Cigarette smoking and risk of Hodgkin lymphoma and its subtypes: a pooled analysis from the International Lymphoma Epidemiology Consortium (InterLymph)

Summary

Background

The etiology of Hodgkin lymphoma (HL) remains incompletely characterized. Studies of the association between smoking and HL have yielded ambiguous results, possibly due to differences between HL subtypes.

Patients and methods

Through the InterLymph Consortium, 12 case–control studies regarding cigarette smoking and HL were identified. Pooled analyses on the association between smoking and HL stratified by tumor histology and Epstein–Barr virus (EBV) status were conducted using random effects models adjusted for confounders. Analyses included 3335 HL cases and 14 278 controls.

Results

Overall, 54.5% of cases and 57.4% of controls were ever cigarette smokers. Compared with never smokers, ever smokers had an odds ratio (OR) of HL of 1.10 [95% confidence interval (CI) 1.01–1.21]. This increased risk reflected associations with mixed cellularity cHL (OR = 1.60, 95% CI 1.29–1.99) and EBV-positive cHL (OR = 1.81, 95% CI 1.27–2.56) among current smokers, whereas risk of nodular sclerosis (OR = 1.09, 95% CI 0.90–1.32) and EBV-negative HL (OR = 1.02, 95% CI 0.72–1.44) was not increased.

Conclusion

These results support the notion of etiologic heterogeneity between HL subtypes, highlighting the need for HL stratification in future studies. Even if not relevant to all subtypes, our study emphasizes that cigarette smoking should be added to the few modifiable HL risk factors identified.

introduction

Despite considerable efforts, the etiology of classical Hodgkin lymphoma (cHL) remains incompletely characterized with few risk factors firmly established [1]. Because cHL is among the most common malignancies among adolescents and younger adults in industrialized countries in the Western hemisphere [2], and because high cure rates have hitherto been accompanied by considerable treatment-related morbidity among survivors, research on risk factors is warranted.

The uncertainty regarding risk factors may reflect that cHL comprises a number of etiologically diverse entities [3, 4], which have been difficult to define precisely. Accordingly, cHL can be grouped according to age at diagnosis (children, younger adults, older adults), histological subtype (nodular sclerosis, mixed cellularity, lymphocyte rich, lymphocyte depleted), and the presence/absence of Epstein–Barr virus (EBV) in the malignant Hodgkin/Reed-Sternberg cells [3–6]. While these criteria are correlated, e.g. the predominant type of cHL in younger adults is nodular sclerosis, the majority of which is EBV negative, they are at the same time too incongruent to be mutually interchangeable. Therefore, the interpretation and comparison of epidemiological studies employing different stratifications of cHL and including different proportions of each subtype in the patient population can be difficult. Thus, studies are warranted which may lead to the definition of etiologically distinct cHL subtypes, e.g. by identifying subtype-specific risk factors.

Tobacco smoking is a potential cHL risk factor worth scrutinizing. Though the number of studies is relatively limited, findings have been inconsistent, e.g. both statistically significantly reduced [7], statistically unchanged [8–19] and statistically significantly increased [20–26] risks of HL overall have been reported among ever smokers. The inconsistent findings may reflect various methodological issues, such as limited statistical power because of small sample sizes, biases related to choice of comparison groups, study participation, exposure assessment, and uncontrolled confounding. However, it may also reflect uneven distributions of cHL variants between the studies, as described above. For example Briggs et al. observed a 1.3-fold significantly increased risk of HL overall among ever versus never smokers [20]. The increased HL risk appeared to be restricted to current smokers, among whom the association was considerably stronger for mixed cellularity HL [odds ratio (OR) = 3.4, 95% confidence interval (CI) 1.8–6.4] than for the nodular sclerosis HL variant (OR = 1.6, 95% CI 1.1–2.3). Two recent meta-analyses reached the conclusion that cigarette smoking increases the overall risk of HL [27, 28]. One meta-analysis of 14 case–control studies and 3 cohort studies reported a stronger association of current smoking with risk of EBV-positive HL (summary relative risk [RR] = 2.26) than EBV-negative HL (summary RR = 1.40) [27]. The other meta-analysis, which included 17 case–control studies and 4 cohort studies of HL, reported a stronger association of current smoking with risk of mixed cellularity HL (summary RR = 2.53) than nodular sclerosis HL (summary RR = 1.35) [28]. Neither of the two, however, simultaneously considered tumor histology and EBV status.

Because HL is a relatively rare disease with annual incidence rates around 2–4 per 100 000 person-years in Western countries, epidemiological investigations have often suffered from limited statistical power to detect risk factors for HL overall, and even more so for HL subtypes. To address this limitation, in the present investigation, we took advantage of an international collaboration across 12 case–control studies totaling more than 3300 patients with HL and more than 14,000 controls without HL with individual level exposure data, information about HL histology and, in a subset, HL EBV status.

materials and methods

Candidate studies were identified through the InterLymph Consortium [29]. Original case–control studies with electronically available data and information regarding cigarette smoking and HL were included in the study pool. From studies with relevant data, we also requested information on tumor histology and EBV status. Smoking history was assessed using the following variables: ever versus never tobacco smoking, type of tobacco smoking (cigarettes, other, or both); and for cigarette smokers, current or former smoking status, age at initiation, average number of cigarettes smoked per day, duration of smoking, and time since cessation. In an included study of twins [30], the interview data were obtained many years after HL diagnosis. The smoking variables, age, and time were transformed to reflect the situation at the time of diagnosis of the case twin.

Smoking variables from individual studies were harmonized to fit common definitions of ‘ever cigarette smoker’, ‘current cigarette smoker’, and ‘former cigarette smoker’. These common definitions aimed at establishing the smoking status at the time of diagnosis in cases, rather than at the time of interview. Former cigarette smoking implied cessation for such a long time (≥1 year) that it was not primarily a matter of reverse causation. However, two American studies had information on time since cessation only for a small fraction of former smokers, making this correction ineffective in these cases. Forty-nine cases and 30 controls were recoded from former to current smokers to fulfill this requirement. We defined participants as ever cigarette smokers based on the criteria available in each study. Inclusion in this study required the participant to either have a cigarette-smoking history or to have never-smoked tobacco. The duration of cigarette smoking was defined by the individual studies, and pack-years were calculated as the product of the duration of cigarette smoking and cigarette-smoking intensity. Categorical analyses for initiation age, cigarettes per day, smoking duration, and pack years were carried out on current compared with never cigarette smokers. Dose–response analyses regarding years since cessation were carried out on the group of former compared with never cigarette smokers. Dose–response trend analyses were restricted to either current or former cigarette smokers as appropriate.

Besides HL and cigarette-smoking-related data, we requested interview data on age at diagnosis (or pseudo-diagnosis for controls), sex, calendar year, locality, race (White, Black, Asian, other), current and childhood socioeconomic status (each coded as 3: low, 2: mid, 1: high), pre-school childcare attendance, number of siblings, history of infectious mononucleosis, and type of home in childhood (single-family or multiunit). Not all variables were available from all studies. For details on data collection and definitions, see references to the individual studies given in Table 1. The categorization of relative socioeconomic status was left to the contributing studies, most often based on education (see original references).

Table 1.

Characteristics of studies of Hodgkin lymphoma and cigarette smoking included in the pooled analyses

Data provider (reference)	Location	Period of interview	Age at (pseudo) diagnosis	Cases (% part)	Controls (% part)	Matching	Source of controls
Chang [22]	Massachusetts, Connecticut (USA)	1998–2002	15–81	500 (85)	614 (56)	Frequency matching by age, sex, and state of residence	Population controls without a prior history of Hodgkin's lymphoma
Cozen [30, 59]	USA and Canada	1961–1992	14–49	86 (70)	87 (70)	Twin controls	Twin controls
Dal Maso [31, 60]	Aviano, Milan (Italy) Aviano, Naples (Italy)	1983–1992 1998–2002	14–85 18–83	156 (97) 62 (97)	1,149 (97) 499 (91)	Frequency matching by center, gender, and age (5-year age groups)	Hospital controls admitted for acute, non-neoplastic, nonimmunological conditions
EpiLymph [24]	Spain, France, Italy, Germany, Ireland, Finland and Czech Republic	1998–2004	16–100	324 (88)	2,511 (81)	Individual matching (Germany and Czech Republic) or frequency matching (France, Ireland, Italy, and Spain) by sex, age (5-year age groups), and residence area	Controls without cancer drawn from population registers (Italy and Germany) or identified in hospitals (Spain, France, Ireland Czech Republic).
Glaser [17]	Northern California (USA)	1990–1996	19–79	310 (87)	325 (72)	Frequency matching on age (5-year age groups), area, and race/ethnicity	Controls without a history of Hodgkin's lymphoma, identified through random digit dialing
Hjalgrim [12]	Denmark and Sweden	1999–2003	17–76	571 (91)	3,085 (71)	Frequency matching by country, age (10-year age-groups), and sex	Controls randomly selected from population registers
Kane [26]	North, East, and West Yorkshire, Lancashire, South Lakeland, Cornwall, South Devon, Dorset, South Hampshire (England)	1998–2004	16–67	250 (79)	234 (69)	Individual matching by sex, date of birth, and area of residence	Controls randomly selected from population registers
Miligi [16]	Italian multi-centre	1990–1997	19–81	360 (85)	1,763 (80)	Frequency matching by age, sex, and area	Controls randomly selected from Demographic or National Health Service registers
Monnereau [9]	Bordeaux, Brest, Caen, Lille, Nantes, Toulouse (France)	2000–2004	18–75	139 (97)	411 (93)	Individual matching by center, sex, age, and residence	Hospital controls with no history of hematological neoplasm
Spinelli [33, 61]	Alberta, Saskatchewan, Manitoba, Quebec, British Columbia, Ontario (Canada)	1991–1993	18–91	262 (68)	1,338 (48)	Frequency matching by age, sex, and province (all males)	Controls randomly selected from: Health Insurance records (Alberta Saskatchewan, Manitoba, Quebec), computerized telephone listings (Ontario), voter's lists (British Columbia)
Zahm [36, 37]	Nebraska (USA) Kansas (USA)	1986–1987 1982–1984	19–98 17–94	69 (96) 116 (92)	1,384 (84) 878 (93)	Frequency matching on age (5-year age groups), race, sex, area, and vital status	Controls without malignancy, identified through random digit dialing, Medicare records, and mortality files

HL was classified histologically according to the Revised European-American classification of Lymphoid Neoplasms and/or the 2001 World Health Organization (WHO) Classification of Hematopoietic and Lymphoid Tumors [12, 22, 24, 31], the Rye classification [16, 30, 32], or ICD for Oncology First [33, 34], Second [17, 33], or Third Editions [9, 26]. HL EBV status typing was assessed by standard immunohistochemistry for latent membrane protein (LMP)-1 and/or Epstein–Barr nuclear antigen (EBNA) and/or by in situ hybridization for EBV-encoded RNA (EBER) (Spanish EpiLymph study and [12, 17, 22, 26]).

A priori it was decided to use random effects models to model variation between studies and localities. As some HL subtype outcomes were available only from a few large studies, some of these were split along state borders to allow better estimation of the random variation between studies, whereas some studies composed of methodologically similar contributions from many small study centers were aggregated to increase the likelihood of being able to provide parameter estimates from each locality. The resulting coding of study locality can be found in Table 2 under the heading ‘substudy’.

Table 2.

Study characteristics and unadjusted odds ratio estimates of Hodgkin lymphoma for ever cigarette smokers versus never smokers and associated 95% confidence limits, according to substudy

Data provider (substudy)	EBV status	Histology	Never smokers (cases/controls)	Ever cigarette smokers (cases/controls)	OR (95% CI)
Chang (Connecticut)	Yes	Yes	85/131	132/154	1.40 (0.97–2.02)
Chang (Massachusetts)	Yes	Yes	129/170	154/159	1.34 (0.97–1.87)
Cozen (USA)	No	Yes	56/59	30/28	1.14 (0.41–3.15)
Dal Maso (Italy 1984–1991)	No	No	79/469	77/680	0.74 (0.51–1.08)
Dal Maso (Italy 1998-2002)	No	Yes	24/182	38/317	1.79 (0.90–3.57)
Epilymph (Spain)	Yes	Yes	20/297	36/326	1.29 (0.68–2.47)
Epilymph (other centres)	No	Yes	117/835	151/1053	1.03 (0.79–1.34)
Glaser (Northern California)	Yes	Yes	181/186	129/139	0.99 (0.72–1.36)
Hjalgrim (Denmark)	Yes	Yes	91/409	147/722	1.63 (1.16–2.30)
Hjalgrim (Sweden)	Yes	Yes	196/952	137/1002	1.08 (0.83–1.42)
Kane (England)a	Yes	Yes	105/111	145/123	1.43 (0.97–2.10)
Miligi (Italy)	No	Yes	160/809	200/954	0.97 (0.75–1.25)
Monnereau (France)	No	Yes	54/150	85/261	0.89 (0.61–1.30)
Spinelli (Canada)	No	Yes	101/393	161/945	1.10 (0.83–1.44)
Zahm (Midwest)	No	Yes	63/925	122/1337	1.31 (0.94–1.83)
Pooled dataset	–	–	1461/6078	1744/8200	1.13 (1.02–1.26)b

^aNot all cases were matched to a control. These excess cases only contributed to the case-series analyses.

^bI² is 6.

We used the methodology suggested by Stukel et al. [35] to adjust the parameter estimates from each substudy for available confounders, rather than a minimal subset of globally available and semantically homogeneous confounders. All confounders were treated as covariates, except race which was coded by indicator variables for being Black, Asian, or Hispanic. Differences in confounder adjustments between studies might then contribute to between-study variation in effect estimates. We anticipated some between-studies variation in effect estimates for this and other reasons, and therefore used random effects models throughout, to quantify this variation.

Many of the original studies were designed and conducted but not always analyzed as individually matched studies. Data for two studies [26, 30] allowed us to obey the original individual matching. These and the other thinly stratified studies [9, 24, 33, 36, 37] were analyzed using conditional logistic regression. Further standard adjustments in the original studies, e.g. adding a linear age trend on top of adjustment for age groups were also accommodated. Most analyses followed the case–control design, but we used case-series analyses when comparing associations between HL subtypes. With respect to evaluating disease association heterogeneity, the case-series approach is preferable to conventional case–control analyses, because it circumvents biases due to different participation and recall between cases and controls [38]. In the case-series comparisons, we adjusted only for sex, age at diagnosis as a trend, and the sociodemographic confounders to retain as many informative strata as possible, under the assumption that cases had been completely sampled (irrespective of age and sex stratum).

Outcomes were HL overall, histology-specific HL (nodular sclerosis or mixed cellularity), EBV-specific cHL (positive or negative cHL), and finally the combination of the latter four. In the latter situation, adjustments were made separately for each combination of outcomes, while the effects of smoking were modeled as histology-specific and EBV status-specific. That is in the latter situation, the effect of some smoking variable z on the risk of each of the four outcomes was modeled as OR (EBV status, histology, z) = OR (histology, z) × OR (EBV status, z). Here, we call OR (histology, z) ‘the association between z and risk of histology-specific HL, taking EBV status into account’ and OR (EBV status, z) ‘the association between z and risk of EBV status-specific HL, taking histology into account’, respectively. We aimed to correctly apportion etiologic heterogeneity in the response to smoking to the correlated outcomes positive EBV status and mixed cellularity histology.

All analyses were carried out using a home-grown macro in SAS version 9.1.3 [39]. For observations analyzed by unconditional logistic regression, this macro avoided enormous or infinite estimates using techniques inspired by Clogg et al. [40] and Heinze and Schemper [41]. In the setting of multiple interest parameters, we assumed the random effects for each independent of the others. Fitting these models, we constrained each random component to have an arbitrarily selected variance of at least 0.0001. Details of the macro are available from the authors. We do not present estimates of the between-studies variation, as they were quite low. Tests for between-study variation were done by naïve likelihood-ratio tests, halving the P-value [42].

results

We obtained data from 12 studies delivered by 11 different providers (Table 1). In total, 3335 HL cases and 14 278 controls were included in the pooled dataset. Data were obtained mainly from frequency or individually matched studies, using population controls [12, 16, 17, 22, 26, 36, 37, 43], hospital controls [9, 31, 32], or both [24, 31]. We also included one twin study in which unaffected twins of cases were used as controls [30]. Data were separated into the 15 substudies covering substantially different geographical areas or time periods as shown in Table 2. Fourteen and seven substudies contained data on histology and EBV status, respectively. Further details can be found in the supplementary Table S1, available at Annals of Oncology online. Overall, 54.5% of cases and 57.4% of controls were ever cigarette smokers, corresponding to a slightly elevated adjusted OR of HL of 1.10 (95% CI 1.01–1.21) (Table 3).

Table 3.

Adjusted odds ratio estimates for Hodgkin lymphoma overall and for EBV and histology-specific Hodgkin lymphoma subtypes and associated 95% confidence limits based on case–control and case-series modeling according to cigarette smoking status by age and sex

	Case–control					Case-series
	All cases	Mixed cellularity	Nodular sclerosis	EBV positive	EBV negative	Mixed cellularity versus nodular sclerosis	Mixed cellularity versus nodular sclerosisa	EBV positive versus EBV neg.	EBV positive versus EBV neg.b
Smoking status
Ever smokers	1.11 (1.01–1.22)A	1.42 (1.17–1.72)C	1.05 (0.94–1.18)	1.57 (1.20–2.05)B	1.06 (0.89–1.26)	1.17 (0.95–1.45)	1.06 (0.79–1.42)	1.43 (1.06–1.93)A	1.30 (0.99–1.71)
Current smokers	1.16 (0.99–1.37)	1.60 (1.29–1.99)C	1.09 (0.90–1.32)	1.81 (1.27–2.56)B	1.02 (0.72–1.44)	1.20 (0.95–1.52)	1.00 (0.71–1.41)	1.63 (1.02–2.61)A	1.45 (1.02–2.05)A
Former smokers	1.00 (0.83–1.21)	1.21 (0.95–1.54)	0.98 (0.84–1.14)	1.28 (0.93–1.78)	1.02 (0.79–1.33)	1.13 (0.86–1.49)	1.13 (0.78–1.62)	1.24 (0.86–1.79)	1.11 (0.79–1.57)
Never smokers	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Men
Ever smokers	1.17 (1.00–1.38)	1.44 (1.13–1.84)B	1.09 (0.92–1.29)	1.85 (1.22–2.80)B	1.09 (0.83–1.44)	1.16 (0.88–1.53)	0.95 (0.63–1.44)	1.67 (1.07–2.60)A	1.44 (0.97–2.14)
Current smokers	1.26 (1.02–1.56)A	1.68 (1.29–2.19)C	1.18 (0.93–1.50)	2.18 (1.42–3.36)C	1.15 (0.73–1.79)	1.20 (0.90–1.62)	0.98 (0.63–1.52)	1.85 (1.08–3.18)A	1.51 (0.99–2.31)
Former smokers	1.01 (0.80–1.26)	1.15 (0.86–1.55)	0.99 (0.80–1.23)	1.36 (0.87–2.13)	1.01 (0.68–1.49)	1.07 (0.75–1.53)	0.92 (0.55–1.53)	1.27 (0.72–2.22)	1.19 (0.73–1.96)
Never smokers	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Women
Ever smokers	1.03 (0.90–1.18)	1.23 (0.91–1.65)	0.99 (0.84–1.16)	1.20 (0.84–1.72)	1.04 (0.84–1.28)	1.12 (0.81–1.53)	1.03 (0.72–1.48)	1.18 (0.80–1.75)	1.08 (0.77–1.52)
Current smokers	1.07 (0.91–1.25)	1.23 (0.86–1.77)	1.04 (0.84–1.28)	1.27 (0.82–1.96)	1.04 (0.79–1.36)	1.09 (0.75–1.58)	0.93 (0.61–1.44)	1.21 (0.75–1.95)	1.14 (0.76–1.71)
Former smokers	1.00 (0.81–1.23)	1.26 (0.87–1.82)	0.97 (0.77–1.22)	1.16 (0.76–1.78)	1.08 (0.80–1.45)	1.23 (0.81–1.87)	1.15 (0.75–1.77)	1.17 (0.72–1.90)	1.03 (0.68–1.56)
Never smokers	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Age 15–44 years
Ever smokers	1.02 (0.91–1.14)	1.28 (1.00–1.65)	0.99 (0.86–1.13)	1.49 (1.08–2.05)A	0.94 (0.77–1.15)	1.23 (0.95–1.60)	1.05 (0.76–1.47)	1.67 (1.19–2.34)B	1.41 (1.04–1.91)A
Current smokers	1.00 (0.85–1.19)	1.33 (1.00–1.75)A	1.00 (0.82–1.22)	1.44 (0.99–2.10)	0.87 (0.62–1.22)	1.22 (0.92–1.62)	1.00 (0.68–1.46)	1.77 (1.17–2.67)B	1.49 (1.05–2.11)A
Former smokers	1.01 (0.77–1.31)	1.25 (0.89–1.77)	0.93 (0.70–1.23)	1.49 (0.97–2.27)	0.91 (0.60–1.39)	1.32 (0.91–1.90)	1.19 (0.76–1.86)	1.52 (0.96–2.41)	1.23 (0.81–1.87)
Never smokers	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Age 45+ years
Ever smokers	1.31 (1.11–1.55)B	1.43 (1.08–1.90)A	1.23 (0.97–1.57)	1.45 (0.95–2.21)	1.46 (1.05–2.02)A	1.03 (0.71–1.49)	0.93 (0.58–1.49)	1.04 (0.58–1.86)	1.01 (0.63–1.62)
Current smokers	1.66 (1.37–2.01)C	1.86 (1.34–2.59)C	1.50 (1.13–1.99)B	2.36 (1.38–4.02)B	1.65 (1.04–2.61)A	1.07 (0.71–1.63)	0.94 (0.54–1.64)	1.50 (0.61–3.66)	1.15 (0.66–1.99)
Former smokers	1.08 (0.89–1.31)	1.18 (0.85–1.63)	1.07 (0.82–1.40)	1.11 (0.70–1.75)	1.34 (0.94–1.92)	0.97 (0.64–1.47)	0.92 (0.56–1.51)	0.97 (0.52–1.82)	0.93 (0.57–1.54)
Never smokers	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Socioeconomic status 1
Ever smokers	1.09 (0.89–1.34)	1.58 (1.10–2.27)A	1.05 (0.81–1.37)	1.34 (0.84–2.15)	0.98 (0.69–1.41)	1.14 (0.77–1.69)	1.05 (0.66–1.67)	1.31 (0.71–2.40)	1.10 (0.70–1.74)
Current smokers	1.23 (0.99–1.54)	1.89 (1.27–2.81)B	1.07 (0.79–1.45)	1.35 (0.79–2.30)	1.05 (0.55–2.01)	1.15 (0.76–1.75)	0.98 (0.60–1.61)	1.36 (0.64–2.88)	1.13 (0.69–1.84)
Former smokers	0.94 (0.70–1.26)	1.43 (0.94–2.19)	1.04 (0.75–1.45)	1.18 (0.69–2.03)	1.03 (0.67–1.59)	1.12 (0.68–1.85)	1.19 (0.68–2.08)	1.19 (0.59–2.41)	1.03 (0.59–1.80)
Never smokers	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Socioeconomic status 2
Ever smokers	1.11 (0.94–1.30)	1.26 (0.92–1.73)	1.11 (0.92–1.35)	1.43 (0.96–2.13)	1.06 (0.80–1.40)	1.14 (0.82–1.59)	1.00 (0.68–1.47)	1.40 (0.91–2.16)	1.28 (0.89–1.86)
Current smokers	1.16 (0.97–1.39)	1.39 (0.95–2.04)	1.17 (0.94–1.46)	1.70 (1.09–2.65)A	1.14 (0.83–1.59)	1.13 (0.79–1.62)	0.97 (0.64–1.48)	1.41 (0.88–2.27)	1.30 (0.87–1.95)
Former smokers	1.07 (0.81–1.42)	1.06 (0.70–1.60)	1.02 (0.79–1.33)	1.09 (0.66–1.82)	1.00 (0.70–1.43)	1.19 (0.75–1.89)	1.08 (0.64–1.82)	1.39 (0.77–2.52)	1.19 (0.72–1.97)
Never smokers	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Socioeconomic status 3
Ever smokers	1.06 (0.87–1.29)	1.26 (0.88–1.79)	0.98 (0.78–1.23)	1.41 (0.89–2.22)	1.13 (0.84–1.52)	1.21 (0.83–1.77)	1.04 (0.67–1.61)	1.28 (0.76–2.14)	1.09 (0.72–1.67)
Current smokers	1.11 (0.83–1.50)	1.29 (0.81–2.05)	1.19 (0.86–1.66)	1.64 (0.87–3.06)	1.36 (0.87–2.12)	1.21 (0.77–1.92)	1.03 (0.58–1.80)	1.66 (0.81–3.43)	1.26 (0.73–2.16)
Former smokers	0.99 (0.77–1.28)	1.33 (0.87–2.04)	0.91 (0.69–1.22)	1.31 (0.78–2.19)	1.11 (0.79–1.56)	1.23 (0.76–2.02)	1.03 (0.62–1.71)	1.09 (0.59–2.02)	1.00 (0.61–1.64)
Never smokers	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)

^aThe association between smoking and risk of histology-specific cHL, taking EBV status into account.

^bThe association between smoking and risk of EBV status-specific cHL, taking histology into account.

*P < 0.05; **P < 0.01; ***P < 0.001.

In Table 3, OR estimates of HL risk according to smoking variables, tumor EBV status, and histology are shown. Overall, compared with never cigarette smokers (OR = 1), current cigarette smoking was associated with a suggestive increased relative risk of HL overall (OR = 1.16, 95% CI 0.98–1.36), whereas the relative risk of HL overall among former cigarette smokers was not increased (OR = 1.00, 95% CI 0.84–1.21).

The increased overall HL risk in current cigarette smokers reflected increased risks of mixed cellularity (OR = 1.60, 95% CI 1.29–1.99) and EBV-positive cHL (OR = 1.81, 95% CI 1.27–2.56), and inconspicuous risks for nodular sclerosis (OR = 1.09, 95% CI 0.90–1.32) and EBV-negative cHL (OR = 1.02, 95% CI 0.72–1.44) (Table 3). The risks for mixed cellularity (OR = 1.21, 95% CI 0.95–1.54) and EBV-positive cHL (OR = 1.28, 95% CI 0.93–1.78) among former cigarette smokers compared with never cigarette smokers were elevated without being statistically significant (Table 3).

Compared with never cigarette smokers, increased risks for mixed cellularity and EBV-positive cHL among current cigarette smokers were seen in both younger and older adults and among men (Table 3). Among female current cigarette smokers, the risk of EBV-positive (OR = 1.27, 95% CI 0.82–1.96) and mixed cellularity cHL (OR = 1.23, 95% CI 0.86–1.77) was elevated without being statistically significant in comparison to never smokers (Table 3).

Compared with the case–control analyses, case-series comparisons found slightly weaker associations for the association of current cigarette smoking both with mixed cellularity versus nodular sclerosis and with EBV positivity versus EBV negativity (Table 3). Taking the correlation between EBV status and histology into account further reduced the associations, leaving current cigarette smoking positively associated only with EBV-positive cHL (Table 3).

In the categorical analyses, statistically significant differences between strata were observed for all variables except years since cessation, when using never smokers as reference group (Table 4). However, when never-smokers were excluded, no statistically significant dose–response trends could be found in the analyses restricted to current smokers.

Table 4.

Adjusted odds ratio estimates of Hodgkin lymphoma and associated 95% confidence limits according to smoking status, EBV status, and histology, stratified on age and sex. Based on case–control and case-series modeling, respectively

	Case–control					Case series
	All cases	Mixed cellularity	Nodular sclerosis	EBV positive	EBV negative	Mixed cellularity versus nodular sclerosis	Mixed cellularity versus nodular sclerosisa	EBV positive versus EBV negative	EBV positive versus EBV negativeb
Initiation age (years)c
≤14	1.08 (0.84–1.38)	1.82 (1.26–2.62)B	1.04 (0.82–1.31)	1.89 (1.18–3.01)B	1.07 (0.65–1.75)	1.24 (0.84–1.85)	1.19 (0.73–1.92)	1.49 (0.91–2.45)	1.26 (0.79–1.99)
15–19	1.24 (1.09–1.42)B	1.64 (1.26–2.14)C	1.16 (0.97–1.39)	1.66 (1.08–2.54)A	1.10 (0.84–1.44)	1.18 (0.89–1.56)	0.90 (0.59–1.37)	1.66 (1.03–2.68)A	1.43 (0.95–2.17)
≥20	1.00 (0.80–1.26)	1.34 (0.90–2.00)	1.03 (0.78–1.37)	2.35 (1.38–4.00)B	1.07 (0.69–1.66)	1.25 (0.74–2.12)	0.98 (0.47–2.04)	2.22 (1.12–4.39)A	1.75 (0.88–3.49)
Never smokers	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
‘Trend per year’	0.99 (0.97–1.01)	0.97 (0.91–1.03)	0.98 (0.94–1.03)	0.96 (0.88–1.05)	1.00 (0.93–1.07)	1.01 (0.96–1.05)	1.01 (0.96–1.07)	1.02 (0.96–1.07)	1.00 (0.95–1.06)
Cigarettes per dayc
1–9	1.07 (0.89–1.30)	1.17 (0.75–1.82)	1.25 (0.98–1.59)	1.31 (0.69–2.47)	1.09 (0.71–1.68)	0.90 (0.58–1.41)	0.67 (0.33–1.36)	1.66 (0.86–3.21)	1.52 (0.84–2.74)
10–19	1.25 (1.04–1.51)A	1.80 (1.32–2.46)C	1.15 (0.93–1.41)	2.32 (1.49–3.60)C	1.41 (1.04–1.91)A	1.20 (0.87–1.66)	0.95 (0.60–1.50)	1.58 (1.00–2.48)	1.48 (0.97–2.26)
≥20	1.26 (1.08–1.48)B	1.68 (1.18–2.39)B	1.11 (0.86–1.42)	2.09 (1.15–3.80)A	1.47 (0.98–2.22)	1.37 (0.98–1.92)	1.07 (0.62–1.83)	1.28 (0.73–2.23)	1.17 (0.70–1.98)
Never smokers	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
‘Trend per 20 cigarettes’	1.03 (0.85–1.24)	1.33 (0.94–1.89)	0.82 (0.59–1.15)	1.15 (0.61–2.19)	0.90 (0.51–1.60)	1.48 (0.99–2.22)	1.31 (0.68–2.51)	1.23 (0.50–3.03)	0.93 (0.49–1.76)
Smoking duration (years)c
0–9	0.96 (0.75–1.23)	1.08 (0.68–1.71)	1.06 (0.80–1.41)	1.71 (0.97–3.00)	1.08 (0.69–1.69)	1.23 (0.79–1.91)	1.09 (0.62–1.90)	1.96 (1.11–3.48)A	1.42 (0.85–2.36)
10–29	1.06 (0.92–1.22)	1.43 (1.04–1.97)A	0.98 (0.82–1.18)	1.37 (0.89–2.11)	0.89 (0.67–1.19)	1.21 (0.90–1.62)	0.87 (0.58–1.31)	1.62 (1.07–2.46)A	1.36 (0.93–1.99)
≥30	1.68 (1.39–2.02)C	2.17 (1.52–3.09)C	1.45 (1.09–1.92)A	2.29 (1.40–3.73)B	1.35 (0.79–2.33)	1.22 (0.82–1.83)	1.19 (0.65–2.17)	1.38 (0.77–2.48)	1.40 (0.79–2.49)
Never smokers	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
‘Trend per10 years’	1.15 (1.00–1.33)	1.24 (0.97–1.59)	0.97 (0.80–1.18)	1.05 (0.76–1.45)	1.07 (0.81–1.41)	1.06 (0.74–1.53)	1.03 (0.66–1.62)	0.86 (0.52–1.43)	0.96 (0.62–1.48)
Pack yearsc
≤14	1.05 (0.91–1.22)	1.11 (0.81–1.53)	1.15 (0.96–1.39)	1.47 (0.91–2.36)	1.22 (0.90–1.65)	1.04 (0.76–1.42)	0.72 (0.45–1.14)	1.70 (0.95–3.01)	1.52 (1.00–2.30)
15–29	1.37 (1.09–1.73)B	2.00 (1.27–3.15)B	1.19 (0.93–1.52)	2.69 (1.57–4.60)C	1.45 (0.98–2.15)	1.33 (0.91–1.94)	1.31 (0.73–2.35)	1.21 (0.67–2.21)	1.14 (0.64–2.00)
≥30	1.65 (1.34–2.03)C	2.34 (1.60–3.42)C	1.21 (0.88–1.67)	2.38 (1.18–4.81)A	1.43 (0.72–2.85)	1.34 (0.87–2.08)	1.06 (0.50–2.26)	1.58 (0.67–2.72)	1.47 (0.70–3.11)
Never smokers	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
‘Trend per year’	1.01 (1.00–1.01)	1.01 (0.99–1.02)	0.99 (0.98–1.01)	1.01 (0.99–1.03)	1.00 (0.98–1.02)	1.01 (0.99–1.02)	1.01 (0.98–1.03)	1.01 (0.97–1.04)	1.00 (0.97–1.02)
Years since cessationd
0–4	1.22 (0.98–1.52)	1.77 (1.12–2.81)A	1.08 (0.80–1.47)	1.11 (0.50–2.46)	0.79 (0.46–1.34)	1.28 (0.79–2.09)	1.14 (0.46–2.80)	1.92 (0.81–4.52)	1.55 (0.69–3.48)
5–9	0.89 (0.69–1.14)	0.87 (0.48–1.57)	0.84 (0.60–1.18)	1.24 (0.59–2.59)	0.81 (0.38–1.73)	0.72 (0.37–1.41)	0.71 (0.27–1.84)	1.22 (0.53–2.81)	0.82 (0.35–1.94)
≥10	0.83 (0.68–1.01)	0.98 (0.68–1.42)	0.85 (0.65–1.09)	1.09 (0.65–1.83)	0.95 (0.67–1.33)	1.19 (0.77–1.83)	1.21 (0.69–2.13)	1.38 (0.77–2.46)	1.31 (0.75–2.29)
Never smokers	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
‘Trend per 10 years’	0.84 (0.71–1.00)A	0.92 (0.75–1.13)	0.86 (0.72–1.03)	1.12 (0.87–1.44)	1.14 (0.93–1.40)	1.05 (0.80–1.38)	0.99 (0.72–1.34)	1.04 (0.70–1.57)	1.08 (0.79–1.46)

^aThe association between smoking and risk of histology-specific cHL, taking EBV status into account.

^bThe association between smoking and risk of EBV status-specific cHL, taking histology into account.

^cAmong current smokers.

^dAmong former smokers.

*P < 0.05; **P < 0.01; ***P < 0.001.

Estimates of between-study variation appeared to be quite low (data not shown). For 79% (257 of 324) of the entries in Tables 3 and 4, the between-studies-variation estimate was at our a priori-defined lower limit. In 10 instances, the test for homogeneity yielded a statistically significant P-value; 10 of 324 is less than the 5% expected by chance.

discussion

In the present analyses, self-reported history of cigarette smoking was associated with a marginally increased risk of HL overall. This is consistent with a recent large cohort study among UK women [44], and of two recent meta-analyses [27, 28] which also included case–control studies contributing to the present investigation. We also carried out detailed case–control and case-series analyses stratified according to tumor and patient characteristics separately and combined to explore the possibility of etiological heterogeneity between HL subtypes. These stratified analyses demonstrated that current cigarette smoking was associated with increased risks of mixed cellularity and EBV-positive cHL across all age groups, whereas the risks of nodular sclerosis and EBV-negative cHL were not systematically increased. These differences by tumor histology and EBV status were largely corroborated in the case-series analyses, which moreover indicated different associations with smoking when simultaneously considered. Thereby, we extend the findings of the two recent meta-analyses, which found stronger associations of smoking with risk of mixed cellularity [28] and EBV-positive HL [27].

While nuances in the association with smoking were apparent in analyses stratified by age and sex, the results suggest that, among persons with cHL, smokers and nonsmokers differ with respect to the histological phenotype and EBV status of their disease. This could imply that the smoking association may pertain predominantly or even exclusively to certain HL subtypes.

Methodological issues and possible causal mechanisms must be considered to explain the observed associations. Involving data from 12 separate studies and with little evidence of statistical heterogeneity between studies, the observed associations are not likely to be chance findings. A variety of biases could have produced spurious associations consistent with the ones observed. These include participation bias, i.e. overrepresentation of individuals with healthy lifestyles among population controls, possibly balanced by the opposite, i.e. over-representation of less healthy individuals in studies using hospital controls, and recall bias, i.e. a tendency for cases to be more likely than controls to recall, e.g. cigarette smoking. However, a number of results argue that these biases are unlikely to be primary explanations for the observed associations. In particular, the association with cigarette smoking was observed specifically for mixed cellularity and EBV-positive cHL, whereas there was no evidence of a consistent association with the larger groups of nodular sclerosis and EBV-negative cHL. Accordingly, there are no strong reasons to believe that participation or recall should differ substantially by histology or EBV status among cases in ways that would explain this. Moreover, there also was good agreement between the case–control and case-series comparisons, the latter being less likely to suffer from the aforementioned biases [38].

There is evidence that cigarette smoking is increasingly associated with lower socioeconomic status in many populations [45]. HL histology and EBV status also vary by socioeconomic status, although the relationship is rather complex and possibly subject to secular changes [5, 46, 47]. Therefore, although all analyses were adjusted for measures of socioeconomic status residual confounding may theoretically have contributed to our observations but is unlikely to account for them entirely.

The increased risk of cHL observed among current cigarette smokers, which is potentially confined to mixed cellularity and/or EBV-positive variants therefore likely reflects a causal association. Biologically, this is plausible. HL, especially the EBV-positive subtype, occurs with increased frequency upon immune incompetence [48, 49], and it is well established that tobacco smoking affects the immune system detrimentally in numerous ways [27, 50]. Altered Th1/Th2 balance, known to be associated with smoking and with HL, might be a possible causal mechanism to explain the association [30, 48]. Epidemiologically, it is also noteworthy that tobacco smoking has been associated with titers of anti-EBV viral capsid antigen immunoglobulin G antibodies [51], which in turn is associated with an increased risk of EBV-positive HL [52].

In contrast to the recent meta-analyses [27, 28], we found no compelling evidence of monotonic dose–response associations between smoking and cHL risk. These conflicting study-specific findings may reflect differences in the analytical approach regarding the reference group. In the present investigation, trend estimates were assessed only within the group of current smokers, whereas, in the meta-analyses, the group of never-smokers was included in the trend estimates [27, 28]. Accordingly, when we mimicked this analytical approach, we too obtained a statistically significant trend in relative risk of all subtypes of HL combined among current smokers of similar magnitude as that of the meta-analyses [27, 28] (data not shown). However, this approach entirely disregards basic and possible smoking unrelated differences between ever and never smokers (in technical terms, the intercept in the statistical models), which was not statistically acceptable in all subanalyses (data not shown).

The absence of an effect of age at cigarette-smoking initiation, pack-years, and duration of cigarette smoking in the present study could also be explained if smoking only influenced one or more late stages in lymphomagenesis [20]. In the extreme, this would imply that only the cigarette-smoking pattern immediately before symptomatic cancer development would influence the risk pattern. The age-specific incidence pattern for HL, with its large fraction of cases in adolescents and younger adults, and evidence that the interval from EBV primary infection to diagnosis of EBV-positive HL may be on the order of a few years [53, 54], make it plausible that recent exposures could be particularly important to risk. This mechanism would explain the absence of dose–response associations between HL risk and age at initiation of cigarette smoking, duration of cigarette smoking, and cumulative cigarette-smoking exposure, and would be consistent with former cigarette smoking being a less strong risk factor for HL than current cigarette smoking. Technically, the inconspicuous trend in HL risk with increasing cigarette smoking intensity could also be due to inadequate statistical power alone or in combination with inhomogeneous exposure information in the contributing studies. Accordingly, despite the large number of cases included, only 197 current cigarette smokers with mixed cellularity HL and 108 current cigarette smokers with EBV-positive HL were available for the dose–response subanalyses.

Stratified analyses showed that smoking history differed between different strata of patients and HL subtypes. The association was stronger in men than in women, stronger in older than in younger adults, stronger for mixed cellularity than for nodular sclerosis cHL, and stronger for EBV-positive than for EBV-negative cHL. The latter variation by EBV status potentially could explain the other differences. Accordingly, the proportion of EBV-positive cases is consistently higher in men than in women, higher in older than in younger adults, and higher in mixed cellularity cHL than in NS cHL [6]. To disentangle the different associations, we carried out case-series analyses that included both cHL histological subtype and EBV status. Albeit with limited statistical precision, these analyses indicated that the association with current cigarette smoking was governed more by HL EBV status than by HL histology. This adds further to the two recent meta-analyses [27, 28], which did not conduct analyses stratified by both histology and EBV status.

However, not all observations were consistent with a restriction of association to EBV-positive HL. Specifically, increased risks were seen across all subtypes of cHL in currently smoking older adults, and accordingly, in this age group, no differences between cHL variants were seen in the case-series analyses. Moreover, though elevated risk estimates were observed for both mixed cellularity and EBV-positive cHL in women, neither of these was formally statistically significant. It therefore remains to be determined whether these findings represent chance variations and/or age and sex-dependent variation in the effect of smoking on HL risk such that, in etiological investigations, HL should be stratified by combining sex, age, histology, and viral status.

Among the strengths of our investigation is the large number of cases and controls, allowing for analyses stratified by HL subtype as well as both case–control and case-series comparisons. In addition, all studies except one [30] included patients with incident HL, and all cases were classified according to standard criteria. Misclassification may nevertheless have occurred between histological subtypes of HL, most likely so that cases of nodular sclerosis cHL were classified as another HL subtype [55, 56]. However, such misclassification cannot explain the association observed with history of cigarette smoking for HL overall, nor for mixed cellularity and EBV-positive cHL.

Another limitation lies in the use of self-reported smoking data. Across studies, history of cigarette smoking was assessed using identically structured questionnaires in cases and controls, and self-reported smoking status has previously been found to be reasonably reliable [57] and valid [58]. However, the covariates available from the included studies may not have been optimal to assess the etiologic association with HL risk. In particular, smoking intensity, derived as an average over a smoking career may be quite different from current or recent smoking intensity, which may be the causally more relevant exposure. Most likely, this misclassification would have driven the observed association with smoking intensity toward the null. Meta-regression of study characteristics on the outcomes from each substudy for some of the most frequent HL outcomes suggested that traditional study characteristics, such as types of controls and prevalence of exposure in controls, were of little importance for the presented results. However, this analysis also suggested some heterogeneity in the way ever cigarette smoking HL cases were classified into former and current smokers, and that this classification occasionally could not avoid reverse causality. Thus, for some American substudies, we found current smoking to be protective, while former smoking increased HL risk. For most European substudies, former smoking seemed protective, which was also counterintuitive (data not shown). Very detailed information about the timing of case-definition, interview, exposure cessation, etc., or prospective studies seems to be the way to avoid this problem in future studies.

To summarize, we observed a positive association between current smoking and risk of cHL, which applied to all subtypes as defined by histology, age, and EBV status, respectively, with the exceptions of nodular sclerosis and EBV-negative cHL in younger adults, respectively. This finding identifies tobacco smoking to be a strong candidate as a modifiable risk factor for a subset of cHL. Importantly, our analyses also support previous speculations of etiological heterogeneity in cHL, which emphasizes the need to take subtype markers into consideration in future investigations.

funding

The study was supported by NIH (grant number 5 ROI CA 69269), Nordic Cancer Union (no grant number), Plan Danmark grant number is 16-02-D, and the Italian Association for Cancer Research (no grant number).

disclosure

The authors have declared no conflicts of interest.