Family history of cancer and risk of pediatric and adolescent Hodgkin lymphoma: A Children's Oncology Group study

Abstract

Family history of lymphoid neoplasms (LN) is a strong and consistently observed Hodgkin lymphoma (HL) risk factor, although it has been only marginally examined in pediatric/adolescent patients. Here healthy control children identified by random digit dialing were matched on sex, race/ethnicity, and age to HL cases diagnosed at 0-14 years at Children's Oncology Group institutions in 1989-2003. Detailed histories were captured by structured telephone interviews with parents of 517 cases and 783 controls. Epstein-Barr virus RNA detection was performed for 355 available case tumors. Two analytic strategies were applied to estimate associations between family cancer history and pediatric/adolescent HL. In a standard case-control approach, multivariate conditional logistic regression was used to calculate odds ratios and 95% confidence intervals (CIs). In a reconstructed cohort approach, each relative was included as a separate observation and multivariate proportional hazards regression was used to produce hazard ratios (HRs) and 95% CIs. Using the latter, pediatric/adolescent HL was associated with a positive family history (HR=1.20, 95%CI: 1.06-1.36), particularly early onset cancers (HR=1.30, 95%CI: 1.06-1.59) and those in the paternal lineage (HR=1.38, 95%CI: 1.16-1.65), with a suggested association for LN in first-degree relatives (HR=3.61, 95%CI: 0.87-15.01). There were no discernable patterns for EBV+ versus EBV– HL. The clustering of LN within pedigrees may signal shared genetic susceptibility or common environmental exposures. Heritable genetic risk variants have only recently begun to be discovered, however. These results are consistent with other studies and provide a compelling rationale for family-based studies to garner information about genetic susceptibility to HL.

Hodgkin lymphoma (HL) is a malignancy of germinal center B lymphocytes occurring in the lymph nodes or other secondary lymph organs and characterized by a small proportion (∼1%) of giant, often binucleated malignant cells in a sea of infiltrating immune cells.¹ In the U.S., HL represents the 8th most common malignancy among children and adolescents <15 years of age and is diagnosed at a rate of 5.6 cases per 1,000,000 person-years.² HL arising in this age group is thought to be an etiologically discrete entity compared to HL in older adolescents and young adults (15-39 years) and older adults (50+ years)³ due to its distinctive demographic, clinical and pathological characteristics.^4,5 The few established HL risk factors identified to date include Epstein-Barr virus (EBV) infection,^2-4 congenital and acquired immunodeficiency,^6,7 and family history of HL and other lymphoid neoplasms (LN; e.g., HL, non-Hodgkin lymphoma (NHL)/chronic lymphocytic leukemia (CLL), acute lymphoblastic leukemia (ALL), and multiple myeloma (MM)).^8-16

The occurrence of familial HL was first noted over a century ago¹⁷ and has been examined in a number of epidemiological studies since that time. Population-based registry studies from the Nordic countries and the Utah Population Database (UPDB) have reported increased risk for HL and other LN among first-degree relatives of HL cases, with standardized incidence ratios (SIRs) on the order of 1.2-3.1 for parents and 4.3-6.2 for siblings.^13-15 Risks among same-sex siblings (SIR=8.0-11.8)¹¹ and monozygotic twins (SIR=99)¹⁶ are higher still. Results for other relatives are less consistent; second-degree relatives had 4.4-fold increased risk in the UPDB,¹³ but risk was not significantly increased in Iceland (SIR=1.85(NS)).¹⁴ Overall, 4-10% of HL cases had at least one relative affected with HL or another LN^13,18 and HL heritability was estimated at 28% in Sweden.¹⁹

Fewer studies have focused specifically on HL in childhood and early adolescence, where familial risks are markedly greater. HL risks in parents and siblings of HL cases ≤36 years were 8.8- and 7.2-fold higher, respectively, than those in the underlying Swedish population, with comparable results in HL patients <15 and ≥15 years.⁸ Similarly, population-based epidemiologic studies in France and England reported 5.4 to 5.8-fold increased HL risks in first- and second-degree or first-degree relatives, respectively,^9,10 and an analysis of 1858 5-yr HL survivors diagnosed at <21 years in the U.S. and Canada indicated a 5.9-fold increased HL risk in siblings.¹²

To our knowledge, no previous studies of pediatric/adolescent HL have explored family cancer history by tumor EBV status, although prior research has clearly demonstrated that there are susceptibility factors that are both shared by and specific to EBV+ HL and EBV– HL.^4,20,21 Our objective was therefore to characterize the association between family cancer history and pediatric/adolescent HL, overall and by tumor EBV status, using data from the largest case-control study of incident pediatric and adolescent HL conducted to date.

Materials and Methods

Data and specimens were collected in Children's Cancer Group (CCG; now Children's Oncology Group (COG)) Protocol E13: “Case-control study of Hodgkin's Disease in children”.⁵

Cases

Pathologically confirmed Hodgkin lymphoma cases diagnosed between 0 and 14 years of age at a participating CCG/COG institution in the United States, Puerto Rico, or Canada during the period January 31, 1989 through July 28, 2003 were eligible if they had physician approval for contact, a telephone in their residence, and at least one biological parent who spoke English or Spanish and consented to participate. Deceased cases meeting these criteria were eligible.

Controls

Unaffected control children were individually matched to cases on sex, race/ethnicity, and date of birth and were identified and recruited via random digit dialing.^22,23 As with cases, controls were also required to have a telephone in their residence, and at least one biological parent who spoke English or Spanish and consented. For cases diagnosed at <5 years of age, controls were matched on birth date ±1 year, and for cases aged 5-14 years, controls were matched on birth date ±3 years. Up to three controls were selected for cases aged <10 years and one control was selected for cases aged ≥10 years. Near the end of the study, a sequential algorithm that allowed for (in order of priority): increased age matching increment, different race/ethnicity, or neighboring area code was used to enhance the matching success rate.

Interviews

An initial phone contact was made with all families to determine interest; study materials (study description, interview guide, consent forms (cases only)) were sent by mail to families that agreed. Family medical history was captured in the initial questionnaire via structured telephone interviews with each parent providing verbal consent; a grandparent or other relative completed a surrogate interview in the absence of a biological parent. Parents of cases and controls were asked to provide the cancer history for first-degree (i.e., parents and full siblings) and second-degree (i.e., half-siblings, grandparents, aunts, and uncles) biological relatives of index children (who are all first-degree relatives of the parents). For each affected relative, the specific information requested included: type and age of onset of the cancer, sex, relationship to the index child, and parental lineage (maternal vs. paternal). Participating subjects were recontacted an average of 8.8 years following the initial interview (range: 0.7-16.7 years) and asked to complete a brief follow-up interview regarding further occurrence of cancers in index subjects and their relatives (and other selected exposures), as a discernible rise in the number of cancer cases was expected over that period.

All cancers reported by parents were subsequently coded with the corresponding International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes (see Supplementary Table 1). We excluded non-melanoma skin cancers, other nonmalignant tumors, and in situ neoplasms to reduce potential information bias due to incomplete recall, as well as secondary cancers, which are likely attributable to treatment effects.

Clinical data and tumor Epstein-Barr virus detection

Clinical and pathologic data for all cases, as well as blood and tumor specimens from a subset of cases providing informed consent, were collected from the diagnosing/treating CCG/COG institutions. Cases were assigned to a HL histologic subtype (nodular sclerosis, NS; mixed cellularity, MC; lymphocyte predominant, LP; other). Archived tumor samples were retrieved for 355 cases for EBV detection. The distributions of cases' age at diagnosis, sex, race/ethnicity, and other factors related to socioeconomic status (SES; i.e., maternal age, maternal education, household income) were similar in those with versus without retrieved tumor samples; however, the retrieved group included a greater proportion of NS HL and fewer “other” cases. A standard digoxigenin‐based in situ hybridization technique was used to detect EBV-encoded RNAs (EBER-1 and EBER-2)²⁴ in available tumor samples. For these assays, positive controls included known EBV+ HL specimens and B95‐8 cells, while negative controls included EBER sense probes. Detection of small nuclear RNA U6 via molecular probes verified the preservation of intact RNA in all tumor specimens.

Protection of Human Subjects

Institutional Review Boards at the University of Pittsburgh and the University of New Mexico (the original coordinating centers), the University of Minnesota, and participating CCG/COG institutions approved the study.

Statistical Analysis

Two complementary approaches were used to examine the association between familial aggregation of cancer and risk of pediatric/adolescent HL.^25,26 The first approach was a traditional case-control analysis, wherein we used conditional logistic regression to model the association between a positive family cancer history, overall and for specific relative and/or cancer subgroups, and HL in the index child, overall and by tumor EBV status (EBV+, EBV−); odds ratios (ORs) and 95% confidence intervals (CIs) were calculated to estimate risk. In addition to accounting for matching factors, we adjusted for maternal education level (<high school (HS) graduate, HS graduate, >HS) to control for confounding due to differences in SES levels between cases and controls, and for number of first- and second-degree relatives, respectively, to minimize confounding by family size/structure.

The logistic regression model described above is the standard case-control analysis method, but it fails to fully account for the family size and structure of each respondent and neglects the information inherent in the affected relative's age at onset. In the second, more powerful approach, each relative was included as a separate observation in the reconstructed cohort. Because each relative is considered a study unit and the case-control status of the index study participant is treated as his/her family history, the family history variable corresponds to one relative; thus, the problem of varying number of relatives does not arise. Follow-up time was calculated for each relative as the period between the date of birth and the end of follow-up, defined by the reported age at cancer diagnosis, age of death, or age at date of last interview, whichever came first. Multivariate proportional hazards regression was used to produce hazard ratios (HRs) and 95% CIs adjusted for the index child's sex, age at diagnosis (where case diagnosis date was assigned as the control pseudo-diagnosis date), and race/ethnicity, and maternal educational attainment; a modification to the generalized estimating equation (GEE), i.e., the robust sandwich estimate of the covariance matrix, was employed to account for the correlation of cancer outcomes within families.²⁷ We examined relatives of all index subjects combined, as well as the subgroups produced by stratification on case tumor EBV status (EBV+, EBV−), HL histologic subtype (NS, MC, LP), and HL diagnosis age (<10, ≥10 years). Models restricted to NS and MC HL (i.e., components of classical HL) generated nearly identical estimates to those for HL overall; thus relatives of all HL cases were retained in the final models. The proportional hazards assumption was evaluated by including the interaction between family cancer history exposure variables and follow-up time in each of the models. Given that there was no evidence against the proportional hazards assumption, interaction terms were not retained.

A sensitivity analysis was conducted to examine the impact of including follow-up interview responses available for a subset of subjects (487 cases, 593 controls) in the primary analyses.

Analyses were performed with Statistical Analysis Software, release 9.3 (SAS Institute Inc., Cary, NC, USA). All statistical tests were two-sided.

Results

Response rates and subject characteristics

A flow diagram depicting subject participation is provided in Supplementary Figure 1. Of the 646 potentially eligible HL cases ascertained at 117 US and Canadian CCG/COG institutions, interviews were completed for 517 (80%), including 324 NS HL, 92 MC HL, and 60 LP HL. Additional cases were excluded from analysis because no matched control was available (68, 11%) or because they did not meet the age criteria (1, 0.2%). The remainder did not complete interviews as a result of parental refusal (4%), inability to locate families (3%), or provider refusal (2%). For the maternal interviews, 451 biological mothers participated, while biological fathers (9) or other willing first-degree relatives (57) completed surrogate interviews when the mother was unavailable. For the paternal interviews, 329 biological fathers responded, and mothers (95) or other relatives (93) completed interviews for the rest.

To identify matched RDD controls, 207,438 telephone calls were made to 88,429 telephone numbers, of which 50,256 numbers were non-residential, 2,822 refused to provide a household census or hung up, and 360 did not otherwise meet the eligibility criteria. Considering the 34,991 households that provided a census, 25,100 had no children in the residence, 8,682 had children who could not be matched to a case, and 1,209 had a matching child (including 136 matched using the relaxed criteria). Maternal and paternal interviews were completed for 784 (64.8%) of the matches, while 323 (26.7%) households actively refused, 101 (8.4%) passively refused, and 1 (0.1%) was ineligible. One control family did not answer the section on family cancer history and was therefore not included in the current analysis.

Selected subject characteristics are shown in Table 1; cases and controls were closely matched on sex, race/ethnicity, and age as dictated by the matching scheme. With respect to family structure, the mean ages of parents and grandparents at the time of the initial interviews were very similar across cases and controls, however, case pedigrees (mean=16 relatives) were slightly larger than control pedigrees (mean=15 relatives, P-value=0.01) on average. In addition, the distributions of measures of SES (i.e., maternal educational attainment and household income) were somewhat higher in controls compared to cases. In the analysis of tumor specimens, EBV RNAs were detected in 16.3% of cases, including 22.7% in the 0-4 year age group, 29.5% in the 5-9 group and 11.5% in the 10-14 group, while 52.4% had no detectable EBV RNA present, and 31.3% did not have tumors available for analysis. Supplementary Table 2 shows the distribution of cases by diagnosis age, HL subtype, and tumor EBV status.

Table 1.

Selected descriptive characteristics of 517 childhood and adolescent Hodgkin lymphoma cases and 783 matched controls.^*

	Controls	Cases
	N (%)	N (%)	Unadjusted OR	95% CI	P-value
Age at diagnosis (years)†
0-4	97 (12.4)	22 (4.3)	-	-	-
5-9	301 (38.4)	122 (23.6)	-	-	-
10-14	326 (41.6)	373 (72.2)	-	-	-
15+	59 (7.5)	0 (0.0)	-	-	-
Sex†
Male	517 (66.0)	320 (61.9)	-	-	-
Female	266 (34.0)	197 (38.1)	-	-	-
Race/ethnicity†
White, Non-Hispanic	631 (80.6)	386 (74.7)	-	-	-
Black, Non-Hispanic	66 (8.4)	54 (10.4)	-	-	-
Hispanic/Asian/Pacific Islander	86 (11.0)	77 (14.9)	-	-	-
Maternal age at child's birth (years)
<25	316 (40.4)	248 (48.1)	Ref
25-29	272 (34.8)	164 (31.8)	0.81	0.62-1.05	0.12
≥30	195 (24.9)	104 (20.2)	0.74	0.54-1.01	0.05
Maternal educational attainment
Less than high school graduate	64 (8.2)	73 (14.2)	Ref
High school graduate	248 (31.8)	175 (34.0)	0.55	0.36-0.83	0.004
Beyond high school	468 (60.0)	267 (51.8)	0.44	0.29-0.65	< 0.0001
Household income at child's birth
0 - $19,999	267 (35.5)	243 (48.4)	Ref
$20,000 - $39,999	333 (44.3)	206 (41.0)	0.69	0.53-0.91	0.009
$40,000+	152 (20.2)	53 (10.6)	0.42	0.28-0.63	< 0.0001
1° and 2° relatives
≤ 12	249 (31.8)	131 (25.3)	Ref
13-15	249 (31.8)	171 (33.1)	1.26	0.93-1.71	0.13
≥ 16	285 (36.4)	215 (41.6)	1.39	1.03-1.86	0.03
Mean (SD) number of 1° and 2° relatives	15 (4.2)	16 (4.5)	1.04	1.01-1.07	0.01
Sibship‡
0	146 (18.7)	86 (16.6)	Ref
1	325 (41.5)	210 (40.6)	1.06	0.75-1.48	0.75
2	200 (25.5)	133 (25.7)	1.08	0.75-1.54	0.69
3+	112 (14.3)	88 (17.0)	1.29	0.86-1.95	0.22
Mean (SD) age of parents at initial interview (years)	38.3 (6.0)	38.4 (5.7)	0.97	0.94-0.99	0.002
Mean (SD) age of grandparents at initial interview (years)	66.0 (8.1)	66.2 (8.3)	0.98	0.97-1.00	0.04

CI = confidence interval; OR = odds ratio

^*Numbers in tables may not sum to total number of cases/controls due to missing values. One control was excluded from analysis due to missing family history information.

^†Cases and controls were matched on sex, race/ethnicity, and age; ORs were not calculated for matching variables.

^‡Sibship is based on full siblings.

Case-control analysis

Overall, 61.0% of case children had a reported family history of any cancer, with 30.9% having two or more cancers in first- or second-degree relatives, while 55.9% of control children had a family history of any cancer and 23.5% reported 2 or more instances of cancer (Table 2). After adjustment for number of relatives and maternal education level in the conditional logistic regression models, these differences produced a modest positive association (OR=1.11, 95% CI: 0.99-1.23) with little evidence for a linear dose response (P-trend=0.16). Similar results were observed for EBV+ and EBV– HL. Significant associations were observed for a family history of testicular cancer (OR=5.89, 95% CI: 1.07-32.33), although this observation was based on few subjects with affected relatives (5 cases, 2 controls) and lacked precision, and the heterogeneous grouping of “other” solid tumors (OR=1.47, 95% CI: 1.04-2.07). Importantly, results of the case-control analysis were in general agreement with those from the reconstructed cohort (Table 3).

Table 2.

The association between family history of cancer and childhood and adolescent Hodgkin lymphoma, overall and by EBV status, estimated via case-control analysis.^*†

	Combined cases					EBV+					EBV-
	Ncontrols	Ncases	OR‡	95% CI	P-value	Ncontrols	Ncases	OR‡	95% CI	P-value	Ncontrols	Ncases	OR‡	95% CI	P-value
1° and 2° relatives with cancer
No	344	201	Ref			86	33	Ref			153	101	Ref
Yes	436	314	1.11	0.99-1.23	0.06	76	51	1.21	0.92-1.59	0.17	226	169	1.12	0.97-1.30	0.12
1° and 2° relatives with cancer
0	344	201	Ref			86	33	Ref			153	101	Ref
1	253	155	0.96	0.72-1.26	0.75	43	30	1.70	0.89-3.22	0.11	133	83	0.95	0.64-1.40	0.78
2+	183	159	1.28	0.94-1.73	0.12	33	21	1.72	0.76-3.90	0.20	93	86	1.28	0.83-1.96	0.26
P-trend					0.16					0.21					0.35
Earliest age of cancer onset in family member
None	344	201	Ref			86	33	Ref			153	101	Ref
<50 years	184	158	1.41	0.97-2.05	0.07	30	25	2.41	1.02-5.70	0.04	98	90	1.41	0.84-2.38	0.20
≥50 years	252	156	0.89	0.65-1.21	0.45	46	26	1.16	0.56-2.42	0.69	128	79	0.87	0.55-1.38	0.55
Malignancies in 1° & 2° Relatives
Hematopoietic cancers	61	43	1.05	0.68-1.60	0.84	13	11	2.04	0.83-5.00	0.12	25	23	1.28	0.69-2.38	0.44
Lymphoid cancers	40	34	1.21	0.74-1.97	0.45	9	7	1.63	0.54-4.91	0.39	16	21	1.72	0.85-3.46	0.13
HL	12	14	1.47	0.65-3.35	0.36	2	2	1.82	0.20-16.39	0.60	5	8	1.92	0.60-6.22	0.27
NHL	21	14	0.96	0.48-1.95	0.92	5	3	1.56	0.31-7.89	0.59	9	11	1.51	0.61-3.77	0.38
ALL	1	4	5.71	0.63-51.60	0.12	0	1	-	-	-	0	2	-	-	-
MM	7	3	0.69	0.16-2.88	0.61	2	1	0.84	0.07-10.21	0.89	3	1	0.49	0.05-5.16	0.55
Myeloid cancers	4	0	-	-	-	1	0	-	-	-	2	0	-	-	-
Solid tumors	402	290	1.05	0.82-1.35	0.70	69	44	1.41	0.74-2.70	0.30	209	157	1.08	0.76-1.52	0.68
Breast	72	57	1.18	0.79-1.74	0.42	11	5	0.85	0.27-2.71	0.78	36	34	1.44	0.84-2.48	0.18
Central nervous system	25	12	0.76	0.37-1.56	0.45	4	1	0.48	0.05-4.74	0.53	14	8	0.87	0.35-2.16	0.76
Cervical/Uterine/Ovarian	62	47	1.07	0.70-1.62	0.77	13	6	0.66	0.22-2.00	0.47	27	25	1.56	0.85-2.86	0.15
Colorectal	41	37	1.18	0.74-1.89	0.48	5	5	2.72	0.71-10.49	0.15	19	20	1.34	0.70-2.56	0.38
Lung	87	60	0.94	0.64-1.37	0.75	19	10	1.05	0.42-2.65	0.91	48	34	0.82	0.49-1.38	0.46
Melanoma	81	69	1.30	0.90-1.87	0.17	12	10	1.27	0.46-3.52	0.65	43	40	1.35	0.83-2.20	0.22
Prostate	35	37	1.53	0.92-2.54	0.10	6	6	1.74	0.52-5.75	0.37	15	16	1.41	0.63-3.14	0.40
Stomach/Small Intestine/Pancreatic	44	26	0.78	0.46-1.30	0.33	8	2	0.40	0.08-2.07	0.27	25	13	0.57	0.27-1.17	0.12
Lip/Oral/Pharyngeal/Esophageal	28	19	0.93	0.50-1.71	0.81	5	3	1.54	0.36-6.70	0.56	16	12	0.93	0.42-2.06	0.86
Bladder/Kidney	21	11	0.83	0.39-1.79	0.64	6	2	0.84	0.16-4.42	0.84	13	7	0.76	0.29-2.00	0.58
Liver	12	10	0.97	0.40-2.38	0.95	3	3	1.67	0.31-9.07	0.55	5	5	0.88	0.23-3.44	0.86
Testicular	2	5	5.89	1.07-32.33	0.04	1	1	4.80	0.28-81.99	0.28	1	4	7.89	0.82-75.46	0.07
Other Solid tumors	82	81	1.47	1.04-2.07	0.03	12	12	2.20	0.88-5.50	0.09	51	45	1.21	0.77-1.92	0.41

ALL = acute lymphoblastic lymphoma; CI = confidence interval; EBV = Epstein-Barr virus; HL = Hodgkin lymphoma; MM = multiple myeloma; NHL = Non-Hodgkin lymphoma; OR = odds ratio

^*Numbers in tables may not sum to total number of cases/controls due to missing values (2 cases and 3 controls missing values for maternal education).

^†EBV status was not determined for 162 cases.

^‡ORs adjusted for number of relatives of corresponding degree (i.e., number of 1° relatives, number of 2° relatives as appropriate), maternal education (<HS graduate, HS, >HS).

Table 3.

The association between family history of cancer and childhood and adolescent Hodgkin lymphoma overall and by EBV status estimated via reconstructed cohort analysis.^*,†

	Combined cases					EBV+					EBV-
	N Relatives of Controls	N Relatives of Cases	HR‡	95% CI	P- value	N Relatives of Controls	N Relatives of Cases	HR‡	95% CI	P- value	N Relatives of Controls	N Relatives of Cases	HR‡	95% CI	P- value
Family history of cancer
1° & 2° Relatives w/ cancer	712	570	1.20	1.06-1.36	0.005	129	84	1.37	0.99-1.90	0.06	367	322	1.19	1.00-1.41	0.05
1° & 2° Relatives w/o cancer	10471	7167	Ref			2220	1198	Ref			4970	3714	Ref
1° Relatives w/ cancer	37	37	1.38	0.85-2.26	0.20	5	4	1.59	0.42-6.03	0.49	17	25	1.82	0.95-3.47	0.07
1° Relatives w/o cancer	2731	1858	Ref			550	298	Ref			1354	975	Ref
2° Relatives w/ cancer	675	533	1.19	1.04-1.36	0.01	124	80	1.37	0.98-1.92	0.07	350	297	1.15	0.96-1.38	0.12
2° Relatives w/o cancer	7740	5309	Ref			1670	900	Ref			3616	2739	Ref
Age of cancer onset
None	10469	7167	Ref			2219	1198	Ref			4970	3714	Ref
< 50 years	219	196	1.30	1.06-1.59	0.01	35	29	1.57	0.95-2.61	0.08	117	113	1.29	0.98-1.70	0.07
≥ 50 years	495	374	1.13	0.96-1.32	0.13	95	55	1.17	0.78-1.75	0.45	250	209	1.13	0.91-1.41	0.26
Parental lineage
Maternal relatives w/ cancer	374	276	1.05	0.88-1.26	0.60	63	35	1.01	0.59-1.72	0.97	185	163	1.14	0.90-1.45	0.29
Maternal relatives w/o cancer	4452	3038	Ref			937	496	Ref			2083	1600	Ref
Paternal relatives w/ cancer	335	290	1.38	1.16-1.65	0.0004	66	48	1.77	1.14-2.75	0.01	181	157	1.24	0.98-1.57	0.08
Paternal relatives w/o cancer	4363	2943	Ref			946	509	Ref			2102	1516	Ref
Malignancies in 1° Relatives
Hematopoietic cancers	4	9	2.97	0.85-10.39	0.09	0	2	-	-	-	2	6	3.95	0.71-21.88	0.12
Lymphoid cancers	3	8	3.61	0.87-15.01	0.08	0	2	-	-	-	2	6	3.95	0.71-21.88	0.12
HL	1	4	5.79	0.50-67.33	0.16	0	0	-	-	-	1	4	5.61	0.52-60.77	0.16
NHL	1	1	1.13	0.08-15.61	0.93	0	0	-	-	-	1	1	1.13	0.10-12.61	0.92
ALL	1	2	2.80	0.25-31.18	0.40	0	0	-	-	-	0	1	-	-	-
MM	0	1	-	-	-	0	1	-	-	-	0	0	-	-	-
Myeloid cancers	1	0	-	-	-	0	0	-	-	-	0	0	-	-	-
Solid tumors	33	28	1.19	0.69-2.03	0.53	5	2	0.84	0.15-4.63	0.84	15	19	1.55	0.77-3.12	0.22
Breast	1	2	2.35	0.22-25.09	0.48	0	0	-	-	-	1	1	1.08	0.07-15.76	0.95
Central nervous system	0	1	-	-	-	0	0	-	-	-	0	1	-	-	-
Cervical/Uterine/Ovarian	8	4	0.71	0.21-2.47	0.59	2	0	-	-	-	1	1	1.49	0.85-2.60	0.16
Colorectal	4	0	-	-	-	0	0	-	-	-	2	0	-	-	-
Lung	1	0	-	-	-	0	0	-	-	-	1	0	-	-	-
Melanoma	9	9	1.55	0.61-3.92	0.36	1	1				4	7	2.04	0.65-6.36	0.22
Prostate	0	0	-	-	-	0	0	-	-	-	0	0	-	-	-
Stomach/Small Intestine/ Pancreatic	3	1	0.47	0.06-3.81	0.48	0	0	-	-	-	2	1	0.77	0.39-1.56	0.48
Lip/Oral/Pharyngeal/ Esophageal	2	1	0.73	0.07-7.56	0.79	0	0	-	-	-	1	1	-	-	-
Bladder/Kidney	2	0	-	-	-	1	0	-	-	-	1	0	-	-	-
Liver	0	0	-	-	-	0	0	-	-	-	0	0	-	-	-
Testicular	1	4	8.09	1.03-63.64	0.05	1	0	-	-	-	0	4	-	-	-
Other Solid tumors	2	6	3.45	0.75-15.86	0.11	0	1	-	-	-	2	3	1.24	0.83-1.85	0.28
Malignancies in 2° Relatives
Hematopoietic cancers	61	40	1.04	0.69-1.57	0.85	13	9	1.85	0.81-4.20	0.14	26	23	1.16	0.63-2.13	0.64
Lymphoid cancers	40	28	1.04	0.64-1.69	0.87	9	5	1.35	0.49-3.73	0.57	17	17	1.19	0.57-2.48	0.64
HL	11	11	1.45	0.65-3.26	0.37	2	2	2.04	0.38-10.97	0.41	4	5	1.68	0.45-6.24	0.44
NHL	21	13	0.92	0.47-1.79	0.80	5	3	1.64	0.43-6.27	0.47	9	10	1.21	0.50-2.97	0.67
ALL	0	2	-	-	-	0	0	-	-	-	0	1	-	-	-
MM	8	2	0.35	0.07-1.75	0.20	2	0	-	-	-	4	1	0.27	0.02-3.58	0.32
Myeloid cancers	3	0	-	-	-	1	0	-	-	-	2	0	-	-	-
Solid tumors	614	493	1.20	1.05-1.38	0.01	111	71	1.32	0.92-1.89	0.14	324	274	1.15	0.95-1.40	0.15
Breast	74	59	1.17	0.82-1.67	0.38	11	6	1.03	0.42-2.52	0.95	38	36	1.25	0.77-2.05	0.37
Central nervous system	25	11	0.75	0.37-1.52	0.43	4	1	0.78	0.11-5.44	0.80	14	7	0.81	0.34-1.93	0.63
Cervical/Uterine/Ovarian	61	47	1.21	0.80-1.83	0.37	15	7	0.75	0.25-2.25	0.61	27	25	1.49	0.85-2.60	0.16
Colorectal	40	39	1.38	0.88-2.15	0.16	5	5	2.75	0.78-9.66	0.11	19	21	1.40	0.74-2.63	0.30
Lung	96	66	1.01	0.73-1.41	0.95	19	11	1.22	0.59-2.53	0.59	54	39	0.95	0.61-1.48	0.83
Melanoma	89	77	1.35	0.94-1.94	0.10	13	12	2.01	0.74-5.42	0.17	47	44	1.26	0.77-2.05	0.35
Prostate	36	40	1.71	1.08-2.69	0.02	6	6	3.23	1.08-9.68	0.04	15	17	1.40	0.70-2.83	0.35
Stomach/Small Intestine/ Pancreatic	42	27	0.91	0.55-1.49	0.70	8	2	0.41	0.10-1.73	0.22	23	13	0.78	0.39-1.56	0.48
Lip/Oral/Pharyngeal/ Esophageal	27	18	1.03	0.55-1.92	0.93	5	3	1.81	0.44-7.35	0.41	16	11	0.99	0.45-2.17	0.97
Bladder/Kidney	20	11	0.90	0.43-1.85	0.77	6	2	0.83	0.17-4.10	0.82	12	7	0.89	0.35-2.25	0.80
Liver	13	10	1.06	0.47-2.39	0.88	4	3	1.21	0.23-6.52	0.82	5	5	1.32	0.42-4.12	0.63
Testicular	1	1	2.19	0.09-51.85	0.63	0	1	-	-	-	1	0	-	-	-
Other solid tumors	90	87	1.40	1.01-1.94	0.05	15	12	1.54	0.51-4.61	0.44	53	49	1.24	0.83-1.85	0.28

ALL = acute lymphoblastic lymphoma; CI = confidence interval; EBV = Epstein-Barr virus; HR = hazard ratio; HL = Hodgkin lymphoma; MM = multiple myeloma; NHL = Non-Hodgkin lymphoma

^*Some relatives had >1 primary cancer, including 1 1° relative of a case, 2 1° relatives of controls, 59 2° relatives of cases, and 64 2° relatives of controls.

^†EBV status was not determined for 162 cases.

^‡HRs adjusted for index child's sex (M vs. F), age at diagnosis (continuous), and race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic/Asian/Pacific Islander), and maternal educational attainment (<HS graduate, HS, >HS).

Reconstructed cohort analysis

A total of 1,895 first- and 5,842 second-degree relatives were included in the 517 case pedigrees, contributing 60,704 and 278,157 person-years, respectively, while 2,768 first- and 8,415 second-degree relatives were described for the 783 control children, contributing 90,106 and 398,889 person-years. When all first- and second-degree relatives were considered, a family history of cancer was associated with HL overall (HR=1.20, 95% CI: 1.06-1.36) and with EBV+ (HR=1.37, 95% CI: 0.99-1.90) and EBV– HL (HR=1.19, 95% CI: 1.00-1.41), respectively, after adjustment for the matching variables and maternal educational attainment in the proportional hazards regression models (Table 3). The increased risk was more pronounced for early onset cancers, i.e., those diagnosed at <50 years of age, (HR=1.30, 95% CI: 1.06-1.59) and those occurring in the paternal line (HR=1.38, 95% CI: 1.16-1.65). Interestingly, the latter association was significant for male (HR=1.50, 95% CI: 1.19-1.88) but not female (HR=1.25, 95% CI: 0.94-1.65) index subjects, although the magnitude of the HRs was similar in both groups (data not shown).

Among first-degree relatives, a family history of LN (including HL, NHL/CLL, ALL, and MM) was associated with a strong positive risk for pediatric/adolescent HL in index children (HR=3.61, 95% CI: 0.87-15.01), although the point estimate lacked precision due to the relatively small number of affected relatives (8 in cases and 3 in controls) and was of borderline significance; no significant associations were detected for any specific LN. In addition, a positive association was observed for history of testicular cancer in a first-degree relative (HR=8.09, 95% CI: 1.03-63.64). It is also noteworthy that 9 HL cases had co-twins (7 same sex, 2 opposite sex, zygosity unknown) and none of the co-twins had developed cancers at the time of last interview (data not shown).

For second-degree relatives, a family history of solid tumors was associated with increased HL risk in index children, which was attributable to prostate cancer (HR=1.71, 95% CI: 1.08-2.69) and other solid tumors (HR=1.40, 95% CI: 1.01-1.94).

There were no substantive differences in familial aggregation patterns between EBV+ and EBV– cases. Likewise, similar results were observed in examining associations for the <10 and ≥10 year age groups, respectively (Supplementary Table 3). Notable exceptions were the strong positive association for a family history of colorectal cancer observed for the <10 year age group (HR=2.52, 95% CI: 1.20-5.30) and the positive association for prostate cancer (HR=1.92, 95% CI: 1.11-3.30) found in those ≥10 years. In examining the HL subtypes separately (Supplementary Table 4), few associations were observed for NS HL, the most common subtype (n=324 cases). Several significant positive associations were observed for MC HL (n=92), including an overall positive family cancer history (HR=1.40, 95% CI: 1.04-1.89), early onset familial cancers (HR=1.59, 95% CI: 1.06-2.37), cancer in the paternal lineage (HR=1.77, 95% CI: 1.16-2.69), and colorectal cancer (HR=3.95, 95% CI: 1.24-12.56). LP HL, the smallest diagnostic subgroup (n=60 cases), was associated with later onset familial cancer (HR=1.57, 95% CI: 0.99-2.50), cancer in paternal relatives (HR=2.03, 95% CI: 1.19-3.46), and a family history of colorectal (HR=7.27, 95% CI: 1.89-27.86) and gynecological (HR=2.85, 95% CI: 1.08-7.51) cancers.

Discussion

In this largest etiologic study of incident HL in children <15 years conducted to date, cancer in one or more first- or second-degree relatives was positively associated with pediatric/adolescent HL; moderate positive associations were observed for earlier (<50 years) onset cancers and cancers occurring in paternal relatives. In the reconstructed cohort analysis, LN in a first-degree relative was associated with a borderline significant 3.6-fold risk of HL in the index child, although this was based on few affected relatives. A smattering of associations with other cancers was also observed among index HL subgroups, as discussed below. These results are highly concordant with those from prior reports in pediatric and adult HL.^8-16

The clustering of LN within pedigrees may signal shared genetic susceptibility, common environmental exposures, or the complex interplay between them. Most notably, a twin study of young adult HL showed a 99-fold increased risk in monozygotic twins of cases, but no increased risk among dizygotic co-twins,¹⁶ strongly implicating genetic susceptibility over environmental effects.

The different patterns of HL/LN aggregation that have been reported suggest different modes of genetic inheritance. In some families, two or more HL cases are seen, with siblings constituting the majority of affected relative pairings, suggesting a recessive genetic trait.^8,15,28-30 In other families, multiple generations are affected with LN,^15,30,31 implying the transmission of highly penetrant, possibly dominant, pleiotropic genetic traits. In a subset of families, there is evidence for genetic anticipation, where the age of diagnosis in subsequent generations is successively lower,^19,32 indicative of a non-Mendelian mechanism. In the current study, there were few siblings affected with cancer and none affected with HL, thus most of the observed LN aggregation was attributable to affected case parents (Supplementary Table 5). Results of genome-wide association studies (GWAS),^20,33 along with those from case-control²¹ and familial linkage studies,³⁴ have consistently implicated the HLA region in HL susceptibility; however, these associations are thought to be insufficient to fully explain the strong familial clustering observed.³⁵ Specific heritable genetic variants conferring increased risk within individual families have only recently begun to be discovered.^36,37 Accordingly, a next logical step in this era of genomic sequencing is to conduct segregation analyses across a number of multiplex families to identify shared genes/pathways in which (probably rare) variants cluster.

Simultaneous familial exposure or shared familial susceptibility to an environmental factor, such as EBV infection, may also lead to familial aggregation of disease. In exploring the role of EBV in familial HL, two small studies failed to show an excess of positive concordance for EBV RNA in paired tumors from multiply affected families (2/17 and 1/5 pairs were concordantly EBV+, respectively), although they did observe concordance for HL subtype in all but 4 of these families.^38,39 Interestingly, in a case report describing a family of five children, 3 children with identical HLA class I haplotypes developed EBV+ HL (2 NS and 1 MC), whereas the 2 children that did not develop HL had other HLA haplotypes, suggesting an interaction between HLA genotype and EBV infection in the pathology of HL in this family.⁴⁰ On balance, we conclude that the underlying causes for most instances of familial LN remain largely unidentified.

Family history of selected other cancers also aggregated with various subsets of pediatric/adolescent HL in the current study, including colorectal, prostate, testicular, and gynecological tumors. The association between pediatric/adolescent HL and family history of colon and rectal cancers observed in children ages 0-9 years in the current study (HR=2.52, 95% CI: 1.20-5.30) was also reported in the aforementioned French case-control study (OR=2.4, 95% CI:1.0-5.4),¹⁰ but not in the English study.⁹ Likewise, the Swedish Cancer Registry observed clustering of HL in parents and testicular seminomas in sons,⁴¹ while other studies did not find an aggregation.⁴² Together these observations may suggest a role for variants in DNA mismatch repair genes or other damage response pathways in the etiology of malignancies in some families, as each of the associated cancers is a common (i.e., tumors of the colon, rectum, and endometrium) or rare (i.e., prostate, testicular cancer, HL) presentation of hereditary nonpolyposis colorectal cancer (HNPCC), for example.^43,44 Indeed, results of an ad hoc analysis revealed a greater burden of these tumors in case versus control families in the current study (Supplementary Table 6). Alternately, given the study design, we cannot rule out recall bias or chance as possible explanations.

As contact with some families was lost between the initial and follow-up interviews, follow-up interviews were conducted in a subset of case (89%) and control (75%) families. We therefore performed a sensitivity analysis to evaluate the possibility that selection bias was introduced by using all available (i.e., initial and follow-up interview) data in our primary analysis. First, we examined the proportion of events and follow-up time contributed by the follow-up interviews and found that the proportions were similar across cases and controls. Specifically, follow-up interviews yielded an additional 54,787 person-years in case (16.2% of total) and 79,080 in control families (16.2%) (P-value=1.00), with 78 additional cancers reported in case relatives (13.7% of total) and 116 (16.3%) in control relatives (P-value=0.20). Second, we compared the mean time lapse between the initial and follow-up interviews and found a somewhat longer increment for controls versus cases (mean (SD) for controls=9.3 (2.5) years and cases=8.0 (2.0) years, P-value=<0.0001). That control families were allowed more time to develop malignancy suggests that any ensuing bias should result in observed associations that were underestimated. Third, we compared the descriptive characteristics in cases and controls who did and did not complete follow-up interviews, respectively, and found similar distributions across the two groups (Supplementary Table 7). Finally, we examined associations among three data subsets (e.g., all available interview data, initial interview data only, and those with follow-up interviews only; Supplementary Table 8) and found that although the absolute numbers of cancers and the resulting statistical significance varied slightly across the three datasets, the inferences derived from each were equivalent. This analysis suggests that little selection bias was likely introduced by the inclusion of the follow-up interview data; we have therefore elected to present results from all available data herein.

To our knowledge, the current study is the first to consider the relationship between family cancer history and childhood HL stratified by tumor EBV status. Our included cases represent a considerable proportion of North American cases diagnosed during the period 1989-2003, given that an estimated ≥70% of lymphoma patients ages 0-14 years were seen at CCG/COG institutions around the time that patients were being recruited for the current study.⁴⁵ To further address the question of representativeness, we compared characteristics of our cases to those from the Surveillance, Epidemiology and End Results Program (SEER13, 1992-2003), revealing similar distributions of sex (male: 62% vs. 58%) and age at diagnosis (0-9 years: 28% vs. 29%), but somewhat different distributions of histologic subtype (NS: 63% vs. 70%, MC: 18% vs. 14%).²

The primary study limitation is the collection of family cancer history through self-report. Recall bias is of perennial concern in interview-based studies, including those of childhood cancer, for which case families may be more primed to remember prior exposures than control families. Accurate ascertainment of family cancer history is not straightforward; validation studies have demonstrated accurate recalls in the range of 64-100% in first-degree relatives and somewhat lower rates in second-degree relatives.^46-49 Reported accuracies vary by tumor type; common malignancies are recalled with higher sensitivities than hematologic and other rarer cancers. Similarly, recall of family histories is expected to be comparable across cases and controls for more common solid tumors, but may be lower for hematopoietic cancers.^48,50 Common errors include missing information regarding site or morphology, naming a benign condition as malignant, and listing metastatic sites as primary cancers.⁴⁷ We captured cancer histories for first- and second-degree relatives, although multiple generations may be affected.^15,30,31 Finally, the young age at onset of the index children means their first- and second-degree family members are also relatively young and susceptible relatives may not have developed a malignancy yet, leading to some degree of misclassification.

Our results should be interpreted with caution given the multiple comparisons made, as well as the small sample sizes and limited number of affected individuals for some subgroups. We did not adjust for the large number of comparisons made, as we thought this would produce adjusted P-values that are overly conservative, given that HL subgroups may have overlapping etiologies, family members are not independent from one another, and different cancers may cluster in individual families in a non-random way.

Conclusions

We confirm that a positive family history of malignancy, particularly early onset cancers and LN in first-degree relatives, is associated with increased risk of pediatric/adolescent HL. Our study design did not permit exploration of specific genetic or environmental risk factors for familial LN aggregation. Nonetheless, the consistency of our findings with those from other published reports, as well as results of the simulation study by Zimmerman et al demonstrating that reconstructed cohort studies grossly underestimate true genetic contributions to risk,²⁶ suggest that a family-based genomic study focused on probands with early onset HL holds promise for the discovery of novel genetic susceptibility variants for HL and other LN.

Novelty and impact of the research.

We analyzed family histories from the largest etiologic study of incident pediatric/adolescent HL conducted to date to evaluate patterns of cancer aggregation. We found a greater number of lymphoid neoplasms in first-degree relatives of cases versus controls, but no discernable differences by case tumor EBV status. Our results, similar to those in adults, suggest shared genetic and environmental cancer etiology among family members and provide support for family-based genomic studies of early onset HL.

Abbreviations

ALL

Acute lymphoblastic leukemia

CCG

Children's Cancer Group

CI

Confidence interval

CLL

Chronic lymphocytic leukemia

COG

Children's Oncology Group

EBER

Epstein-Barr virus-encoded small RNA

EBV

Epstein-Barr virus

GEE

Generalized estimating equation

HL

Hodgkin lymphoma

HR

Hazard ratio

HS

High school

ICD

International Classification of Diseases

LN

Lymphoid neoplasm

LP

Lymphocyte predominant

MC

Mixed cellularity

MM

Multiple myeloma

NHL

Non-Hodgkin lymphoma

NS

Nodular sclerosis

OR

Odds ratio

RDD

Random digit dialing

RNA

Ribonucleic acid

RS

Robust score

SES

Socio-economic status

SIR

Standardized incidence ratio

UPDB

Utah Population Database