Lifetime never smokers with lung cancer likely represent a genetically susceptible subgroup distinct from ever smokers with lung cancer.1 Familial aggregation may provide indirect evidence for a role of genetic factors in lung cancer susceptibility. The majority of studies that have evaluated familial aggregation in never-smoking lung cancer patients reported increased but not statistically significant lung cancer risk.2, 3, 4, 5, 6, 7, 8, 9 Several recent studies that analyzed aggregation of lung cancer in relatives of lung cancer patients reported no increased risk in first-degree relatives of never-smoking lung cancer cases.8, 9 However, one study suggested a significant 6-fold increased lung cancer risk in relatives of younger (40–59 years old) nonsmoking lung cancer patients, yet based on a small number of individuals (47 lung cancer cases and 50 controls).3 Only a few studies have evaluated occurrence of other specific cancers in relatives of never-smoking lung cancer patients.7, 10, 11
We estimated the risk of lung cancer and other cancers in first-degree relatives of never-smoking lung cancer cases compared with first-degree relatives of matched controls, using self-reported data on family history derived from an ongoing lung cancer case–control study, controlling for relatives' smoking status as reported by the proband.
Material and methods
From September 1995 through December 2003, patients with lung cancer were accrued for an ongoing and previously described molecular epidemiological study on susceptibility markers for lung cancer from The University of Texas M. D. Anderson Cancer Center.8, 12, 13 There were no age, gender, ethnic or stage restrictions. There were 316 patients with histologically confirmed lung cancer, who reported themselves to be lifetime never smokers (defined as those who had smoked less than 100 cigarettes in their lifetime). The response rate for cases was about 80%. The reasons for refusal to participate included patient too ill, patient referred only for second opinion to M. D. Anderson Cancer Center or patient unwilling to donate blood for the study and complete the interview. Healthy controls who were also lifetime never smokers (n = 318) without a previous diagnosis of cancer (except for nonmelanoma skin cancer) were recruited from the Kelsey–Seybold Clinics, Houston's largest private multispecialty physician group, that includes a network of 23 clinics and more than 300 physicians in the Houston metropolitan area. Patients arriving at the clinics were given a short survey form to determine their eligibility for the study. The completion of the form was strictly voluntary. Based on the survey forms, individuals most suitable for frequency matching to the recruited cases were identified and contacted to schedule the interview and specimen collection. The matching criteria included age (±5 years), gender and ethnicity. The response rate for controls was about 75% when approached for an interview. The main reasons for declining participation included lack of time or difficulties related to transportation. All cases and controls were U.S. residents. This research was approved by the M. D. Anderson Cancer Center and Kelsey–SeyboldInstitutional Review Boards.
Collection of epidemiological data
After study participants were briefed on the study and signed an informed consent, a 45-min structured personal interview was conducted by M. D. Anderson research interviewers, during which they obtained information on socio-demographic characteristics, lung cancer risk factors and family history. Family history included cancer histories of all first-degree relatives (parents, siblings, and offspring), their year of birth, age at the time of study or at death and smoking status (yes or no), collected for each relative, as well as type of cancer and age at diagnosis for affected relatives.
Descriptive statistical analyses were performed to compare demographic characteristics of the lung cancer cases and controls and their relatives. To evaluate whether there was an excess risk of cancer among the first-degree relatives of cases, we performed unconditional logistic regression using generalized estimating equations, which accounts for relatedness within families, to obtain odds ratios (OR) and 95% confidence intervals, treating cancer status among first-degree relatives as the outcome. Although probands were not included in the analyses, we controlled all analyses for age and gender of the proband and of the relatives, ethnicity of the proband, type of relationship to the proband, smoking status and birth cohort of the relative, as appropriate. We adjusted the estimates for birth cohorts defined as born: (i) in or before 1900; (ii) between 1901 and 1940; (iii) between 1941 and 1960 and (iv) in or after 1961. This stratification was chosen so as to approximately reflect prevailing cigarette smoking patterns.14 For relatives with multiple reported cancers, only the first primary cancer was analyzed, since distinguishing between metastatic and second primary tumors was not possible without validation.
In the analyses of young onset cancers, relatives were treated as ‘affected’ if they were diagnosed with cancer before age 50; otherwise they were treated as “unaffected.” Excluding relatives with later-onset cancers from the “unaffected” did not materially change the estimates, which are thus not shown. Likewise, in the site-specific analyses, relatives were treated as “affected” if they had the specific cancer; otherwise they were considered “unaffected.” Again, when relatives with other cancers were excluded from the “unaffected” group, the risk estimates remained virtually identical (not shown). Kaplan–Meier survival analysis was used to estimate the cumulative risk of lung cancer in case and control relatives, and both Kaplan–Meier survival and Cox proportional hazards model analyses were used to evaluate the difference in time to cancer diagnosis between case and control relatives and between relatives of young and older onset cases. In the proportional hazards analysis, we verified the assumptions of the proportional hazards model and also applied a robust variance correction to allow for relatedness among the relatives.
Unless otherwise stated, analyses were performed using the SAS 9.1 statistical software package (The SAS system for Windows Release V9.1, SAS Institute, 2002–2003). The robust variance correction in the proportional hazards modeling was applied using STATA (V8.2).
Characteristics of the probands are summarized in Table I. There was no statistically significant case–control difference in age, gender, and ethnicity. The average age of the cases and controls was 61 years, about two-thirds of both cases and controls were women, and about 80% were Caucasian. There was no case–control difference in the number of reported first-degree relatives per proband (Table I). Information on histology was available for 299 (94.6%) of the patients. The most common cell type diagnosed was adenocarcinoma (64.9%), followed by bronchioalveolar carcinoma/adenocarcinoma with bronchioalveolar carcinoma features (13.0%) and squamous cell carcinoma (8.0%). Small cell carcinoma was rare (found in 2.3% of probands), and nonsmall cell lung cancer not otherwise specified was reported in 11.7% of probands.
Table I. Select Demographic Characteristics of Probands and Their Relatives
Cases were significantly more likely to have 2+ first-degree relatives with cancer compared to controls (OR = 1.76, 95% CI [1.15–2.68], adjusted for the number of smoking relatives and the total number of relatives).
Characteristics of first-degree relatives
Young probands (<50)
Older probands (50+)
The average number of cancer-affected relatives per family was significantly higher for the cases than for the controls (p = 0.006; Table I). Among the cases, 30.7% reported no cancer-affected first-degree relatives, 35.4% reported 1 and 33.9% reported 2 or more such relatives, while for the controls, the corresponding percentages were 37.7, 36.8 and 25.5%. Cases were significantly more likely to have 2 or more relatives with cancer compared with controls [OR = 1.76, 95% CI (1.15–2.68)] (Table I).
Family history data were reported for 2,465 first-degree relatives of the cases and 2,441 first-degree relatives of the controls. The characteristics of the relatives are presented in Table I. There were no statistically significant differences in age or in smoking prevalence between case and control first-degree relatives.
Risk of all cancers
Risk of any cancer was 1.25-fold higher in first-degree relatives of cases compared with relatives of controls [95% CI (1.05–1.50)] (Table II). The risk was elevated, but not statistically significant, in fathers, brothers and sisters, but not mothers of case probands (Table II). Offspring of the cases exhibited a statistically significant 2.05-fold increased risk for any cancer [95% CI (1.03–4.10)]. Smoking relatives of cases had a 36% excess risk of any cancer compared with smoking control relatives [95% CI (1.03–1.81)], while nonsmoking relatives of cases showed no significant risk compared with nonsmoking control relatives [OR = 1.11, 95% CI (0.87–1.42)]. No significant excess risk was observed for relatives of young case probands (age below 50), although there was a significant 24% excess risk [95% CI (1.01–1.53)] in relatives of older case probands. There was a 44% excess risk [95% CI (1.05–1.97)] of young onset cancers (diagnosed at an age less than 50) among case relatives.
Table II. Familial Risks of All Cancers and Lung Cancer
Lung cancer risk was elevated, although nonsignificantly, in fathers, brothers, and sisters, but not in mothers of the cases (Table II). Smoking relatives of cases had a 68% excess risk of lung cancer compared with smoking control relatives, although the estimate did not reach statistical significance [95% CI (0.99–2.63), p = 0.055], whereas no excess lung cancer risk was noted in nonsmoking case relatives. There was no significant excess risk among the relatives of young case probands (age < 50), while a nonsignificant 54% excess risk (0.96–2.45) was noted for relatives of older case probands. There was a strikingly high risk [OR = 5.52, (1.19–25.51)] of young onset lung cancer among relatives of cases compared with relatives of controls. We also evaluated these risks in relatives of lung adenocarcinoma probands separately, and similar results were noted (data not shown).
Risks of other specific cancers
The top 10 cancers that occurred in first-degree relatives of either cases or controls included breast, lung, prostate, colorectal, nonmelanoma skin, melanoma, head and neck, leukemia, lymphoma, stomach, brain, ovarian, thyroid and testicular cancer (Table III). In addition, we also analyzed all smoking-related cancers (lung, head and neck, bladder, kidney and pancreas) as a separate category.
Table III. Ten Most Frequent1 Cancers (Other Than Lung2) and Smoking-Related Cancers Occurring in Relatives
Top ten cancers in case relatives and top ten cancers in control relatives.
Lung cancer is second most common.
Adjusted for ethnicity, gender and age of the proband, age, smoking status, birth cohort of the relative and type of relationship to proband, where appropriate. Sex-specific cancers are analyzed in the appropriate gender only.
Included lung, head and neck, bladder, kidney and pancreatic cancers.
The risk of smoking-related cancers among relatives showed a pattern similar to that for lung cancer, however, none of the risk estimates reached statistical significance (data not shown). The overall risk estimate was 1.27 [0.91–1.77] (Table III). Stratification by smoking status resulted in identical nonsignificant risk estimates (OR for both smoking and nonsmoking relatives was 1.23; Table III).
We noted an elevated 1.58-fold risk [CI (1.04–2.43)] of breast cancer in the first-degree female relatives of cases and a 2.57-fold increased risk in mothers of cases [CI (1.32–5.04)] (Table III).
In addition to breast cancer, significant excess risk in case relatives was found for testicular cancer [OR = 12.32, 95% CI (1.71–88.9)], although based on only 8 case relatives and 1 control relative. The same analysis was performed for young onset (before age 50) cancers in relatives (data not shown). Besides young onset lung cancer (mentioned earlier), only young onset testicular cancer was associated with case status of the proband, with a more than 12-fold risk [OR = 12.98, 95% CI (1.44–121.47)].
Age at cancer onset in relatives
As mentioned earlier, case relatives tended to exhibit an earlier age at diagnosis of lung cancer than control relatives (Table IV). This difference was more pronounced among nonsmoking relatives (60.6 years in case relatives vs.74.2 years in control relatives, p = 0.055) (Table IV). Although there was no difference in relatives' age at lung cancer diagnosis by proband's age at diagnosis, relatives of young onset cases (<50) presented with an earlier age at diagnosis of any type of cancer (53.7 years, SD = 11.7) compared with the relatives of older onset probands (61.4 years, SD = 16; p = 0.0002) or to the relatives of controls (60.7, SD = 14.8; p = 0.003) (Table IV). Also, nonsmoking relatives of younger cases (<50) were diagnosed with any cancer at a significantly earlier age (49.4, SD = 11.3) than nonsmoking relatives of young (<50) controls (58.7, SD = 11.5; p = 0.017) (Table IV). These results were confirmed with Kaplan–Meier survival analysis (Fig. 1a,b withlog-rank p-values). To control for covariates (ethnicity, type of relationship, birth cohort, gender of the proband and of the relative, age of proband, and smoking status, as appropriate), we performed a Cox proportional hazard regression analysis (Table IV), to show that nonsmoking relatives of young onset cases had an increased risk of cancer, compared with nonsmoking relatives of older onset cases [HR = 2.08, 95% CI (1.15–3.76)] and to nonsmoking relatives of controls [HR = 1.90, 95% CI (1.06–3.37)].
Table IV. Age at Cancer Onset in Relatives of Never Smoking Probands by Case-Control Status and by Age-at-Onset of the Case Proband, and Risk of Any Cancer by Proband's Age at Diagnosis, Using Proportional Hazards Model
Type of cancer
Category of relatives
Relatives of cases age ≥50
Relatives of cases age <50
Age at cancer diagnosis in relatives of never smoking probands
The difference in average age of relatives of young (<50) cases and relatives of older cases was significant overall (41.3, SD 20.8, vs. 56.4, SD 20.8; p < 0.001), for smokers (52.7, SD 15.9 vs. 58.8, SD 17.2; p < 0.001), and for nonsmokers (35.2, SD 20.6, vs. 55.1, SD 22.3; p< 0.001).
The difference in average age of relatives of young cases and relatives of young controls was not significant overall (41.3, SD 20.8 vs. 42.0, SD 21.1, p = 0.63), for smokers (50.7, SD 17.5 vs. 52.7, SD 15.9, p = 0.30), and for nonsmokers (35.16, SD 20.61 vs. 36.7, SD 21.5, p = 0.41).
Data for smokers are not significant and not shown; estimates adjusted for ethnicity, type of relationship, birth cohort, gender of proband and of relative and age of proband.
Relatives' risk of any cancer by proband's age at diagnosis3
Cumulative lung cancer risk in case and control relatives
Figure 1c illustrates the cumulative risk of lung cancer in smoking relatives by proband case–control status. Smoking relatives of cases showed a significantly higher cumulative risk of lung cancer by age 85 compared with smoking control relatives: 18 and 13%, respectively (p = 0.0388). However, no difference in cumulative lung cancer risk was found in nonsmoking relatives of cases compared with nonsmoking control relatives (Fig. 1d) (p = 0.8).
We found an aggregation of cancers overall and lung cancer specifically among first-degree relatives of never-smoking lung cancer patients. Case relatives were at a higher risk of young onset lung cancer compared with control relatives, and smoking case relatives had a higher cumulative risk of lung cancer than smoking relatives of controls. An excess of breast cancers was observed among female relatives and especially mothers of cases, and an excess of testicular cancer was noted in male case relatives.
The existence of genetic susceptibility to lung cancer is suggested in a number of familial aggregation studies4, 5, 6, 8, 15, 16, 17, 18, 19 as well as in segregation analyses20, 21, 22, 23, 24 and in a linkage study25 of lung cancer pedigrees. A recent meta-analysis26 of 28 case–control and 4 cohort studies published on familial aggregation of lung cancer, regardless of the smoking status of the proband, suggested a 2-fold increase in lung cancer risk associated with a lung cancer family history.
Familial risks may reflect not only genetic predisposition, but also shared exposures, such as smoking habits,27, 28, 29 or exposure to cooking fumes.7, 30 To address this issue, we (i) included only relatives of never-smoking probands; (ii) took into account smoking habits of the relatives, both by controlling for smoking status and by conducting analyses stratified by smoking status. Potentially, in never-smoking lung cancer probands, relatives who smoke are likely to have exposed the probands to secondhand smoke, putting them at an increased lung cancer risk. In our study, this factor does not seem to play a major role, since a slightly higher percent of smokers was observed among control rather than case relatives, including parents. Thus, it is also unlikely that parental smoking (and thus childhood secondhand smoke exposure) was an appreciable risk factor for lung cancer in this group of never smokers. Interestingly, it was recently suggested that most familial cases of lung cancer could not be attributed to shared smoking habits.31 Thus, genetic factors rather than shared smoking habits are likely to contribute to the observed familial aggregation of cancer in our study.
Only a few studies have focused specifically on cancer/lung cancer family history in lung cancer patients who have never smoked or are long-term quitters, with the same purpose of minimizing the impact of shared smoking habits on familial lung cancer risk (Schwartz et al.,3 257 cases and 277 controls; Wu et al.,7 646 cases and 1,252 controls, all women; Brownson et al.,2 432 cases and 1,168 controls never smokers, all women; Mayne et al.,10 197 cases and 197 controls never smokers; Wu et al.,32 108 cases and 108 controls, all women). Matakidou et al.,26 in a meta-analysis of 11 studies that considered familial lung cancer aggregation specifically in never smokers, reported a 1.51-fold significant risk. However, most of the individual studies reported increased, but not statistically significant lung cancer risks.2, 3, 4, 5, 6, 7, 8, 9 In our study, lung cancer risk in all first-degree case relatives combined also was nonsignificantly elevated (OR = 1.39, p = 0.145).
Younger age at cancer diagnosis may indicate the existence of genetic susceptibility. Several segregation studies have suggested the presence of a major gene affecting lung cancer risk in families with young nonsmoking probands.32, 33 A higher lung cancer risk in relatives of younger lung cancer probands has been documented in case–control studies.3, 7, 16 In particular, Schwartz et al.3 observed that for never smoking probands aged 40–59 years there was a 6.1-fold risk of lung cancer in first-degree relatives. We did not observe an increased lung cancer risk in relatives of young onset probands defined as those diagnosed with lung cancer before age 50. One reason may be that the relatives were still young (41.3 years on average). However, lung cancer occurring at an early age was significantly more frequent among relatives of cases compared with control relatives (OR = 5.52, p = 0.042). Interestingly, we also observed that relatives of younger onset lung cancer probands were diagnosed with any cancer at significantly earlier ages than relatives of later onset lung cancer cases or relatives of controls. This may reflect the predisposition of the former to both young onset lung cancer and to a spectrum of cancers characterized by young onset (e.g. testicular cancer, see below).
We noted about 2-fold nonsignificantly elevated lung cancer risks in fathers, sisters and brothers of cases, while case mothers were, unexpectedly, less likely to be affected with lung cancer compared with mothers of controls. Another observation was a considerable difference in age at lung cancer diagnosis between case mothers (51.9, SD = 11.6) and mothers of controls (74.1, SD = 9.4; p < 0.001; the pattern was identical in smoking and nonsmoking mothers); in particular, there were 4 case mothers and no control mothers with young onset (<50) lung cancer. In other words, fewer case mothers than control mothers developed lung cancer, but those case mothers who did, developed it much earlier. We suggest 2 not mutually exclusive hypotheses to explain these facts. The first is that case mothers are predisposed to lung cancer with early onset. The second is that a competing risk of breast cancer might be a factor underlying the deficit of lung cancers among case mothers. Indeed, many more case mothers, compared with control mothers, developed breast cancer (OR = 2.75, Table III), and the ages at diagnosis with breast cancer were similar in mothers of cases (60.6, SD = 15.8) and controls (62.5, SD = 10.5; p = 0.64). Competing risk might also explain the results of Mayne et al.34 who reported a nonsignificantly decreased lung cancer risk in case mothers (OR = 0.67) (or no excess maternal lung cancer risk in a cohort analysis), while breast cancer risk was significantly elevated [OR = 2.00, 95% CI (1.00–4.00)]. Fathers, as in our study, were at a nonsignificantly increased 1.87-fold lung cancer risk. In both our and Mayne et al's study there was no difference between cases and controls in the percent of smoking fathers or mothers.
We observed a 25% excess risk of any cancer in relatives of lung cancer cases. Similar findings have been reported by some investigators6, 8, 32, 35, 36 but not others.7, 15 As mentioned earlier, several studies reported excess risks of breast cancer in relatives of nonsmoking lung cancer patients8, 10, 11, 37 and colon, digestive or aerodigestive cancers.32, 33, 37 Etzel et al.8 in an analysis of a subset of our study subjects observed a strong significant 5-fold increased breast cancer risk in mothers of never-smoking lung cancer cases. In our analysis the number of never smokers has substantially increased and the association with maternal breast cancer was still observed, although decreased in magnitude to about 2.5-fold; for all female case relatives combined there was a 58% significant excess breast cancer risk.
Few studies have reported a familial association of lung cancer with testicular cancer. This may be due to the fact that most studies considered “all cancers combined,” “smoking-related cancers,” “nonlung cancers” or different groupings of cancers. Here we selected only the 10 most frequent cancers reported either among case or control relatives. Although testicular cancer occurred in only 1 control relative (who was the brother of a control but had a father with lung cancer), it was the ninth most common cancer in case relatives. Hemminki and Li38 reported that testicular cancer was significantly associated with a parental history of lung cancer (with no indication of the smoking status of the affected parent) and several other cancers (including breast cancer). Bromen et al.39 found an increased (close to 10-fold) breast cancer risk in sisters of testicular cancer patients. There may exist shared genetic pathway(s), e.g. carcinogen metabolic,40, 41 DNA repair,42, 43, 44, 45, 46 or estrogen-receptor-signaling pathways47, 48, 49, 50 that contribute to evidence of familial aggregation of cancer (such as lung, breast and testicular cancers).
Limitations of the study
This study has several limitations. Family history was self-reported, and verification of the diagnosis through medical record review was not performed. However, nondifferential misclassification of familial cancer would usually lead to an underestimation of the true risk and would therefore not explain the risk elevation noted. Studies have validated patient report of family history by medical records51 or through comparison with expected numbers of cancers based on population incidence.52 Bondy et al.51 have shown that subjects correctly identified the primary site of cancer for 88% of cases in first-degree relatives and for lung cancer in 85%. Pinsky et al.52 observed that the ratio of reported to expected number of cancers in family members was about 0.7 (30% underreporting), with female probands tending to underreport less often than male probands. In our study, most of the probands were women, which possibly make the problem of underreporting less severe. Also, the probands in our study are gender-matched, and in all stratified analyses the gender of the proband was controlled for, which should eliminate or reduce any potential bias caused by gender differences in reporting.
Another limitation of our study was that 52 cases (about 16%) had a previous cancer other than nonmelanoma skin cancer. Since a history of a prior cancer (other than nonmelanoma skin cancer) was an exclusion criterion for controls, this may have biased our estimates away from the null. We reran our analyses after exclusion of cases with prior cancer and did not note any material changes from the presented results.
Most case probands in our study presented with adenocarcinoma or bronchioalveolar carcinoma, while small cell lung cancer, being strongly associated with smoking,53, 54 was very rare in this group of never smokers. Thus, our study provides little information on familial aggregation of cancer in relatives of small cell lung cancer patients who never smoked. Another limitation is that we used the ethnicity of probands as a proxy for the ethnicity of the relatives, since the latter was not reported.
The degree of missing data on characteristics of the relatives was relatively low: less than 1% for gender and presence of a cancer diagnosis, and less than 3% for relatives' age and smoking status. Among the relatives with cancer, information was missing on age in 1% of control relatives and 3.8% of case relatives, on smoking status in no control relatives and in 4.8% of case relatives, on cancer site in 3.8% of control relatives and 3.3% of case relatives, and on age at diagnosis in 7% of control relatives and 12% of case relatives. Individuals with missing information were excluded from the analyses. Since a slightly higher number of case than control relatives with cancer were thus excluded, our estimates are likely to be conservative.
Recall bias is possible since diagnosis of (lung) cancer in a proband might result in an increased knowledge or awareness of cancer in the patient's relatives.
When controlling for smoking, we only adjusted for the smoking status of relatives in our study, instead of smoking intensity and duration, which are likely to be inaccurately reported and could lead to misclassification.55 Thus, for smoking related cancers, our results might be inflated should case relatives smoke more than control relatives. Nothing in our data, however, indicates that this should be the case. On the contrary, the prevalence of smoking was slightly higher in the relatives of controls compared with the relatives of cases. Another limitation, inherent to a case–control study, is our inability to fully evaluate cancer risk in relatives who are still young, for example in sisters, brothers and especially offspring or relatives of young probands.
In conclusion, our analysis shows that shared genetic factors may affect susceptibility to lung cancer and several other cancers. These factors tend to manifest themselves at a relatively young age. Further studies of genetic determinants of lung cancer in never smokers, as well as of common genetic pathway(s) for cancers showing familial aggregation are warranted.
National Cancer Institute grant R01 CA55769 (Dr. M.R. Spitz, Dr. C.I. Amos), Flight Attendant Medical Research Institute YCSA grant (Dr. O.Y. Gorlova) and a Flight Attendant Medical Research Institute professorship (Dr. M.R. Spitz).