Aetiologic heterogeneity in pediatric Hodgkin lymphoma? Evidence from the Nordic countries, 1978–2010

Abstract

Background. In epidemiological studies, Hodgkin lymphomas (HL) in children younger than 15 years and HL in adolescents and younger adults age 15–35 years has traditionally been studied separately, under the assumption that HL at age 0–14 constitute a homogeneous entity. However, the continued validity of this research practice in affluent settings may be questioned. Specifically, the boundary at age 15 years may not be epidemiologically justified, and therefore also questionable clinically. We therefore updated and further characterised recent HL incidence patterns among Nordic children.

Material and methods. We obtained HL incidence data in children aged 0–14 years for the period 1978–2010 from the five nationwide Nordic cancer registries. The data were analysed by log-linear and/or a mixture of Poisson regression models.

Results. The analyses showed statistically significantly decreasing and increasing HL incidence rates in children younger and older than eight years, respectively during the study period. Statistical modelling suggested that cases in children age 0–6 years constituted a disease entity of its own, whereas cases in older children were more likely to belong to the younger adult HL entity.

Conclusion. Diverging incidence trends and statistical modelling suggest that HL in children age 0–14 years cannot be assumed to constitute a homogeneous disease entity in affluent settings.

In western industrialised countries, a bimodal age-distribution with distinctive incidence peaks in younger adults and older adults has long been considered a hall-mark of Hodgkin lymphoma (HL) epidemiology [1,2]. Early surveys indicated that age-specific HL incidence also was bi-modally distributed in socio-economically deprived populations, and only under such conditions the youngest incidence peak was seen in children, more specifically in boys, rather than in younger adults [3]. The variations in age-specific HL incidence within and between populations were instrumental to the formulation of two epidemiological models for HL. The multiple disease model proposes that HL in children (0–14 years), younger adults (15–34 years), and older adults (35+ years) are aetiologically heterogeneous [1,2]. The late infection model proposes that cases of HL in children and younger adults are caused by the same infectious organism and that the risk of HL is modified by age at infection with this agent [3–5]. Accordingly, the association between the age-group encompassed by the youngest HL incidence peak and socio-economic development suggested by incidence surveys has been attributed to corresponding differences in childhood infectious pressure.

However, early childhood remains a period of relatively intense infectious exposure also in affluent countries, e.g. [6]. This also concerns infection with the Epstein-Barr virus (EBV), which is strongly suspected of being causally related with the majority of HL cases in young children [7], notably boys, see Barros et al. [8]. Thus, in a recent survey of primary EBV infection in Denmark using infectious mononucleosis as proxy we observed an incidence peak around the age of four years, which was particularly pronounced among boys [9].

This led us to hypothesise that also in affluent countries early childhood is a period with a relatively high incidence of HL. We speculated that this would manifest as a discernable peak in age-specific incidence data, the demonstration of which would require more detailed analyses than conventional application of broad, i.e. five-year age categories on low incidence data. To test this hypothesis, we took advantage of data from nationwide cancer and population registries in the Nordic countries. The epidemiological null hypothesis is that in an affluent setting all pediatric HL cases are just very young cases of the young adult HL entity, while a distinct childhood peak suggests different epidemiologies and/or aetiologies of potential relevance for treatment strategies in younger and older children with HL.

Material and methods

We obtained number of HL cases and time at risk in the population by sex, age (0, 1, 2, 3,…, 14 years), calendar year (1978, 1979,…, 2010), and country from the nationwide cancer registries and bureaus of statistics in the Nordic countries (Denmark, Finland, Iceland, Norway and Sweden) as in the NORDCAN database [10].

We tested the hypothesis of the existence of a childhood HL entity in a modelling approach which required HL incidence data on younger adults. To this end available information on number of cases and time at risk in the Danish population by sex, age (15, 16,…., 34 years) and calendar year (1968–2008) from the Danish registers proved sufficient. Information on sex- and age-specific distribution of HL EBV status was also needed for the modelling exercise. This is not available from any registers, and we therefore used the distributions observed among 241 patients of relevant ages (age 18–34 years at diagnosis), participating in a previous investigation of incident HL in Denmark and Sweden between 1999 and 2003 [11].

The raw data for all our analyses consisted of person-time at risk (pyrs) and counts of events (events) stratified by sex and age (one-year categories) and other factors, such as country and calendar year as needed. These data were modelled by Poisson regression, i.e. the expected number of events in each cell was a rate function of the parameters multiplied by person time at risk.

First, we assessed time trends and variation by sex and country. Here, a complicated starting model that, e.g. included country and interactions between age, sex and country was reduced by backward elimination to a much simpler model, the parameters of which are reported here (in the results section). Next, we assessed the existence of a peak in the overall childhood HL incidence rate by comparing incidence models made up by one and two gamma-density shaped components, respectively. Thus, we determined if the left (younger) tail of the younger adult HL entity sufficed statistically to describe HL incidence in children or if the addition of a separate childhood HL entity was required. Each of the components was in the form c_i(a) = α_i × exp[(ν_i + 1) × log(ν_i/θ_i)+ν_i × log(a + 0.5) + (a + 0.5) × ν_i/θ_i]/Γ(ν_i + 1), where a is the age, Γ is the gamma function, ν_i is a shape parameter, α_i and θ_i are scale parameters, scaling c_i vertically (magnitude) and horizontally (time/age), respectively [11,12]. With an adequate data fit and under a rare disease assumption, θ_i becomes the mode (peak) of component c_i(a). Subsequently, we refined the above model by taking age- and sex-specific distributions of HL EBV status into consideration. Thus, the modelling was expanded to include three components, i.e. one EBV-negative component peaking in younger adults, one EBV-positive component also peaking in early adulthood, and one hypothesised component of undetermined EBV-status peaking in childhood. Here, model building was informed by age- and sex-specific data about both HL incidence rates and the balance between EBV-positive and EBV-negative cases. The combined model for data on incident HL in children (Poisson regression), incident HL in younger adults (Poisson regression) and the balance between EBV-positive and EBV-negative HL in EBV-typed adult incident HL cases (logistic regression) was fitted as a function of the parameters above (α_i, ν_i, θ_i, i = 1–3). All these component models were fitted separately for each sex.

All modelling was performed using SAS proc genmod (trends etc.) and SAS proc nlmixed (components) in SAS version 9.3. All significance tests were likelihood-ratio based. As the likelihood function for the mixture models cannot be guaranteed to be unimodal we used a variant of simulated annealing as our heuristic for searching for the maximum likelihood estimates. Details are available from the authors.

Results

The distribution by age, sex and calendar period of the 748 incident cases of HL registered among Nordic children age 0–14 years in the period 1978–2010 is presented in Table I.

Table I. Number of registered cases of Hodgkin lymphoma in the Nordic cancer registers 1978–2010 by age and sex.

Age (years)	Registered cases of Hodgkin lymphoma in the Nordic cancer registries								All
	Sex
	Boys				Girls
	1978–1989	1990–1999	2000–2010	1978–2010	1978–1989	1990–1999	2000–2010	1978–2010
0	0	0	0	0	0	0	0	0	0
1	0	0	0	0	0	1	0	1	1
2	3	1	1	5	0	0	0	0	5
3	1	4	1	6	0	0	0	0	6
4	12	8	3	23	0	1	3	4	27
5	6	9	8	23	2	5	1	8	31
6	6	5	6	17	5	1	2	8	25
7	12	6	6	24	4	4	1	9	33
8	5	10	6	21	4	3	0	7	28
9	12	5	11	28	5	3	2	10	38
10	15	9	21	45	3	5	10	18	63
11	17	12	9	38	9	9	13	31	69
12	11	19	26	56	12	16	13	41	97
13	23	22	32	77	20	15	22	57	134
14	29	41	31	101	27	27	36	90	191
0–14	152	151	161	464	91	90	103	284	748

Using 1995 as a reference year for the trend analyses, the final and simple model for presenting the pattern of childhood and adolescent HL incidence rates consisted of sex- and age-specific rates multiplied by exp[(yr-1995)×(b1 + b2×age)]. Here, b1 represents a common period trend for all age and sex-specific strata and b2 represents an interaction term between age and period. Incidence rates corresponding to the reference year are presented in Figure 1. Thus, statistically both incidence and trend patterns could be assumed to be similar between the Nordic countries. According to the analyses, b1 was −0.0253 (95% CI −0.0528–0.0022, p = 0.07) and b2 was 0.0032 (95% CI 0.0008–0.0056, p = 0.01). Thus, the data suggested a net decrease in HL incidence for the youngest children and a net increase in HL incidence for the oldest children, both effects larger the further away from age eight years (where b1 + b2 × age = 0) for which age there was no time trend (Figure 1).

Figure 1. Modelled age-specific incidence rates of Hodgkin lymphoma in boys (left) and girls (right) age 0–14 years in the Nordic countries in 1978 (blue), 1995 (red), and 2010 (green).

We next examined if the overall age-specific HL incidence pattern was compatible with the existence of separate childhood and adolescent/younger adult HL entities by comparing models assuming one and two disease entities, respectively. These analyses favoured the two entities assumption both for males (p = 4×10⁻⁹) and females (p = 6 ×10⁻⁵) (Figure 2). Put differently the statistical analyses indicated that overall HL incidence in children comprised contributions from both a childhood HL entity and a left (younger) tail of the younger adult HL entity. The incidence of the childhood HL entity peaked at age 5.0 years (95% CI 4.4–5.7 years) for boys, and age 5.7 (95% CI 4.9–6.5) for girls. The age-specific contributions to the overall HL incidence from the childhood entity as predicted by the models are presented in Table II for children aged less than 10 years.

Figure 2. Observed incidence rates (broken lines (smoothed)) of Hodgkin lymphoma in boys (left) and girls (right) age 0–14 years in the Nordic countries 1978–2010, and modelled entities (childhood entity red lines; younger adult entity blue lines; childhood and young adult entities combined purple lines). See Material and methods for description of the model.

Table II. Estimated proportions (%) of all Hodgkin lymphoma cases belonging to the modelled childhood HL entity; by age and sex.

Estimated proportions of Hodgkin lymphoma cases belonging to the modelled childhood entity
Age (years)	Sex
	Males (%)	Females (%)
0	20	100
1	89	100
2	94	99
3	94	98
4	90	94
5	81	85
6	63	69
7	38	46
8	17	25
9	6	12

In the more refined model where the younger adult HL entity was split into an EBV-positive and an EBV-negative entity (allowing the model more flexibility), the statistical need for a separate childhood peak remained significant in males (p = 0.0004), but not in females (p = 0.07).

Discussion

The present analyses showed that age-dependent variation in overall HL incidence in children in the Nordic countries is compatible with the continued existence of a distinct childhood HL entity in these populations. In other words, the age-specific incidence curve for childhood HL in the Nordic countries is unlikely to merely represent the left (younger) tail of the well established younger adult HL entity in the same populations [13]. Indeed, up until and including the age of six years HL cases were more likely to belong to the hypothesised childhood HL entity than to the adolescent/younger adult entity. Conversely, cases of HL diagnosed in children age seven years or older more likely belonged to the adolescent/younger adult HL entity.

The observed childhood HL entity is interesting from both epidemiological and clinical points of view. Epidemiologically, the notion of a separate childhood HL entity is not novel [2,3,14], but its presence has conventionally been reserved for or interpreted in the context of socio-economical deprivation or transition at a population level [2,3,5,15]. The age characteristics of the childhood HL entity suggested by the present analyses may offer an explanation for it having escaped attention or discussion in incidence surveys in affluent populations. Accordingly, cases belonging to the supposed childhood HL entity were essentially symmetrically distributed on either side of the peak incidence ages around five years. This age-distribution combined with a generally low overall HL incidence in children, the use of broad (five-year) age categories in statistical analyses and the superposition of HL cases belonging to the epidemiological younger adult HL entity may have left the childhood incidence peak obscure, at least in more recent data [16].

At the same time as our analyses demonstrate that a childhood HL entity exists even in affluent populations, they also add further to its delineation vis-a-vis the younger adult HL entity characteristic of the same populations. Thus for almost 50 years epidemiologists have adhered to the tri-partition of HL into different entities by age at diagnosis proposed by MacMahon, i.e. 0–14 years, 15–35 years, and 35 + years [2]. This view has been only slightly updated around the turn of the century to reflect the realisation of a causal role for EBV in a large fraction of HL cases, as articulated in the four disease model of HL by Jarrett [14]. This model posits that the childhood HL entity and the young adult HL entity have different aetiology, at the very least causing the childhood HL entity to be EBV-positive and the majority of the young adult HL entity to be EBV-negative. Although it is likely influenced by environmental factors and may have changed over time, our findings indicate that age seven years may currently be a more relevant discriminator between cases belonging to the supposedly different entities of HL, at least in the Nordic countries.

We have previously observed that primary EBV infection presenting as infectious mononucleosis is associated with a transiently increased risk of EBV-positive HL peaking approximately three years after the infection [7]. The present investigation was sparked by the observation of a prominent incidence peak for infectious mononucleosis around the age of four years among Danish boys [9]. The childhood peak in infectious mononucleosis incidence most likely reflects an age interval with a particularly high EBV infection pressure consistent with earlier serologic cross-sectional analyses [17]. It is therefore of interest that the incidence of the childhood HL entity appeared to peak at the age of five years, i.e. after the observed peak age for childhood infectious mononucleosis. While we did not have access to information on EBV status of the cases making up the childhood HL entity, we speculate that in boys, among whom the evidence of the peak was the strongest, this would mostly be EBV-positive [8]. In a recent Swedish survey, the 0–4 year old HL cases were all boys and mostly of mixed cellularity histology, supporting this speculation [18]. With this inspiration, it should be a manageable project to reexamine hospital records and tissue samples from very young (0–6 years) children, and older children (10–14 years) in order to possibly find statistically significant differences in the distribution of histology, stage, tumour EBV status etc. by age, in accordance with other studies with a sufficiently detailed stratification by age at pediatric HL [19,20], see also Barros et al. [8].

The simplest and most obvious interpretation of our data is that continued socio-demographic development have made the childhood HL peak incidence decrease over time, as is evident in the raw numbers in Table I. A similar story can be observed in age- and sex-specific HL rates over a 40-year timespan in Singapore [21]. An alternative explanation would be that the childhood HL entity was largely imported from developing countries through immigration. We did not have access to data about ethnicity of the HL cases. However, the childhood HL peak is also clearly evident in boys in Finland (data not shown), although Finland has the lowest percentage (3%) of foreign-born in the population of the large Nordic countries, a small fraction of whom are from the developing countries and arrived at a time when the childhood HL peak had already more or less vanished [22,23].

The distinction between cases belonging to the childhood and the younger adult HL entities, respectively, may also be relevant to the clinical situation. Specifically, the data suggested that the vast majority of HL among children older than 10 years epidemiologically belong to the younger adult HL entity. This would argue that although there may be obvious reasons dictated by patient physiology there are not necessarily any lymphoma-related reasons to follow different treatment protocols for HL in younger and older children and adolescents [24]. In the same vein, studies of prognostic markers should also take into account that cases occurring before the age of 15 years do not represent a single, uniform HL entity.

Our investigation has a number of strengths and limitations. We identified cases of HL in the Nordic cancer registers known for their high validity and completeness and with established procedures for ensuring comparability over time and between countries [10]. There is therefore no reason to assume that our findings in any material way should be attributed to biases related to registration procedures or other data quality issues, including diagnostic misclassification. Covering a period of 33 years in a population of more than 20 million people, the number of cases identified was sufficient for incidence rates being stratified into one-year age groups required for the analyses. To this end, corresponding information on the composition of the underlying populations were also available from the national bureaus of statistics. As our analyses were based on a priori assumptions, we also consider the findings unlikely to have resulted from chance alone. We did not expand our analyses to include information on tumour histology because such data was not available for the entire material and further analyses therefore statistically meaningless. Also, information on HL EBV status was not available from any resources.

In conclusion, data suggest that the incidence of HL in children less than eight years has decreased in the past decades in the Nordic countries, while at the same time it has increased in older children. Statistically, we found evidence of a distinct childhood HL entity, predominantly affecting boys, which comprised the majority of cases under the age of seven years. This observation suggests that at least in males the age-specific incidence pattern of HL in these countries in fact has three peaks, i.e. in children, in younger adults and in older adults, rather than only the latter two. The childhood entity appears to peak in the wake of an age-interval with a high intensity of EBV infection, and we therefore speculate that it comprises mostly EBV-positive cases.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.