry in Europe found the absence of an overall HWE in both genders and also: 1) a stronger “healthy hire effect” in men versus women; 2) a slight “time since hire” effect in men only; and 3) a stronger “healthy worker survivor effect” (mortality lower in those who keep jobs) in women versus men.34
In our cohort, there is a somewhat stronger overall HWE in men16 compared with women (overall SMRs of 0.66 and 0.75, respectively), with no gender difference for circulatory disease and large differences for cancer and external causes of death. We see in other cohorts from this and other companies that overall HWE gender differences (when there are sufficient numbers of women for study) are often slight and can vary in either direction. This is understandable, considering the many possible modifiers of the HWE that have been discussed.35
It was important to include recently hired employees and other short-duration employees in this cohort because of their relevance to analyses of external causes of death and other short-latency causes. However, some have argued that SMRs in industry studies are biased downward because of inclusion of people with insufficient time to cause or detect longer-term work-related effects. To address this possibility, we compared all-cancer SMRs in groups that did, and did not, include women with shorter tenures or follow-up times (data not shown), and we did not find the hypothesized downward bias. Rather, these analyses found essentially the same results whether or not short-duration women were excluded (excluding durations of at least 1, 5, and 10 years, respectively). A similar analysis that excluded short-latency subjects (requiring a latency of 5, 10, and 20 years, respectively) also did not find an increasing all-cancer SMR as the latency requirement increased. In fact, the SMR was decreased in the analyses restricted to those with at least 20 years of latency; this suggests a healthy work survivor effect among women, as mentioned above.34 We also conducted conventional duration and latency analyses using cut points of 0 to 9.9, 10 to 19.9, and 20+ years and found slight, statistically not significant trends toward unity. These results are not suggestive of work-related effects.
In SMR comparisons between women hired before 1960 and those hired during 1960 to 2000, the earlier-era group has a 3% higher all-cause SMR and 6% and 7% higher SMRs, respectively, for cancer and circulatory diseases (all below 1.0). This is expected, considering that differences between employed people and the general population diminish in older age groups. Interestingly, for external causes of death, all but 9 of the 203 deaths occurred among women hired in 1960 and later, with an SMR of 0.90 (95% CI = 0.77 to 1.03), compared with an SMR of 0.53 (95% CI = 0.24 to 1.01) for women hired before 1960.
To the best of our knowledge, this is the largest mortality surveillance study of women in the petroleum industry. Findings indicate a generally favorable mortality picture compared with an external population, and this is similar to results for women in our earlier U.S. study.15 The current study has been able to more closely examine specific causes of death, job type, and operating segment. Among the results, a small number of elevations of cancer subtypes are not suggestive of workplace etiologies. In addition, MVAs and other external causes of death are increased for some subgroups of blue-collar women who are younger and have worked in the company for shorter periods. Elevations are also seen in some of the same subgroups for chronic diseases such as cerebrovascular and respiratory disease. None of these findings show work-related patterns, and all are based on 15 or fewer deaths. Taken together with SMRs near unity among operators and laborers for heart disease (ie, not a deficit as we might expect in light of the HWE), the total picture suggests lifestyle influences. Future updates will have increased power to confirm or refute these patterns. Increased power will also allow the study of a broader array of specific causes of death and of more subsets of the women in this cohort. Finally, these results and future updates provide information for consideration in developing and evaluating prevention and intervention programs that can be directed to the appropriate groups.
We thank everyone involved in the conduct of this study (previously noted in the earlier report on men's results16). We also thank Brian Doll, Eileen Pearlman, and Jeff Lewis for their helpful comments on this manuscript of women's results. We thank the dedicated professionals in the U.S. state health departments for supplying death certificates for the Health Status Registry, and we also acknowledge the state disclaimers of responsibility for any analyses, interpretations, or conclusions. Death certificates from Pennsylvania, New York State, and New York City were obtained for purposes of this study only.
This study was funded by Exxon Mobil Corporation.
1.Hanis NM, Stavraky KM, Fowler JL. Cancer mortality in oil refinery workers. J Occup Med. 1979;21:167–174.
2.Hanis NM, Shallenberger LG, Donaleski D, Sales EA. A retrospective mortality study of workers in three major U.S. refineries and chemical plants. Part 1: comparisons with U.S. population. J Occup Med. 1985;27:283–292.
3.Shallenberger LG, Acquavella JF, Donaleski D. An updated mortality study of workers in three major United States refineries and chemical plants. Br J Ind Med. 1992;49:345–354.
4.Schnatter AR, Theriault G, Katz AM, Thompson FS, Donaleski D, Murray N. A retrospective mortality study within operating segments of a petroleum company. Am J Ind Med. 1992;22:209–229.
5.Schnatter AR, Katz AM, Nicolich MJ, Theriault G. A retrospective mortality study among Canadian petroleum marketing and distribution workers. Environ Health Perspect. 1993;101:85–99.
6.Collingwood KW, Raabe GK, Wong O. An updated cohort mortality study of workers at a northeastern United States petroleum refinery. Int Arch Occup Environ Health. 1996;68:277–288.
7.Lewis RJ, Yarborough CM, Nicolich MJ, Friedlander BR. Lymphopoietic cancer and other major causes of death among petrochemical researchers: an update. Occup Med. 1999;49:71–77.
8.Lewis RJ, Gamble JF, Jorgensen G. Mortality among three refinery/petrochemical plant cohorts. I. 1970 to 1982 active/terminated workers. J Occup Environ Med. 2000;42:721–729.
9.Gamble JF, Lewis RJ, Jorgensen G. Mortality among three refinery/petrochemical plant cohorts. II. Retirees. J Occup Environ Med. 2000;42:730–736.
10.Lewis RJ, Schnatter AR, Katz AM, et al. Updated mortality among diverse operating segments of a petroleum company. Occup Environ Med. 2000;57:595–604.
11.Wong O, Harris F, Rosamilia K, Raabe GK. An updated mortality study of workers at a petroleum refinery in Beaumont, Texas, 1945–1996. J Occup Environ Med. 2001;43:384–401.
12.Wong O, Harris F, Rosamilia K, Raabe GK. Updated mortality study of workers at a petroleum refinery in Torrance, California, 1959 to 1997. J Occup Environ Med. 2001;43:1089–1102.
13.Huebner WW, Wojcik NC, Rosamilia K, Jorgensen G, Milano CA. Mortality updates (1970–1997) of two refinery/petrochemical plant cohorts at Baton Rouge, Louisiana, and Baytown, Texas. J Occup Environ Med. 2004;46:1229–1245.
14.Lewis RJ, Schnatter AR, Drummond I, et al. Mortality and cancer morbidity in a cohort of Canadian petroleum workers. Occup Environ Med. 2003;60:918–928.
15.Huebner WW, Schnatter AR, Nicolich MJ, Jorgensen G. Mortality experience of a young petrochemical industry cohort: 1979–1992 follow-up study of U.S.-based employees. J Occup Environ Med. 1997;39:970–982.
16.Huebner WW, Wojcik NC, Jorgensen G, Nicolich MJ, Marcella SP. Mortality patterns and trends among 127,266 U.S.-based men in a petroleum company: update 1979–2000. J Occup Environ Med. 2009;51:1333–1348.
17.World Health Organization (WHO). Ninth Revision of the International Statistical Classification of Diseases, Injuries, and Causes of Death. Vol 1. Geneva: World Health Organization; 1977.
18.World Health Organization (WHO). International Statistical Classification of Diseases and Related Health Problems: 10th Revision. Geneva: World Health Organization; 1992.
19.Marsh GM, Youk AO, Stone RA, Sefcik S, Alcorn C. OCMAP-PLUS, a program for the comprehensive analysis of occupational cohort data. J Occup Environ Med. 1998;40:351–362.
20.Chemical Manufacturers Association Epidemiology Task Group. Guidelines for good epidemiology practices for occupational and environmental epidemiologic research. J Occup Med. 1991;33:1221–1229.
21.Tsai SP, Wendt JK, Hunter RB. Trends in cigarette smoking among refinery and petrochemical plant employees with a discussion of the potential impact on lung cancer. Int Arch Occup Environ Health. 2001;74:477–482.
22.Divine BJ, Hartman CM, Wendt JK. Update of the Texaco mortality study 1947–93: Part I. Analysis of overall patterns of mortality among refining, research, and petrochemical workers. Occup Environ Med. 1999;56:167–173.
23.Bond GG, McLaren EA, Cartmill JB, Wymer KT, Lipps TE, Cook RR. Mortality among female employees of a chemical company. Am J Ind Med. 1987;12:563–578.
24.Burns CJ, Carson ML, Cartmill JB. Cause-specific mortality among employees of a Texas-based chemical manufacturing facility, 1940 through 1996. Tex Med. 2003:67–73.
25.Gunnarsdottir H, Rafnsson V. Mortality among female manual workers. J Epidemiol Comm Health. 1992;46:601–604.
26.Schottenfeld D, Fraumeni JF. Cancer Epidemiology and Prevention. New York: Oxford University Press; 2006.
27.Steenland K, Hu S, Walker J. All-cause and cause-specific mortality by socioeconomic status among employed persons in 27 US states, 1984–1997. Am J Pub Health. 2004;94:1037–1042.
28.Watkins DK, Chiazze L, Fryar C. Historical cohort mortality study of a continuous filament fiberglass manufacturing plant: II. Women and minorities. J Occup Med. 1997;39:548–555.
29.Steenland K, Halperin W, Hu S, Walker JT. Deaths due to injuries among employed adults: the effects of socioeconomic class. Epidemiology. 2003;14:74–79.
30.Greenwald P, Friedlander BR, Lawrence CE, Hearne T, Earle K. Diagnostic sensitivity bias: an epidemiologic explanation for an apparent brain tumor excess. J Occup Med. 1981;23:690–694.
31.Baillargeon J. Characteristics of the healthy worker effect. Occup Med. 2001;16:359–366.
32.Checkoway H, Pearce N, Kriebel D. Research Methods in Occupational Epidemiology. New York: Oxford University Press; 2004.
33.Choi BCK. Definition, sources, magnitude, effect modifiers and strategies of reduction of the healthy worker effect. J Occup Med. 1992;34:979–988.
34.Lea CS, Hertz-Picciotto I, Andersen A, et al. Gender differences in the healthy worker effect among synthetic vitreous fiber workers. Am J Epidemiol. 1999;10:1099–1106.
35.Howe GR, Chiarelli AM, Lindsay JP. Components and modifiers of the Healthy Worker Effect: evidence from three occupational cohorts and implications for industrial compensation. Am J Epidemiol. 1988;128:1364–1375.
36.Gridley G, Nyren O, Dosemeci M, et al. Is there a healthy worker effect for cancer incidence among women in Sweden? Am J Ind Med. 1999;36:193–199.