Genetic Epidemiology 8: 199-208 (1991)
Association of Incident Lung Cancer With Family History of Female Reproductive Cancers: The Iowa Women’s Health Study Thomas A. Sellers, John D. Potter, and Aaron R. Folsom
Division of Epidemiology, School of Public Health (T.A.S., J. D.P., A. R.F.), and the Institute of Human Genetics (T.A. S.), University of Minnesota, Minneapolis, Minnesota A number of studies have documented the familial aggregation of lung cancer; there is at least one report that female reproductive cancers are also increased in these families. To determine if the risk exists for all reproductive cancer sites, we conducted a nested case-control study of lung cancer incidence in a cohort of 41,837 women ages 55-69 years. Women were recruited by mail and asked to provide information on education, occupation, smoking habits, physical activity, and family history of specific cancer sites among female relatives. Four year follow-up for cancer incidence was conducted using a state-wide tumor registry. Compared to random controls (n = 1900),cases ( n = 152) were more likely to have reported at baseline a sister affected with cancer of the uterus [crude odds ratio (OR) = 3.4, 95% C1 = I.7-7.0, P < 0.011, cervix (OR = 3.2,95% CI 1.2-8.6, P < 0.05), or cancer at any site (OR = 1.6, 95% C1 1.1-2.4, P < 0.05). A family history of an affected mother with a female reproductive cancer was also more common among the cases, but not statistically significant. Cases were less educated, more likely to work in a technicaliindustrial setting, less physically active, more likely to smoke, and to smoke for a longer period of time than the controls (all P < 0.01). These differences reduced the magnitude of the family history risk indicators; only the combined category of reproductive cancer at all sites among sisters remained statistically significant. Additional family studies should be done to assess environmental factors in the relatives of the cases and controls to disentangle the influence of shared genes and shared environmental factors in these associations. Key words: lung cancer, family history, case-control, reproductive cancers
Received for publication October 29, 1991; revision accepted May 14, 1991 Address reprint requests to Dr. Thomas A . Sellers, Division of Epidemiology, Health Sciences Unit A, 1-210 Moos Tower, 515 Delaware Street S.E., Minneapolis, MN 55455.
0 1991 Wiley-Liss, Inc.
200
Sellers et al.
INTRODUCTION
The existence of families with multiple members affected with lung cancer [Goffman et al., 1982; Brisman et al., 1967; Joishy et al., 19771 is consistent with the hypothesis, first suggested almost 40 years ago [Parnell, 1951; Goodhart, 1959, that genetically determined susceptibility factors may be of pathogenic importance. Epidemiologic studies have shown that occupational exposures and smoking habits cannot completely account for these familial clusterings [Tokuhata and Lilienfeld, 1963; Tokuhata, 1964; Ooi et al., 1986; Wu et al., 1988; Sellers et al., 19881. Metabolic studies have revealed interindividual differences in response to carcinogens [Heighway et al., 1986; Roots et al., 1988; Ayesh et al., 1984; Kellerman et al., 1973; Kouri et al., 19821, cytogenic studies provide evidence that a candidate susceptibility gene for lung cancer is likely to reside on chromosome 3p21 [Whang-Peng et al., 1982; Mooibrack et al., 1987; Brauch et al., 1987; Kok et al., 19871, and segregation analyses have provided evidence that susceptibility may be inherited in Mendelian fashion [Sellers et al., 19901. The increased occurrence of malignancies among relatives of lung cancer patients does not appear to be limited to lung cancer alone; an increased risk of other respiratory system cancers has been consistently observed [Tokuhata and Lilienfeld, 1963; Tokuhata, 1964; Ooi et al., 19861 and an increased risk at all sites has been occasionally observed [Lynch et al., 1986; Samet et al., 1986; Sellers et al., 1987; McDuffie et al. 19891. The impetus for the present study was a report of a 2-fold excess familial risk of female reproductive tract cancers in families with lung cancer clustering [Sellers et al., 19871. Because of small numbers of affected relatives, it could not be determined whether this risk was for cervical cancer, which is also known to be smoking related [Brinton et al., 19861, or cancer of the ovaries, uterus, or endometrium, which are considered to be under hormonal influence [Miller, 19781. Glucocorticoids and estrogens regulate mammalian lung differentiation and maturation [Farrell, 1977; Mendelson et al., 1982; Torday et al., 19751. High affinity receptors for these hormones are present in histologically diverse bronchogenic carcinomas [Chaudhuri et al., 1982; Jones et al., 1984; Kobayashi et al., 1982; Liu et al., 19801, and women receiving hormone replacement therapy have demonstrated an increased risk of lung cancer [Adami et al., 19891. Hormones may thus be important in the etiology of lung cancer. Because hormones contribute to the risk of cancers of the ovary and uterus [Miller, 1978; Henderson et al. 19821, the clustering of female noncervical reproductive cancers and lung cancer in some families seems biologically plausible, perhaps representing some shared pattern (kind, interval, duration, intensity) of hormonal exposure. The present report utilizes data from the Iowa Women’s Health Study to determine if (1) the presence of an affected first-degree relative with female reproductive cancer is a risk indicator for lung cancer, and (2) the risk can be partitioned according to the anatomic site of the malignancy. METHODS
Study Population The Iowa Women’s Health Study is a prospective cohort study of a sample of 5 5 to 69-year-old women who had a valid Iowa driver’s license in 1985. A questionnaire was mailed in January 1986 to 99,826 women, to which 41,837 responded. Excluding 1,796 who were ineligible (deceased or wrong age or sex), this represents a response
Familial Reproductive Cancers and Lung Cancer Risk
201
rate of 42.7%. Based on 1980 census data, responders were about 3 months older than nonresponders and were slightly more likely to live in rural, less-affluentcounties [Folsom et al., 19901. Data collection. Self-reported data were collected on the history of cancers of the breast, ovary, uterus, or endometrium (“lining of the womb”), cervix, “female reproductive organs, but specific site of the cancer is unknown,” or “cancer of another site” among mothers, grandmothers, aunts, sisters, and daughters of respondents. For mothers, grandmothers and daughters the specific “cancer of another site” was queried. Data were not collected on family size, ages at onset, or cancer in male relatives. No attempt was made to validate cancers in family members. Among other self-reported items on the 16-page questionnaire were sociodemographic variables (including race, occupation, education, and residence), smoking habits, gynecologic history, leisure-time physical activity, and other medical conditions or diseases. Physical activity was assessed by asking whether participants undertook any leisure exercise and, if so, the frequency of (defined) moderate intensity and heavy intensity activities. The frequency-intensity questions were combined to create a three level physical activity score (low, medium, high). Follow-up for cancer occurrence in the cohort was performed using the State Health Registry of Iowa, part of the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) program. Incident cases of lung cancer (ICD 162) were identified through a computer program that matched 1986-1989 Registry cases and study participants on name, zip code, birthdate, and social security number. All but one cancer was confirmed by microscopic examination. Data analysis. Data were analyzed as a nested case-control study [Mantel, 19731, an efficient approach when the person-years contributed by noncases is very large. Controls ( n = 1900) were randomly selected from the women who did not develop lung cancer during the follow-up period. Frequency distributions of the study variables were examined for cases and controls. The difference in mean age was tested for significance using a two-sided t test. To determine whether cases and controls differed with respect to sociodemographic variables, contingency table chi-square tests were used. Design variables were assigned as 1 for the presence and 0 for the absence of a positive family history (i.e., at least one affected first-degree female relative) of the following cancers among the cases and controls: ovarian, uterine, cervical, reproductive organ-site unknown, or cancer at any site. To refine the source of the familial risk, similar design variables were created to represent the occurrence of cancer at the same sites by generation, i.e., an affected mother, at least one affected sister, or at least one affected daughter. Unconditional logistic regression analysis was used to predict case or control status based on these design variables, each family history variable being evaluated independently in separate models. To determine whether the effect of a positive family history could be mediated by differences in personal characteristics between the cases and controls, additional logistic models were constructed with the family history design variables and the following covariates: education (< high school versus 3 high school), packyears of cigarette consumption (denoted by three indicator variables to represent 0, >O to 19, 20 to 39, and >40 packyears), residence (urban versus rural), occupation (industrial/technical versus all others), and physical activity (low or medium versus high). Adjusted odds ratios and corresponding 95% confidence
202
Sellers et al.
intervals were then computed to describe the effect of variables found to be statistically significant at the 0.05 level in the logistic models. RESULTS A total of 179 women developed lung cancer during approximately 4 years of follow-up. Twenty-seven reported a prior primary malignancy (8 breast, 7 lung, 3 endometrium, 3 cervix, 2 lymphoma, 1 larynx, 1 large bowel, 1 uterine, 1 multiple) and were excluded because of the potential diagnostic confusion between recurrent malignancy, metastases, or second primaries. Thus, a total of 152 incident cases were available for analysis. The distribution by cell type was adenocarcinoma (35%), small cell (28%), large cell (19%), and squamous (18%). The mean age was 62.4 years among the cases and 61.7 years among the controls. Cases were more likely to smoke, and those who did smoke reported greater packyears of tobacco exposure than controls (Table I). Cases were less educated, more likely to work in a technical or industrial occupation, more likely to reside in urban areas with greater than 2,500 population, and more sedentary than the controls. Multiple logistic regression models were fitted to evaluate which of the variables TABLE I. Summary Characteristics of Incident Lung Cancer Cases and Controls: The Iowa Women's Health Study, 1986-1989 Cases = 152)
(n Variable Smoking status Never Former Current Pack years' None >Oto 19 20 to 39 240
Education attained >High school
Association of Incident Lung Cancer With Family History of Female Reproductive Cancers: The Iowa Women’s Health Study Thomas A. Sellers, John D. Potter, and Aaron R. Folsom
Division of Epidemiology, School of Public Health (T.A.S., J. D.P., A. R.F.), and the Institute of Human Genetics (T.A. S.), University of Minnesota, Minneapolis, Minnesota A number of studies have documented the familial aggregation of lung cancer; there is at least one report that female reproductive cancers are also increased in these families. To determine if the risk exists for all reproductive cancer sites, we conducted a nested case-control study of lung cancer incidence in a cohort of 41,837 women ages 55-69 years. Women were recruited by mail and asked to provide information on education, occupation, smoking habits, physical activity, and family history of specific cancer sites among female relatives. Four year follow-up for cancer incidence was conducted using a state-wide tumor registry. Compared to random controls (n = 1900),cases ( n = 152) were more likely to have reported at baseline a sister affected with cancer of the uterus [crude odds ratio (OR) = 3.4, 95% C1 = I.7-7.0, P < 0.011, cervix (OR = 3.2,95% CI 1.2-8.6, P < 0.05), or cancer at any site (OR = 1.6, 95% C1 1.1-2.4, P < 0.05). A family history of an affected mother with a female reproductive cancer was also more common among the cases, but not statistically significant. Cases were less educated, more likely to work in a technicaliindustrial setting, less physically active, more likely to smoke, and to smoke for a longer period of time than the controls (all P < 0.01). These differences reduced the magnitude of the family history risk indicators; only the combined category of reproductive cancer at all sites among sisters remained statistically significant. Additional family studies should be done to assess environmental factors in the relatives of the cases and controls to disentangle the influence of shared genes and shared environmental factors in these associations. Key words: lung cancer, family history, case-control, reproductive cancers
Received for publication October 29, 1991; revision accepted May 14, 1991 Address reprint requests to Dr. Thomas A . Sellers, Division of Epidemiology, Health Sciences Unit A, 1-210 Moos Tower, 515 Delaware Street S.E., Minneapolis, MN 55455.
0 1991 Wiley-Liss, Inc.
200
Sellers et al.
INTRODUCTION
The existence of families with multiple members affected with lung cancer [Goffman et al., 1982; Brisman et al., 1967; Joishy et al., 19771 is consistent with the hypothesis, first suggested almost 40 years ago [Parnell, 1951; Goodhart, 1959, that genetically determined susceptibility factors may be of pathogenic importance. Epidemiologic studies have shown that occupational exposures and smoking habits cannot completely account for these familial clusterings [Tokuhata and Lilienfeld, 1963; Tokuhata, 1964; Ooi et al., 1986; Wu et al., 1988; Sellers et al., 19881. Metabolic studies have revealed interindividual differences in response to carcinogens [Heighway et al., 1986; Roots et al., 1988; Ayesh et al., 1984; Kellerman et al., 1973; Kouri et al., 19821, cytogenic studies provide evidence that a candidate susceptibility gene for lung cancer is likely to reside on chromosome 3p21 [Whang-Peng et al., 1982; Mooibrack et al., 1987; Brauch et al., 1987; Kok et al., 19871, and segregation analyses have provided evidence that susceptibility may be inherited in Mendelian fashion [Sellers et al., 19901. The increased occurrence of malignancies among relatives of lung cancer patients does not appear to be limited to lung cancer alone; an increased risk of other respiratory system cancers has been consistently observed [Tokuhata and Lilienfeld, 1963; Tokuhata, 1964; Ooi et al., 19861 and an increased risk at all sites has been occasionally observed [Lynch et al., 1986; Samet et al., 1986; Sellers et al., 1987; McDuffie et al. 19891. The impetus for the present study was a report of a 2-fold excess familial risk of female reproductive tract cancers in families with lung cancer clustering [Sellers et al., 19871. Because of small numbers of affected relatives, it could not be determined whether this risk was for cervical cancer, which is also known to be smoking related [Brinton et al., 19861, or cancer of the ovaries, uterus, or endometrium, which are considered to be under hormonal influence [Miller, 19781. Glucocorticoids and estrogens regulate mammalian lung differentiation and maturation [Farrell, 1977; Mendelson et al., 1982; Torday et al., 19751. High affinity receptors for these hormones are present in histologically diverse bronchogenic carcinomas [Chaudhuri et al., 1982; Jones et al., 1984; Kobayashi et al., 1982; Liu et al., 19801, and women receiving hormone replacement therapy have demonstrated an increased risk of lung cancer [Adami et al., 19891. Hormones may thus be important in the etiology of lung cancer. Because hormones contribute to the risk of cancers of the ovary and uterus [Miller, 1978; Henderson et al. 19821, the clustering of female noncervical reproductive cancers and lung cancer in some families seems biologically plausible, perhaps representing some shared pattern (kind, interval, duration, intensity) of hormonal exposure. The present report utilizes data from the Iowa Women’s Health Study to determine if (1) the presence of an affected first-degree relative with female reproductive cancer is a risk indicator for lung cancer, and (2) the risk can be partitioned according to the anatomic site of the malignancy. METHODS
Study Population The Iowa Women’s Health Study is a prospective cohort study of a sample of 5 5 to 69-year-old women who had a valid Iowa driver’s license in 1985. A questionnaire was mailed in January 1986 to 99,826 women, to which 41,837 responded. Excluding 1,796 who were ineligible (deceased or wrong age or sex), this represents a response
Familial Reproductive Cancers and Lung Cancer Risk
201
rate of 42.7%. Based on 1980 census data, responders were about 3 months older than nonresponders and were slightly more likely to live in rural, less-affluentcounties [Folsom et al., 19901. Data collection. Self-reported data were collected on the history of cancers of the breast, ovary, uterus, or endometrium (“lining of the womb”), cervix, “female reproductive organs, but specific site of the cancer is unknown,” or “cancer of another site” among mothers, grandmothers, aunts, sisters, and daughters of respondents. For mothers, grandmothers and daughters the specific “cancer of another site” was queried. Data were not collected on family size, ages at onset, or cancer in male relatives. No attempt was made to validate cancers in family members. Among other self-reported items on the 16-page questionnaire were sociodemographic variables (including race, occupation, education, and residence), smoking habits, gynecologic history, leisure-time physical activity, and other medical conditions or diseases. Physical activity was assessed by asking whether participants undertook any leisure exercise and, if so, the frequency of (defined) moderate intensity and heavy intensity activities. The frequency-intensity questions were combined to create a three level physical activity score (low, medium, high). Follow-up for cancer occurrence in the cohort was performed using the State Health Registry of Iowa, part of the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) program. Incident cases of lung cancer (ICD 162) were identified through a computer program that matched 1986-1989 Registry cases and study participants on name, zip code, birthdate, and social security number. All but one cancer was confirmed by microscopic examination. Data analysis. Data were analyzed as a nested case-control study [Mantel, 19731, an efficient approach when the person-years contributed by noncases is very large. Controls ( n = 1900) were randomly selected from the women who did not develop lung cancer during the follow-up period. Frequency distributions of the study variables were examined for cases and controls. The difference in mean age was tested for significance using a two-sided t test. To determine whether cases and controls differed with respect to sociodemographic variables, contingency table chi-square tests were used. Design variables were assigned as 1 for the presence and 0 for the absence of a positive family history (i.e., at least one affected first-degree female relative) of the following cancers among the cases and controls: ovarian, uterine, cervical, reproductive organ-site unknown, or cancer at any site. To refine the source of the familial risk, similar design variables were created to represent the occurrence of cancer at the same sites by generation, i.e., an affected mother, at least one affected sister, or at least one affected daughter. Unconditional logistic regression analysis was used to predict case or control status based on these design variables, each family history variable being evaluated independently in separate models. To determine whether the effect of a positive family history could be mediated by differences in personal characteristics between the cases and controls, additional logistic models were constructed with the family history design variables and the following covariates: education (< high school versus 3 high school), packyears of cigarette consumption (denoted by three indicator variables to represent 0, >O to 19, 20 to 39, and >40 packyears), residence (urban versus rural), occupation (industrial/technical versus all others), and physical activity (low or medium versus high). Adjusted odds ratios and corresponding 95% confidence
202
Sellers et al.
intervals were then computed to describe the effect of variables found to be statistically significant at the 0.05 level in the logistic models. RESULTS A total of 179 women developed lung cancer during approximately 4 years of follow-up. Twenty-seven reported a prior primary malignancy (8 breast, 7 lung, 3 endometrium, 3 cervix, 2 lymphoma, 1 larynx, 1 large bowel, 1 uterine, 1 multiple) and were excluded because of the potential diagnostic confusion between recurrent malignancy, metastases, or second primaries. Thus, a total of 152 incident cases were available for analysis. The distribution by cell type was adenocarcinoma (35%), small cell (28%), large cell (19%), and squamous (18%). The mean age was 62.4 years among the cases and 61.7 years among the controls. Cases were more likely to smoke, and those who did smoke reported greater packyears of tobacco exposure than controls (Table I). Cases were less educated, more likely to work in a technical or industrial occupation, more likely to reside in urban areas with greater than 2,500 population, and more sedentary than the controls. Multiple logistic regression models were fitted to evaluate which of the variables TABLE I. Summary Characteristics of Incident Lung Cancer Cases and Controls: The Iowa Women's Health Study, 1986-1989 Cases = 152)
(n Variable Smoking status Never Former Current Pack years' None >Oto 19 20 to 39 240
Education attained >High school
