Birth Order, Sibship Size, Childhood Environment and Immune-Related Disorders, and Risk of Lymphoma in Lymphoid Cancer Families

Lymphoid cancers are a heterogeneous group of neoplasms that arise from immune cells. Collectively, they represent the fifth highest global incidence of cancer. Established risk factors include older age, male sex, ethnicity, compromised immune function, and family history of lymphoproliferative disorders (LPD; refs. 1, 2). Low-penetrance common genetic polymorphisms that affect pathways related to DNA integrity, B-cell growth, and survival and xenobiotic metabolism have also been implicated (3–5). Early-life environment may modulate risk of immune-related disorders, such as allergies and autoimmune conditions as well as some lymphoid cancers (6).

The hygiene hypothesis proposes that an early life environment that has a relative lack of exposure to microorganisms and infectious disease inhibits a child's immune system from maturing optimally (6). Consequently, such individuals are more susceptible to adult-onset immune-related disorders. Measures of family structure and crowding relate to the hygiene hypothesis as they may affect age and extent of exposure to infectious diseases, with low birth order, and smaller families correlating with higher risk.

Associations between early birth order and/or smaller sibship size and increased risk of lymphoma have been reported for lymphoid cancers collectively (7), and separately for non-Hodgkin lymphoma (NHL; refs. 7–11) and Hodgkin lymphoma (HL; refs. 7, 12–16). However, many other studies report no association between family structure and risk of NHL (10, 16–19), HL (10–12, 16, 19, 20), chronic lymphocytic leukemia (CLL; refs. 16, 17, 21), or multiple myeloma (MM; refs. 7, 16, 19). A few studies have observed a positive association between later birth order and NHL (17, 21, 22), and larger sibship size and risk of NHL (17, 18, 21, 22), and MM (16). The discordant findings among studies may be partly explained by variations in study design, study population, participant response rate, selection bias, hematologic subtypes assessed, and socioeconomic status (SES; ref. 23).

Few studies have examined family structure and environmental factors in the context of multiple-case lymphoid cancer families. Jønsson and colleagues observed a paternal birth order effect among 24 parent–offspring pairs in 32 families enriched for CLL and B-cell malignancies (24). Royer and colleagues (25) found that familial Waldenström macroglobulinemia (WM) cases were more likely to have immune-related disorders such as autoimmune diseases, allergies, and specific infections among 103 familial WM and related B-cell disorders (25).

Currently, there remains a large gap in our understanding of the etiology of familial lymphoid cancers (24). We examined family structure, childhood lifestyle, and immune-related disorders among a large collection of multiple-case lymphoid cancer families, in relation to risk of lymphoid cancer.

This study was approved by the BC Cancer-University of British Columbia Clinical Research Ethics Board. All participants provided written informed consent. Families were eligible for inclusion if they contained a member diagnosed with lymphoma and at least one additional relative with a lymphoid cancer. Patients with the cancers of interest, all of their first-degree relatives, and additional relatives that connect affected family members were invited to participate. Participants were recruited by physician-, self-, or genetic counsellor referral. Family participation was not limited to within Canada, although most families were identified through a member residing in British Columbia, Canada. Families were ascertained between 2006 and 2018.

Information about lymphoid malignancies, family structure, and demographics was obtained systematically using a questionnaire and phone interviews with multiple family members. Family structure and early lifestyle environment information, such as parental education, family income, farm residence, and urban/rural residential location was reported by sibship. Personal information regarding education, medical history (allergies, autoimmune diseases, surgical procedures) and early lifestyle data was obtained from a self-administered questionnaire. Allergies were classified as drug, environmental, or food. Autoimmune diseases were categorized as systemic, organ-specific, or conditions without detectable autoantibodies (25, 26).

We report on 196 families with 524 lymphoid cancer cases among 418 sibships. Of these 418 sibships, 52 lacking family structure (birth order, sibship size) and 17 only-child cases were excluded. The remaining sibships contained 453 cases and 1,112 siblings, from which 3 (0.7%) cases and 94 (8%) siblings were removed due to missing age of enrollment or sex. Analyses were conducted on 450 cases and 1,018 siblings among 346 sibships.

Lymphoid cancer diagnoses were confirmed histologically or through medical records for 241 of the 450 (54%) cases. Cancers were classified according to the InterLymph hierarchical classification of lymphoid neoplasms for epidemiologic research (27).

Our study examined multiple-case families with a history of hematologic malignancies and does not represent a population-based collection.

A χ² goodness-of-fit test was performed to assess whether the observed sex distribution of the families resembled that of the Canadian population (28). American population data were used in instances where distinct histologic subtype information was unavailable (29).

The relationship between lifestyle factors and risk of lymphoid cancer was examined using logistic regression with a generalized estimating equation (GEE) to accommodate correlated family data. ORs and 95% confidence intervals (CI) were clustered by family and adjusted for age (continuous) and sex. Potential confounding effects of ethnicity did not change risk estimates ≥ 10% and were not retained in the final analysis. Independent, exchangeable, and autoregressive correlation structures performed similarly; the autoregressive correlation structure was used in subsequent analyses. Covariates assessed include sex, age of enrollment (n = 1,468), highest level of participant education (n = 494), maternal and paternal education (n = 759 and n = 770, respectively), family income during childhood (n = 756), childhood farm residence (n = 801), childhood residential location (n = 751), allergies (n = 354), asthma (n = 378), autoimmune diseases (n = 378), appendectomy (n = 353), and tonsillectomy (n = 353). Age of death was used in replacement of age of enrollment for nonliving participants. Individuals with missing age, sex, or family structure data were removed from the dataset. Because of their structural dependence, birth order and sibship size were investigated using separate GEE models. Covariates were independently assessed within each GEE model. Statistical analysis was performed using R version 3.5.

Because of the variability between age-of-onset patterns, additional analyses were done with HL cases separated into childhood, young-adult, or adult onset, according to Cozen and colleagues (age-of-diagnosis ≤50, and <40; ref. 30) and Westergaard and colleagues (age-of-diagnosis <15, 15–42, and >42; ref. 15). Results from these sensitivity analyses were essentially identical (Supplementary Table S1) to those of all HL cases together; only outcomes based on all HL cases are presented. Additional analyses were stratified by sibship size (≤3 siblings and >3 siblings) according to Cozen and colleagues (ref. 18; Supplementary Table S2).

The availability of early life and disease information varied because not all members of a given sibship completed the self-administered questionnaire. For example, if one member of a sibship reported the mother's level of education, that datum was applied to all members of the sibship. In contrast, allergy information was only available if it was self-reported. The variables had different amounts of missing data and so they were separated into three groups of comparable sample size to retain the most information. Three GEE models were built in a stepwise manner to investigate the relationships between lymphoid cancer risk and lifestyle factors. Complete family structure (birth order and sibship size), age of enrollment, and sex information was available for 1,468 participants, which constituted the base model. The middle model contained the base model variables and childhood environment variables: maternal education, paternal education, family income, farm residence, and residential location. The full model was comprised of the middle model variables in addition to education, allergies, autoimmune diseases, asthma, appendectomy, and tonsillectomy.

The association between family structure and lymphoid cancer risk was also evaluated using standard logistic regression and χ² tests for trend using permuted pairs of independent cases and controls. Each family member was assigned a generation number relative to the founding lymphoid cancer case or presumed carrier (31). One family (of 196) was excluded from permutations due to an unmatched generation variable. The data set supported the random sampling of 95 generation-matched case/control pairs of independent families, such that a maximum of one individual per family was selected (without resampling). Ninety-five pairs were permuted 10,000 times in quadruplicate for use in: (i) logistic regression with birth order, (ii) logistic regression with sibship size, (iii) a χ² test for trend in proportion with birth order, and (iv) a χ² test for trend in proportion with sibship size. The 10,000 P values and coefficient estimates from each permuted analysis were compared with those observed with the full family dataset. The logistic models contained base model variables (age of enrollment, sex, and birth order or sibship size).

We report on 346 sibships with a lymphoid cancer–affected individual within 196 multiple-case families (Table 1). The median age of enrollment for cases and unaffected siblings was 62 and 63 years, respectively. 241 (54%) cases were confirmed histologically or by medical records, all of which supported the reported diagnosis. Familial cases were 54% male; in comparison, Canadian population NHL, HL, CLL, and MM cases are 55%, 57%, 61%, and 59% male, respectively (28). The sex distributions of familial MM (P = 0.0454) and HL (P = 0.0126) cases were significantly less frequently male than population cases.

Birth order was inversely associated with lymphoid cancer (OR = 0.83; 95% CI, 0.78–0.89), such that earlier birth order has a higher risk of lymphoma (Tables 2 and 3). The ORs were 0.62 (95% CI: 0.41–0.82) for fourth born compared with first-born individuals, and 0.41 (95% CI: 0.30–0.57) for fifth or later born compared with first born. We also observed a strong inverse relationship between sibship size and lymphoid cancer (OR = 0.82; 95% CI, 0.79–0.85); smaller sibships had a higher risk of lymphoma. The OR was 0.58 (95% CI, 0.46–0.72) for sibships of 3, compared with sibships of 2. The ORs for sibships of 4 and 5 were 0.39 and 0.23, respectively. The adjusted effect estimates for birth order changed slightly after stratifying by sibship size (Supplementary Table S2).

Table 3 shows the effects of family structure on the risk of lymphoid cancer types. Larger sibships were significantly associated with a lower risk of lymphoma and several histologic subtypes. Birth order was inversely associated with risk of most major lymphoma entities [NHL, B-cell NHL, CLL, follicular lymphoma (FL), Mantle cell lymphoma (MCL), marginal zone lymphoma (MZL), and MM] but was not significant for HL, diffuse large B-cell lymphoma (DLBCL), or T-cell NHL. We observed no differences in the risk patterns associated with childhood, young-adult, or older-adult onset HL (Supplementary Table S1). Sibship size and birth order effects were similar among families with 2, 3, or 4 or more lymphoid cancer cases.

To estimate the probability of a chance association with birth order position or sibship size, 40,000 independent permutation tests were performed. Despite a smaller sample size and lower power of permuted data (n = 95 case/control pairs), both the regression and χ² tests for trend supported our findings, with approximately 51% and 93% of P values achieving statistical significance (P < 0.05) for birth order and sibship size, respectively (Supplementary Fig. S1). Without the family dependence, the OR estimates remained comparable with those from GEE models for sibship size (median OR = 0.82) and birth order (median OR = 0.82), validating our observations.

Higher maternal education and an above average level of income during childhood was associated with increasing risk of lymphoid cancer (Table 4). Childhood farm residents had a lower risk of lymphoma (OR = 0.65; 95% CI, 0.48–0.88), which was not significant after adjusting for sibship size (OR = 0.87; 95% CI, 0.70–1.08). Cases were less likely than their unaffected siblings to have a post-secondary education (OR = 0.62; 95% CI, 0.38–0.99; Table 4), even when adjusting for family structure. There was no relationship between paternal education or childhood house location (urban vs. rural) and lymphoma or subtypes.

Allergies and tonsillectomy were independent risk factors for most major lymphoma entities (Table 4). Lymphoid cancer risk was increased for individuals with environmental (e.g., hay fever) and drug allergies for several lymphoma entities, whereas food allergies were exclusively associated with risk of Nodular sclerosis classical HL (NSCHL; Table 4). History of appendectomy was significantly associated with a 9.7-fold increase in risk of DLBCL. Asthma was not significantly associated with risk of lymphoma with the exception of MM where small sample size makes it unclear. There was no significant association between personal history of collective autoimmune diseases and lymphoma. However, familial lymphoid cancer cases were significantly less likely than their siblings to have had organ-specific autoimmune disease (OR = 0.44; 95% CI, 0.20–0.98) after adjusting for sibship size.

Three GEE models encompassed selection of family structure, early-life environment, and immune-related disorders. Because the availability of lifestyle and disease information varied among participants, three GEE models were built in a stepwise manner (Table 5). The base model contained 1,468 individuals, in which birth order and sibship size were independent significant predictors of lymphoid cancer status. The middle model (n = 682) retained family income during childhood in addition to birth order and sibship size as significant predictors of lymphoid cancer status, while the full model (n = 321) included allergies, autoimmune diseases, tonsillectomy, and family structure, with maternal education included in the sibship size (but not birth order) model.

We assessed associations of family structure and childhood environment with disease in families with multiple lymphoid cancers. We observed an inverse relationship between birth order and cancer risk that was similar for lymphoid cancers collectively and most major subtypes (NHL, CLL, FL, and MM). Sibship size was also inversely associated with risk of lymphoma and all entities, with the exception of lymphoplasmacytic lymphoma. High maternal education, above average income during childhood, allergies, and tonsillectomy were independent risk factors for lymphoma. To our knowledge, this is the largest multiple-case family study to date that supports the hygiene hypothesis contributing to lymphoid cancers.

Familial lymphoid cancer cases were more likely to be male, which is consistent among population studies (7–10, 12, 17, 18, 23, 32) but not always true among multiple-case family studies (31, 33, 34). In this study, NHL (and subtypes) resembled the population sex distribution (28); however, HL and MM cases were significantly less likely to be male. Familial cases may reflect a different and potentially more genetic etiology in comparison with population cases (of which the majority are sporadic; ref. 31). Lower rates of B-cell malignancies in women may be attributed to nongenetic factors including body size and sex/reproductive hormones (35).

In this study, individuals earlier in birth order and/or of smaller sibships had higher risk of lymphoma. With the exception of a few studies (16, 17, 21–23), many have shown inverse associations between birth order and lymphoid cancers, including NHL (7, 8, 11), HL (7, 12, 14–16), and major subtypes (CLL, DLBCL, FL; refs. 7, 8). Opposite risk patterns for childhood- and adult-onset HL have been documented in population-based studies (15); however, we observed no difference in risk among young-adult or older-adult HL, while childhood-onset cases were limited in sample size. In our study, larger sibships were protective of lymphoid cancers, suggesting that multiple-case families may have a different disease etiology than sporadic/nonfamilial lymphomas in the population. Our finding that MCL, MZL, and mucosa-associated lymphoid tissue (MALT) lymphoma were more frequent among earlier born siblings has not been previously reported. Birth order and family size are inevitably correlated and distinguishing between their effects is difficult. Generally, eldest siblings receive more prenatal care and medical surveillance, and may be better nourished than later born siblings (36). Children from smaller sibships are traditionally of higher SES and have an older age at first bacterial or viral disease (36, 37).

Familial predisposition to lymphoma has been extensively investigated, but few studies examine the effect of family size in multiple-case families (31). The effects of birth order were similar among families with 2, 3, and 4 or more lymphoid cancer cases. We would expect families with more cases to have a more genetic etiology. Because the effects of birth order and sibship size do not vary with the number of affected individuals in the family, we suggest that they correlate with an exposure that affects lymphoid cancer risk.

Indicators of infectious exposures that are correlated with childhood SES were also supportive of the hygiene hypothesis, such that individuals with a high childhood SES were at an elevated risk of lymphoid cancer (38). Strong indicators of childhood SES include parental education and income as they capture knowledge-related behaviors that influence the age, extent, and response to infectious agents (39). More protected or cleaner environments associated with higher SES may delay infectious exposure and increase adult-onset immune-related disease risk, which is consistent with population-based associations (14, 17, 40), and our observations. In this study, childhood farm residents had a lower risk of lymphoma and FL, which is consistent with the hygiene hypothesis and epidemiologic population-based studies (11, 41). Early and frequent farm visits and animal contact (0–4 years of age) are thought to trigger an early immune response and strong immune competence suggested to prevent childhood lymphomas (11, 41, 42). Our study did not differentiate between age at childhood exposure, nor include farm-related exposures as an adult.

There is limited and contradictory information on the associations between education and risk of lymphoma (43). In this study, familial cases were more likely to have lower educational attainment than their unaffected siblings, which is consistent with sporadic DLBCL (43) and MM (43, 44) cases, but not all population-based studies (14, 38, 43). The relationship between education and lymphoid cancer is complex and may be influenced by age of diagnosis, treatment regimens, and childhood SES. Lower educational attainment may be attributed to neurocognitive impairments from chemotherapy as observed among survivors of childhood HL, adult breast, and primary central nervous system lymphoma (45).

Familial cases were significantly more likely than their unaffected siblings to report a history of allergies and a tonsillectomy, which may indicate defective immune regulation (46). A positive association between lymphoid cancer and a tonsillectomy has been described among population NHL (47, 48), HL (47–49), and CLL (47) cases, but not among multiple-case families (25). A tonsillectomy in younger children may indicate severe recurrent tonsillitis (47, 49) caused by an altered or impaired immune response that affects lymphogenic mechanisms in adulthood (47). Lymphoid cancer risk may be more pronounced in tonsillectomized children because of the declining immunologic function of the tonsils from early childhood to adulthood (48, 49). Viruses, such as EBV, have been implicated in this role, as it is associated with recurrent bouts of tonsillitis (40, 47, 48). The association between tonsillectomy and lymphoma may be confounded by SES, educational attainment, and family size although conflicting evidence has been reported (14, 38, 48).

An elevated risk of allergies has been observed among case–control (7, 14, 18) and cohort studies (46, 50–52) of nonfamilial occurrences of lymphoma (50), including NHL (18, 46, 51) and mature B-cell subtypes (18, 50, 53), HL (14, 52), MM (7, 50), and familial WM (103 cases; ref. 25); however, some case–control (but no cohort) studies observed the opposite effect (7, 18, 21, 22, 47, 54, 55). To date, most epidemiologic studies examining immunologic factors and lymphoma have been case–control studies that may be biased from immune-altering effects of preclinical lymphoid cancer (52), which would be mitigated in cohort studies. Explicit correlations between lymphoid cancer subtypes and high molecular weight allergens (54), serum IgE levels (55), and type of allergy (e.g., food, environment; refs. 7, 18, 21, 46, 47, 50) complicate the elucidation of these relationships. Both disorders are polygenic multifactorial diseases with a heritable component and environmental modifiers of risk with variable disease outcomes among populations, exposed groups, and families (56, 57). Various explanations have been proposed to account for the observed increase in incidence of lymphoma among individuals with immune dysregulation (50, 52). Families with heterogeneous lymphoid cancers may have genetic susceptibility factors that perturb optimal immune development in a way that increases risk of both allergies and lymphoid cancers. These predisposition factors may be different than those that contribute to sporadic lymphoid cancers. Nonfamilial lymphoid cancers may be more environmentally triggered, and less genetic, and therefore result from a different combination of etiologic factors than multiple-case families. In our study, affected sibships were multigenerational and had a higher frequency of allergies (55.6%) relative to the expected population frequency of 30%–40% (58), suggesting that susceptibility factors may play a bigger role than environmental influences. In addition to genetic and environmental risk factors, chronic stimulation and proliferation of lymphocytes may increase the chance of oncogenic mutations and subsequent cancer development as described by the antigenic stimulation hypothesis (50, 59).

We observed no association among collective autoimmune disorders and familial lymphoid cancer occurrence. Personal and family history of autoimmune conditions are strong established risk factors for lymphoid cancers (5, 25, 26, 60, 61), so this finding is unexpected. However, we were unable to examine subtype-specific associations among the biologically diverse autoimmune diseases, and a personal history of organ-specific autoimmune disease was associated with lower risk of lymphoma. Among individuals with an organ-specific autoimmune disease, unaffected siblings were on average 11 years younger than lymphoid cases, suggesting they may not be truly lymphoid unaffected due to shorter duration of follow-up, and/or ascertainment bias. In our study, an autoimmune condition was observed in 18.8% of individuals in a lymphoid-affected sibship, which was higher than the expected population frequency of 3%–6% (62), suggesting that these families may be enriched for genetic factors that predispose to immune dysregulation and associated conditions.

With the possible exception of MM, we observed no association between familial lymphoid cancer and asthma, consistent with the literature (9, 11, 13, 18, 21, 46). Some studies, including ours, were unable to differentiate between allergic and non-allergic asthma which may explain inconsistency among associations/studies (5). In this study, a higher risk of DLBCL (but no other subtype) was associated with an appendectomy, which is consistent with some (30, 63), but not all epidemiological population-based studies (9, 64). An appendectomy/appendicitis may reflect susceptibility to infection/inflammation; however, this information was unavailable in our study cohort. The removal of the appendix and surrounding lymphoid tissue may alter the natural immune response to pathogenic microorganisms (63).

Our observations support the antigen stimulation hypothesis (46, 59), wherein chronic immune stimulation progressively leads to random oncogenic mutations and subsequent cancer development (7, 14, 18, 25, 46, 50–52). In contrast, the immune surveillance hypothesis proposes that allergic conditions enhance the ability of the immune system to detect and eliminate malignant cells (46, 59), and is also well supported (7, 18, 21, 22, 47, 54, 55). Inconsistencies among studies may be partially attributable to differences in study designs, reverse causality, gender differences, selection bias, diverse definition and measurement of allergy, hematologic subtypes assessed, reliance on self-reported data/recall bias, and participant characteristics (e.g., families with a genetic etiology, sporadic cases; refs. 7, 23, 31, 50).

Our study has several strengths, including extensive demographic, family structure, and exposure data, and inclusion of unaffected family members. Participation rates of cases and unaffected siblings were not differentiated by SES or education. Furthermore, case–control studies can suffer from response bias due to education and SES disparities among participants; our study design is able to differentiate risk among such categories. Despite the rarity of familial hematologic malignancies, this study included a relatively large number of families (196). We were able to detect effects of birth order, sibship size, and childhood environment among familial lymphoid cancer cases while controlling for known lymphoma risk factors (age, sex, ethnicity). Limitations include use of self-reported data which may be subject to recall/response biases. We did not have complete atopic disease data or direct markers of infectious exposure, such as number and type of infections, age at infection, or serologic data. Shorter duration of follow-up may have biased some associations because insufficient time elapsed for disease development among siblings and children. Families were not ascertained by means of a systematic population-based study, which may limit the generalizability of the findings to nonfamilial lymphoma. However, this study represents the largest, and in terms of demographic and lifestyle information, the most extensively characterized cohort of lymphoid cancer families reported to date.

This investigation represents the first multiple-case family study to quantify the effects of family structure according to lymphoid cancer type. This is the first study to establish an inverse relationship between family structure (birth order and sibship size) and risk of CLL and MM in the context of families with heterogeneous lymphoid cancers. The observed inverse relationship between family structure and risk of lymphoma is supportive of the hygiene hypothesis, and that childhood exposure to infectious agents may play a role in the risk of multiple types of lymphoid cancers. Our observations indicate that lifestyle factors such as SES and education also correlate with risk of lymphoma. The familial nature of these cancers implies a role of shared genetic and/or environmental factors. Such effects may be modified by lifestyle factors that correlate with birth order and family structure, and could lead to the identification of modifiable factors that protect against lymphoid cancers, even in the context of multiple-case families.

No potential conflicts of interest were disclosed.

Conception and design: S.J. Jones, J.M. Connors, A.R. Brooks-Wilson

Development of methodology: S.J. Jones, J.J. Spinelli

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.M. Connors, A.R. Brooks-Wilson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.J. Jones, S. Stroshein, A.M. Williams, D. Liu, J.J. Spinelli, J.M. Connors

Writing, review, and/or revision of the manuscript: S.J. Jones, S. Stroshein, A.M. Williams, J.J. Spinelli, J.M. Connors, A.R. Brooks-Wilson

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.M. Connors

Study supervision: A.R. Brooks-Wilson

We thank the families for their participation. We thank Susan Slager for her advice on statistical analyses. S. Stroshein was supported by a Vice-President Research Undergraduate Student Research Award from Simon Fraser University. This study is supported by the research grant MOP-130311 (awarded to A.R. Brooks-Wilson) by the Canadian Institutes of Health Research.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.