The findings from our systematic review and meta-analysis of the published literature on familial aggregation of lung cancer are consistent with a two-fold increase associated with family history with evidence of risk being related to early age of diagnosis and number of relatives affected.
The interpretation of these studies requires caution: while familial risks are compatible with genetic predisposition, they could reflect common exposures. Smoking is the most important environmental risk factor of lung cancer, and the association between a person's smoking habits and that of his parents or siblings has been well documented (Salber and Macmahon, 1961
). Unless adjustment is made for smoking habits, an above-expected incidence of lung cancer in relatives of lung cancer patients may be found, in the absence of any genetic effect. To date only four investigators () have attempted to address this issue by taking into account the smoking habits of both the study subjects and their family members, reporting RRs comparable with those in studies making no such adjustment.
To minimise the impact of shared smoking habits in families, a number of studies have estimated familial risks associated with nonsmoker status (). Pooling of the data in never-smokers resulted in an elevated risk of lung cancer associated with a family history of the disease that was statistically significant, supporting the view that genetic or other environmental factors may play a role in familial aggregations.
The contribution of shared environmental risk factors to familial lung cancer risk may also be assessed through risk estimation associated with an affected spouse since concordance of smoking habits between spouse pairs has been reported (Macken et al, 2000
). Indeed, risk was significantly elevated in probands with an affected spouse, but remained lower than the risk associated with an affected relative, consistent with possible genetic factors.
Cohort studies of twins are classically used to separate genetic and environmental influences on familial aggregation of a disease. A critical assumption is that MZ and DZ twins display a comparable degree of similarity because of shared environmental factors, so that any difference in concordance rates only reflects genetic factors. The reported concordance ratios of lung cancer among male twins are almost equal, suggesting a strong environmental effect shared by twins (i.e. smoking behaviour) rather than a genetic component, which was widely cited to counter the propositions that an inherited basis exists for lung cancer or that the predisposition to smoke was itself genetic. Twin studies have, however, consistently shown greater concordance for smoking in MZ than DZ twins (Carmelli et al, 1992
), suggesting that environmental exposure is being confounded by genetic influence. Yet, paradoxically, this concordance difference in smoking behaviour is not reflected in a concordance difference for lung cancer, although in female twins, where the prevalence is much lower, it did appear to follow a more conventional genetic pattern with risks in MZ being greater than in DZ twins, pointing to genetic predisposition (Lichtenstein et al, 2000
One caveat to our meta-analysis is the significant heterogeneity observed between studies, although its impact on summary risk estimates is difficult to assess. Given the differences in location, design and control selection of the various studies, some degree of heterogeneity may be expected. Some of it is also likely to reflect differences in statistical methodology between studies, particularly in the adjustment for smoking habits. The presence or absence of adjustment for the smoking habits of study participants or their relatives did not appear to impact significantly on the results of our meta-analysis, although when adjustment was performed there was a trend towards reporting lower RR. A further issue inherent in many case–control studies is that of recall bias. The diagnosis of lung cancer in an individual may bring to light knowledge or awareness of lung cancer in relatives. Bias from this source can be eliminated by collecting the family history data before diagnosis (prospective/cohort study design). Alternatively, verification of cancer or cause of death among relatives from medical records or death certificates will eliminate recall bias. Where possible, we examined the impact of such verified data and noted that such studies reported higher rather than lower RRs; support that recall bias is unlikely to represent a significant confounder.
The only characteristic found to significantly impact on the heterogeneity observed between studies was the date of study publication. Studies published before 1993 reported higher RRs of lung cancer associated with positive family history, indicating time lag bias and possibly publication bias. However, formal testing showed no evidence of publication bias between case–control or cohort studies. Further statistical analysis of studies published before and after 1993 showed adjustment for family size to be a significant confounder. Individuals with large families are more likely to have an affected relative than those with small families; where average family size differs between cases and controls, failure to adjust for this might inflate the reported RR, as observed in the earlier studies. Univariate regression analysis of all the studies for the presence or absence of adjustment for family size did not, however, appear to account for the heterogeneity observed between studies, making it unlikely to significantly impact on the combined RR.
Type of control, type of relative studied and gender of participants were examined for their effect on the summary statistics with no significant associations detected. Although there were indications that some of these may have contributed to heterogeneity, each study possessed different combinations of both desirable and undesirable methodological features, such that no single factor, other than publication year, consistently increased or decreased RRs. Sample size limitations prevented detailed multivariate analysis, so that other important sources of heterogeneity may have become apparent if appropriate adjustment for confounding had been possible.
In summary, this systematic review finds a significant increase of lung cancer risk associated with having an affected relative, the risk being further increased with earlier age of onset of the disease and with multiple affected family members. This suggests that lung cancer risk may be in part genetically determined. However, familial studies of lung cancer are problematic as they display high heterogeneity and it is usually impossible to make a suitable adjustment for smoking, the major risk factor. Furthermore, the twin studies and the elevated lung cancer risk associated with an affected spouse do not favour a genetic susceptibility. Such limitations formally preclude the drawing of strong inferences about any genetic influences on lung cancer outside the context of rare Mendelian disorders. Ultimately, verification of a genetic predisposition must come from the identification of causal mutations. Recently, following a genomewide linkage scan, a candidate locus for lung cancer predisposition has been reported (Bailey-Wilson et al, 2004
). If confirmed, this would provide the most convincing evidence to date of a genetic susceptibility outside rare Mendelian disorders.