The association between birth order and childhood leukemia is modified by paternal age and birth weight. Pooled results from the International Childhood Cancer Cohort Consortium (I4C)

Abstract

The “delayed infection hypothesis” states that a paucity of infections in early childhood may lead to higher risks of childhood leukemia (CL), especially acute lymphoblastic leukemia (ALL).

Using prospectively collected data from six population-based birth cohorts we studied the association between birth order (a proxy for pathogen exposure) and CL. We explored whether other birth or parental characteristics modify this association.

With 2.2X10⁶ person-years of follow-up, 185 CL and 136 ALL cases were ascertained. In Cox proportional hazards models, increasing birth order (continuous) was inversely associated with CL and ALL; hazard ratios (HR) = 0.88, 95% confidence interval (CI): (0.77– 0.99) and 0.85: (0.73–0.99), respectively. Being later-born was associated with similarly reduced hazards of CL and ALL compared to being first-born; HRs=0.78: 95% CI: 0.58–1.05 and 0.73: 0.52–1.03, respectively. Successive birth orders were associated with decreased CL and ALL risks (P for trend 0.047 and 0.055, respectively). Multivariable adjustment somewhat attenuated the associations. We found statistically significant and borderline interactions between birth weight (P=0.024) and paternal age (P=0.067), respectively, in associations between being later-born and CL, with the lowest risk observed for children born at <3 kg with fathers aged 35+ (HR=0.18, 95% CI: 0.06–0.50). This study strengthens the theory that increasing birth order confers protection against CL and ALL risks,but suggests that this association may be modified among subsets of children with different characteristics, notably advanced paternal age and lower birth weight. It is unclear whether these findings can be explained solely by infectious exposures.

Introduction

Childhood leukemia is thought to be related to environmental exposures during the prenatal, postnatal, and possibly the preconception periods. Current theories on the etiology of childhood leukemia, especially acute lymphoblastic leukemia (ALL), include the “delayed infection hypothesis”, which states that reduced exposure to infection in early childhood results in an abnormal response to common pathogens, leading to lymphoproliferation and clonal expansion of B or T cell clones¹. There is limited direct evidence for reduced exposure to specific infections among children who develop leukemia² as early childhood exposure to multiple pathogens is difficult to measure precisely, particularly in retrospective studies relying on maternal recall. Therefore, proxy measures have been evaluated, including daycare attendance, breastfeeding, vaccination history, hospitalization or prescriptions for infection, as well as birth order or sibship size^{3–5among others}. In fact, later-born children have increased serological evidence of several common infections⁶.

The association between family size and childhood cancer was first raised in 1962 by MacMahon and Newill, who showed a monotonic decrease in childhood cancer mortality for birth order 1–5⁷. Since then, studies from many countries have explored the relationship between birth order and childhood cancer in general or leukemia in particular, with inconsistent results^{3,8–10among others}. These inconsistencies may be due to differences in study design or heterogeneity regarding sociodemographic or perinatal characteristics among populations studied.

International data demonstrate secular trends of increasing birth weight¹¹, increasing average paternal age¹² as well as decreased total fertility rates and family size¹³. The current and future effects of these recent developments on childhood leukemia risk have received little attention. Birth weight has emerged as a clear and consistent risk factor for childhood leukemia¹⁴. Furthermore, increased paternal age has been raised as a potential risk factor for childhood ALL¹⁵. In most epidemiologic studies, these two factors have been considered as potential confounders of the association between birth order and leukemia, but have not been assessed as modifiers of this relationship. The search for modifiers may uncover specific subgroups in which biological or environmental determinants have differential impacts; this knowledge may enhance the investigation of the etiologic pathways leading to childhood leukemia.

Since childhood cancers are rare and risk factors assessed to date explain only a small part of their etiology, large collaborative efforts are required to accumulate sufficient cases for epidemiologic studies. Most epidemiologic evidence for the etiology of childhood cancer is based on retrospective studies or birth registries. The International Childhood Cancer Cohort Consortium (I4C) includes birth cohorts spanning four continents and 50 years of data collection¹⁶, each cohort providing rich and detailed perinatal and postnatal data (see supplemental data). This study aims to explore the association between birth order and childhood leukemia in the I4C and to determine whether other etiologic factors, such as birth weight, paternal age, child sex and age at diagnosis modify the effect of birth order on childhood leukemia risk.

Methods

Study population:

We pooled data from six I4C-member population-based birth cohorts (Table 1 and supplemental data), ALSPAC, CPP, DNBC, JPS, MoBa and TIHS, as detailed in our previous work on birth weight¹⁶. All childhood leukemia cases ascertained to date from participating cohorts were included in the analysis. Regarding non-cases, for four of the cohorts all children were included, while for Scandinavian cohorts, MoBa and DNBC, a random 10% sample of non-cases contributed person-time (due to cohort data-sharing policies). Children with Down syndrome and non-singleton births were excluded.

Table 1-.

Participating cohorts,^* baseline characteristics¹ and incident childhood cancer cases.

	ALSPAC	CPP4	DNBC3	JPS	MoBa3	TIHS	Total
Recruitment years	1991–1992	1959–1966	1996–2003	1964–1976	1999–2009	1987–1995	1959–2009
Total number of live births in cohort	14,062	58,000	96,860	92,408	108,487	10,628	380,445 (180,549 included in analysis)
Years follow-up, mean (range)	14.9 (0.5–15)	5.6 (0.0–8.0)	14.0 (11.6–15.0)	15.0 (15–15)	9.5 (5.6–15.0)	14.7 (12.7–15.0)	12.3 (0–15)
Person-years	204,819	322,414	129,370	1,317,498	98,986	137,514	2,210,600
% Later born	56.2	71.6	67.5	70.8	55.2	53.1	68.0
Mean sibship size [SD] (range)	0.8 [1.0] (0 – 13)	2.0 [2.1] (0 – 21)	0.9 [0.8] (0 – 6)	2.0 [2.3] (0 – 16)	0.8 [0.9] (0 – 4)	N/A	1.8 [2.1] (0 – 21)
% Male	51.6	50.7	51.0	51.5	50.2	71.3	52.2
Mean birth weight [SD]	3,410 [551]	3,177 [531]	3,586 [567]	3,272 [523]	3,604 [561]	3,195 [751]	3,288 [560]
Mean Paternal Age at Birth [SD](range)	30.7 [5.8] (15 – 65)	28.3 [7.0] (14 – 66)	32.7 [5.2] (18 – 64)	31.6 [6.8] (16 – 80)	32.7 [5.3] (19 – 50)	26.5 [5.6] (14 – 63)	30.6 [6.8] (14 – 80)
Mean Maternal Age at Birth [SD] (range)	28.0 [5.0] (15 – 44)	24.1 [5.9] (12 – 48)	30.5 [4.3] (18 – 47)	27.6 [5.7] (13 – 59)	30.2 [4.6] (16 – 47)	23.6 [4.4] (13 – 45)	26.8 [5.9] (12 – 59)
% Maternal secondary education	35.32	41.4	65.5	42.4	79.7	18.1	43.2
Total number of cancer cases	22	49	190	168	190	24	643
Leukemia	3	16	61	39	62	4	185
ALL	3	11	44	27	49	2	136
U=unknown

^*Reference describing each cohort can be found in Supporting Information A. ALL = Acute lymphoblastic leukemia; SD = Standard Deviation; N/A = Not available.

^1.Proportion with missing values shown in Supporting Information Table S1.

^2.Proportion staying in secondary education after compulsory school-leaving age.

^3.Follow-up not complete to age 15; 10% sample of “noncases” included in analysis.

^4.Follow-up for cancer to age 8 only.

Outcome assessment:

Cancer was ascertained using national or regional cancer registries, where available, or by clinical follow-up as previously described¹⁶.We calculated time to leukemia onset from birth to date of first cancer diagnosis, censoring at date of last follow-up (or age 15 years) or date of death for non-cases (see table 1). The dependent variables were all leukemia and acute lymphoblastic leukemia (ALL), in particular, as ascertained using ICD-0-3 morphology codes 9800-9946 and 9820-9837, respectively. Sample sizes were insufficient to assess other subtypes.

Exposure Assessment:

Using data recorded at baseline in the cohorts, the main independent variable was birth order analyzed as three metrics: later-born vs. first-born (dichotomous); continuous; and ordinal categories. Due to outliers in family size, we performed a sensitivity analysis curtailing continuous birth order at ≤6. Previous multiple births were counted as the number of resultant live births. Previous stillbirths were included as previous births. Paternal age and birth weight were modeled as continuous variables and, in separate analyses, as categorical variables.

The covariates assessed included sex of child, year of birth, gestational age, and birth weight, in addition to parental ages, age difference (father’s age minus mother’s age), maternal smoking and maternal education. Criteria for covariate entry into the models were their importance in the literature or change in the birth-order-cancer or birth-order-leukemia hazard ratios (HR) by more than 10% in either direction. Covariates with missing data were imputed using a chained multiple [N = 20] imputation procedure. The final set of covariates for the full dataset included birth weight (continuous), sex of the child, maternal secondary education (yes vs. no), maternal age (continuous), and paternal age (continuous). Since only 23.2% of the children in the JPS had information on gestational age, an additional analysis was performed for the subcohort with information on this covariate.

Statistical analysis:

Analysis was performed using Cox proportional hazards models, stratified by cohort. We examined whether there was heterogeneity of the association between leukemia and birth order by cohort, and found none. The pooled associations between birth order and leukemia are expressed as hazard ratios (HR) and their corresponding 95% confidence intervals (95% CI). For all analyses, P ≤ 0.05 was considered statistically significant. After testing the main-effects of birth order, we looked for first-order interactions between birth order*paternal age, and birth order*birth weight, and birth order*child sex on their relationship with leukemia and ALL, respectively. We explored the relation between birth order and leukemia diagnosed at successive ages.

Proportional hazards assumptions were confirmed using a test of the Schoenfeld residuals and by graphically observing that the transformed survival probabilities plotted over the natural log of the analysis time for both later-born and first-born were roughly parallel.

Power:

A sample size of 185 cases of leukemia was sufficient to detect a hazard ratio of 0.645 for later-born vs. first-born, with alpha=0.05 and power of 80%, assuming that two-thirds of children would be later-born.

Results

A total of 180,549 children were followed from birth until first cancer diagnosis, death, or end of follow-up. This resulted in a total of 2,210,600 person-years of follow-up and 643 incident cases of childhood cancer, of which 185 were leukemias and 136 were ALL. Years of recruitment and baseline parameters for the cohorts are shown in Table 1. We note differences in characteristics and contribution of cases depending on the size and the age of the cohort. In particular, the Scandinavian cohorts were characterized by later recruitment periods, older paternal age at birth, higher birth weights and higher number of incident cancer cases compared to the earlier cohorts; TIHS was predominantly male, with lowest mean parental ages.

Table 2 shows the unadjusted and multivariable-adjusted relationships between leukemia and birth order using the various metrics. Increasing birth order (continuous) was associated with a decreased risk of leukemia as well as ALL. The results for leukemia and ALL were especially marked after truncating birth order to ≤6 (adjusted HR 0.81 and 0.78, respectively, P=0.017).

Table 2:

Hazard ratios for childhood leukemia, and acute lymphoblastic leukemia (ALL) by birth order as a continuous, dichotomous or ordinal variable.

Cancer Type	Number of cases	Unadjusted Hazard ratio	95% Confidence Interval	P-value	Adjusted† Hazard ratio	95% Confidence Interval	P-value
Continuous birth order (N = 180,549) #
Leukemia	185	0.88	0.77 – 0.99	0.037	0.88	0.76 – 1.02	0.091
ALL	136	0.85	0.73 – 0.99	0.038	0.84	0.70 – 1.01	0.067
Continuous birth order truncated at ≤ 6 (N = 170,894)
Leukemia	177	0.82	0.70 – 0.95	0.009	0.81	0.68 – 0.96	0.017
ALL	131	0.80	0.66 – 0.95	0.013	0.78	0.63 – 0.96	0.017
Later born vs. first born where first born is the referent group (N = 180,549)
Leukemia	185	0.78	0.58 – 1.05	0.097	0.80	0.58 – 1.11	0.180
ALL	136	0.73	0.52 – 1.03	0.077	0.74	0.51 – 1.09	0.129
Successive birth order (N = 180,549)
Leukemia 1 2 3 ≥ 4	185 79 68 22 16	1.00 0.86 0.62 0.60	--- 0.61 – 1.19 0.38 – 1.02 0.33 – 1.07	--- 0.357 0.058 0.082*	1.00 0.84 0.61 0.61	--- 0.59 – 1.18 0.36 – 1.04 0.31 – 1.20	--- 0.309 0.068 0.152
ALL 1 2 3 ≥ 4	136 60 53 12 11	1.00 0.87 0.46 0.61	--- 0.60 – 1.27 0.24 – 0.87 0.31 – 1.20	--- 0.481 0.016 0.154*	1.00 0.85 0.44 0.59	--- 0.57 – 1.26 0.23 – 0.87 0.27 – 1.31	--- 0.407 0.018 0.197

^†The hazard ratio is adjusted for birth weight (continuous), sex of the child, maternal secondary education (yes vs. no), maternal age (continuous) and paternal age (continuous)

^#For TIHS cohort birth order beyond first-born was imputed.

^*P for trend = 0.047 and 0.055 for leukemia and ALL, respectively

Examining later-born vs. first-born we noted a borderline significant inverse relation for leukemia and ALL after adjusting for covariates. Successive birth order was associated with decreased leukemia and ALL risk (P for linear trend 0.047 and 0.055, respectively in unadjusted analyses), Table 2 and supplementary Figure 1. The association was attenuated slightly in the subcohort with available gestational age (Supplemental figure 2). The peak inverse effect of birth order on leukemia (HR = 0.62. 95% CI: 0.38–1.02) and ALL (HR = 0.46, 95% CI: 0.24 – 0.87) was particularly noted at birth order 3. Adjustment for parental age, maternal education, birth weight and child sex did not materially change the HRs but slightly affected the precision of the findings (Table 2 and Supplemental Figure 1).

A statistically significant interaction was found between birth order and continuous birth weight (P=0.024), and a borderline interaction between birth order and continuous paternal age (P=0.067), in their association with time to leukemia. In a model containing main effects of birth order as well as interactions between later-born and continuous paternal age and between later-born and continuous birth weight, the interaction P-values were 0.033 and 0.015, respectively (Supplementary Table 2).

In order to evaluate these interactions further, we conducted a stratified analysis. We examined the effect of birth order in strata of birth weight and paternal age categories individually, with a further analysis combining strata of both these variables (Table 3). In these analyses, we observe that the effect of birth order was strongest among children with lower birth weight (<3kg), in those with older fathers (≥30 years old), and the strongest inverse relations (HR = 0.26 and 0.18) was seen in children with birth weights <3 kg with older fathers (aged 30–35 and >35 years, respectively). This pattern of effect modification was unchanged after adjusting for covariates. We did not find any evidence of a time-varying relationship between birth order and age at diagnosis, using cut-offs of 1, 2, 3, 4, or 5 years of age, nor of effect modification by child’s sex (not shown).

Table 3:

Unadjusted hazard ratios and 95% confidence interval (CI) for leukemia association with birth order (later-born vs. first-born [reference category]) stratified by paternal age and birth weight categories.

Paternal age, years	Birth weight, kg	Observations N	Leukemia N	Hazard ratio for later-born vs first-born	95% CI	P-value
Stratified by paternal age
≤ 25		36,628	20	0.91	0.37 – 2.25	0.833
25–30		49,585	68	1.15	0.71 – 1.87	0.560
30–35		38,507	49	0.70	0.39 – 1.25	0.226
> 35		36,881	41	0.49	0.26 – 0.96	0.037
	Stratified by birth weight
	< 3.0	49,539	26	0.31	0.14–0.72	0.006
	3.0-<4.0	115,545	199	0.83	0.57–1.26	0.316
	≥ 4.0	14,733	40	0.97	0.48–1.94	0.922
	Stratified by both paternal age and birth weight
≤ 25	< 3.0	11,755	6	0.39	0.12–1.23	0.109
	3.0-<4.0	22,780	10	1.08	0.43–2.74	0.870
	≥ 4.0	1,991	4	1.31	0.42–4.06	0.639
25–30	< 3.0	13,233	11	0.44	0.18–1.11	0.082
	3.0-<4.0	32,443	43	1.23	0.72–2.08	0.449
	≥ 4.0	3,719	14	1.49	0.67–3.29	0.326
30–35	< 3.0	8,995	4	0.26	0.10–0.70	0.008
	3.0-<4.0	25,521	30	0.73	0.39–1.35	0.311
	≥ 4.0	3,841	15	0.88	0.38–2.04	0.766
> 35	< 3.0	8,451	4	0.18	0.06–0.50	0.001
	3.0-<4.0	24,054	31	0.50	0.25–1.00	0.051
	≥ 4.0	4,234	6	0.61	0.25–1.50	0.280

Discussion

In this pooled analysis of six cohort studies spanning the globe and over 40 years of recruitment, we found a clear inverse relation between childhood leukemia (and ALL) and birth order. Being a later-born child conferred a reduction in childhood leukemia risk that did not appear to be confounded by parental age, maternal education, child’s sex or birth weight. The relation did not vary by age at diagnosis. The pattern of decreased ALL risk with increasing birth order is similar to those reported in a pooled analysis from five US states⁸. Being a later-born rather than a first-born child implies increased early opportunities for exposure to pathogens post-partum. In general, these findings support the delayed infection hypothesis, although it is important to note that later-born children also have a different immunologic intrauterine milieu than first-borns, with evidence of differences in T-cell programming¹⁷. There may also be other environmental and biological exposures associated with birth order, such as fetal microchimerism.

Marcotte et al³ reported that birth order effects were modified by ethnicity. A novel finding in this study is the apparent modifying effect of increasing paternal age on the relation between birth order and childhood leukemia and a strong modifying effect of birth weight. Although based on small numbers, it appears that the “protective” effect of having siblings is most pronounced in children of lower birth weight and those with older fathers.

Older paternal age is associated with a variety of offspring outcomes including single gene and chromosomal abnormalities, complex disorders such as autism, schizophrenia, and type I diabetes and low birth weight¹⁸. Oksuzyan et al¹⁹ reported a significant association between paternal age (odds ratio 1.23) and childhood ALL in offspring of fathers aged 35–45 compared with fathers aged <25 years, however, no analysis of interactions was reported. Yip et al also reported an increased risk of leukemia, specifically ALL, among older fathers²⁰.

Older fathers, besides the obvious fact that they are likely to have more offspring, more experience in childrearing, and higher incomes than younger fathers, may also bear a higher lifetime burden of environmental exposures along with higher risk of germ cell spontaneous mutations. These factors should seemingly increase the child’s leukemia risk, not the converse. Furthermore, older fathers give birth to offspring with longer telomeres²¹, and telomere length may, in itself, be related to childhood ALL risk²². However, increased paternal age has also been associated with increased rates of fetal loss, both spontaneous abortion and stillbirths¹⁸, this may imply that there is enhanced prenatal loss of infants at risk for leukemia or cancer conceived by older fathers, and raises the possibility of “live-birth bias”²³.

As for birth weight, most research has focused on high birth weight as a risk factor for childhood leukemia, especially ALL. However Roman and colleagues²⁴ have drawn attention to the paucity of childhood ALL cases among children with low birth weight. Whether these children are “protected “ from leukemia due to low somatic growth, or whether the low leukemia rates also reflect in utero loss of fetuses with poor growth characteristics and higher risks of leukemia is unclear. That older siblings may be further protected by having lower birth weight is intriguing, especially since they are at lower risk for impaired fetal growth. It is known that several types of neonatal infection are increased in low birth weight or small-for-gestational-age babies, and this increased risk may persist throughout childhood²⁵. Infection may trigger preterm birth, itself a cause of low birth weight, however controlling for gestational age did not materially affect our results. Is it plausible that the enhanced “protection” conferred by lower birth weight in children with at least one older sibling is due to the compounding effect of increased early exposure to infectious agents, both in utero and post-partum, resulting in a broader immune repertoire? Ongoing studies in our consortium aim to identify DNA-based molecular signatures of infection and how they relate to childhood cancer. Once identified, these signatures may help determine to what extent the associations observed between birth order and leukemia are mediated by infection versus other mechanisms or other exposures.

The strengths of our study are the prospective nature of data collection spanning several countries and several decades. Birth weight, paternal age and birth order were ascertained in all cohorts in an unbiased fashion without reference to later cancer onset. Adjustment for multiple potential confounders did not substantially alter our findings nor did analysis by age of leukemia onset.

Our study is limited by relatively low numbers of events, with insufficient power to examine leukemia subtypes other than ALL. In the earlier cohorts misclassification of leukemia subtype may have occurred due to lack of immunophenotyping. Moreover, we had no information on immune or genetic subtypes of ALL, which may have distinct etiologies. We were not always able to distinguish previous live- from still-births. In some of the included cohorts, children have not all reached 15 years of age, so some findings may change when complete case ascertainment data become available. Further studies undertaken when all the cohorts have matured will provide more information on the entire spectrum of childhood leukemia through to adolescence as well as providing new opportunities for analyses within sibships, and regarding birth spacing.

In conclusion, we have found a clear protective association between having older siblings on childhood leukemia risk and two heretofore unreported possible effect modifiers. It is possible that recent large population-based registry studies reporting null or weak associations between birth order and leukemia^9,10 did not discern the substantial effect reported herein because paternal age and birth weight were examined as covariates and not as potential modifiers. We look forward to efforts to determine whether our findings can be replicated in other populations. Additionally, it would be important to examine whether specific molecular or clinical subtypes of childhood leukemia are associated with these striking inverse relations.

Novelty and Impact:

Most studies of childhood leukemia (CL) are retrospective or registry-based. Using pooled, prospectively-collected data from six population-based birth cohorts we assessed associations between CL and birth order (proxy for pathogen exposure), exploring interactions with other birth and parental characteristics. We observed an inverse association between birth order and CL with effect modification by paternal age and birth weight . “Delayed infection” may provide a partial explanation for diminished CL risk in children with older siblings.